[SPON: Ugreen Nexode Power Bank 20000mAh 130W 30% OFF: amzn.to/4ciEoHi ] [SPON: Ugreen Nexode Power Bank 12000mAh 100W 20% OFF: amzn.to/4cpUUVS ] [SPON: Check Out Ugreen Nexode Power Bank lineup: amzn.to/4dDZ79F ] #Ugreen #powerbank
I had to check, if "Moore's Law" is the name of the theory stating that the number of transistors in ICCs doubles in a certain interval. Then I had to check again, if this channel's name is "Moore's Law is dead". And I must admit I found myself confounded by this complaint video, demanding nothing less than a relaunch of an entire processor family, predicated on the assessment that doing more with much less power isn't good enough. I had to unsubscribe, because there are a dozen YT-channels where I can listen to the same narrow point of view without the confusion about where we are with this technology. The name "Moore's Law is dead" at least suggests that the premise of the channel is that assessments are based on the realization that the semi-conductor technology is inevitably approaching a plateau and thus "generational uplifts" will become smaller and smaller as time goes on. P.S.: Windows 11 is rubbish.
This is why I bought Intel stock yesterday on the cheap. AMD will always massively fumble at some point and so it's only a matter of time before Intel is back in the game.
@Kryptic1046 yeah but MLD already indicated that Intel's next generation CPUs might be delayed until 2025 already so let's see. Plau they are late to AI which the bubble might already be busted. Let's see if they have anything for ARM.
@@Kryptic1046 Except if Intel ends up declaring bankruptcy and reorganizes the stock, you'll receive a pittance of shares of the new stock in exchange for your existing stock and basically be left holding nothing. Don't go too crazy on buying cheap stocks that might not be long for this world. This happened to me with a large number of shares I head in Weatherford years ago when they went bankrupt.
@@idontneedthis66 I appreciate the warning. Nah I didn't do anything too crazy, 50 shares to start. If it rebounds, great, but if not, I'll still be fine.
Oh I'm sure there was pressure.....but it was internally and from marketing. Never attribute to incompetence that which is adequately explained by an overly aggressive marketing team.
My tinfoil hat theory is that they either intentionally neutered the standard SKUs to make their upcoming X3D look better, or they don't care because they assume anyone waiting for X3D was never going to buy the nonX3D parts anyway. Whatever gaming performance uplift they offer with these new 9000 series would need to be significantly outperformed by the 9000X3D chips in order for AMD to charge a meaningful premium for said X3D parts. Put differently, if any of the non-X3D parts outperformed the 7800X3D, that just makes "life" harder for the 9000X3D chips.
Who ever is leading that team should absofuckinglutely be fired, can't play nice with the rest of the company and seems like they turned zen 5 into a disaster.
I see this all the time in my company. A dev team who wants to do things their way, and where the project manager came from the same team and gives them preferential treatment.
Welcome to Big Company 101. I've worked at many big (10,000+ employee) companies. This is, sadly, very common. Someone gets the CEO/VP/Board seat, demands 'kiss the ring' from underlings, and all the good people nope out or get laid off for refusing to tongue-punch fartboxes. Also, in such large companies, getting teams to work together is discouraged because silos drive efficiency (in the addled minds of leadership that can't lead, and managers that can't manage), so tribalism isn't just unfortunate, it's a direct result of...just...garbage tier leadership.
@@spiralout112 Don't assume its only project management. I've also seen how much power the marketing team can and frequently does have over a dev team. Think crap movies you've seen. Sometimes the director is being forced to make decisions because of studio interference....WB I'm looking right. at. you.
What am I missing from the Level1Tech’s CS2 data, at 1080p it’s more CPU limited, at 4K you’re more GPU limited. That said it’s really odd that all CPUs can throughput well over 300 fps at 1080p, but at 4K the 7950X is 5% slower than the 9950X, I can’t say I’ve ever seen results like that.
AMD did some weird shenanigans with Zen 5. So much so that in so many cases Zen 5 is slower than Zen 4. Maybe it's a software issue, we hope it's one, cuz this just isn't it...
It's so weird, especially if they could've worked more on the microcode before launch. AMD should learn with Gaben about how to not do anything and still win, they could've let Intel catching strays in the press and kept quiet, instead they outted themselves with this garbage release for no reason at all lol.
The thing that interests me is the .1% lows - much better on the 9950 at 4k and that might make me think that it would have the edge over the 14900 as resolution drops.
Bro they made a completely new design and they got a few things wrong, (lets not forget that it was meant to be on TSMC's 3nm, but was back ported to 4nm), but it's by no means implies we should panic, Zen 1 was pretty crap, and 4th gen Bulldozer got a 40% IPC improvement on it's last generation that no one seems to remember, and Intel is still on 10nm, everyone is having issues, that's just a fact of trying to put trillions of transitions on a chip.
@@woobilicious. I got a 7900XT and couldnt run FFXIV for 7 months after I got it because the drivers were shit. I literally put my 3060 back in my PC because of it. Don't give me excuses for a multi million dollar company. I also plan to go to AM5 when either my 5900X shits the bed or zen 6 x3d comes out. I buy AMD products, I know their products and I know that they CAN NOT afford to do to their CPUs what they have done to their GPUs.
@@woobilicious. Excavator was 20% IPC at most and it came with a clock speed regression from Steamroller. In many cases, performance was barely improved.
@@Azureskies01 thank you. I'm so tired of people volunteering to lick the boots of billion dollar companies. It's disgusting. You should always have adversarial relationships with these companies. They are not there to do what's best for you, they only want your money. They are happy to take it from you while doing less and less for it. We all need to hold them accountable for everything everytime. No matter what company it is
CZcams comments missed the opportunity to call this channel out for flopping too. From months of claiming "AMD is sandbagging with Zen 5, it'll be great!", to this video where he says "I knew they won't deliver. AMD, do this and that!!". A decent person would do a bit of self-reflection and apologize.
See that slide at 7:31. We were expecting FCLK of 2400MHz, and DDR5 7200 to be the sweet spot and here we are with same old 2000MHz FCLK and DDDR5 6000 the sweetspot.
@@woobilicious. infinity fabric currently transfers data in a DDR fashion, i.e. in both edges of every clock cycle. QDR is double that. So, 3000Mhz QDR would let you transfer at 12GT/s.
@@andreaspatsalides1914 i've heard otherwise from a programmer friend, but even if it is true, what's the point if Linux doesn't have all the software I need and games too, and using VM or emulators will cut performance and convenience
The person behind y-cruncher did a long blog with observations on the zen5 core, relevant to programmers. While most things are good, sse/avx/avx2/avx512 latency has doubled due to a 'core hazard'. I occasionally write this kind of code - 2c latency is far harder to optimize for, and existing code will hit penalties. Opinion: The design team can't have wanted this. Something either interacted badly, or they had to chicken-bit something important.
I think its penny-pinching to save area. They implemented vpintersect and vpconflictd but made and,or,add,xor,nand... half speed. It makes no sense to me.
Lisa Su needs to get ALL of her employees on the same page, they have the ability to make great hardware but seem to be lacking true leadership and communication.
I think issues like that are symptom of AMD going back to being a large company. During Zen 1 design they were pretty much skeleton crew, so the company was easy to manage, now with many teams working on many projects in parallel and competing for budget, promotions etc. they are likely to start making mistakes like Intel did for many years.
@@mrmarecki1 ...and like Intel continue to do. Corporations tend to get uglier and nastier as they grow up, and have to work harder to mask it with public appeal.
We're getting the release now, patch later mentality we see in the gaming sphere in the CPU release world? My oh my. Intel and AMD both out here taking Ls.
Perfect cut at the end. You could just keep getting angrier and angrier at the nonsensicality of it all. AMD not only had no reason to release this early, they had every reason _not_ to release this early. Quick, look at Jessie rolling on the floor to avoid getting too wound up. x)
I mean Intel was down, all they needed was either a good generation or "do nothing" and ride the wave of 7000 series and launch zen 5 when it was ready. Its like AMD wanted to kick Intel while they were down but forgot to watch its own back.
Zen5 is now radioactive and nobody is going to want to touch it. WTF indeed. I was going to grab an X870E with a 9800x3d. Instead, I will upgrade my 3800x to 5800x3d, and I will see you all for Zen7 on AM6.
GN had the frequency bouncing all over while gaming. The frequency was higher than Zen4 parts, but the average performance was lower. Power consumption was also higher because of the unproductive higher frequency. It's like it's boosting at the wrong times, plus other weirdness.
Whiskey Tango Foxtrot indeed 😐I can't imagine working on a newer generation of a product and refusing to look at the last generation because "I know my code/product better". That just seems insane to me and now I'm wondering if holding out for the 9000X3D in my new build is actually something I should do vs just going with a Ryzen 7 7800X3D. I really hope AMD can fix this with an update and get Zen5 back on track to where it should be.
@@d8l835 What kind of build do you have and how do you find the heat? I'm looking at a Fractal Ridge build and it seems like everyone that goes with a higher end AMD X3D in that case needs to undervolt the CPU. My hope is that the 9000X3D chips run much cooler and I can avoid undervolting altogether.
@@stedavid13 I forget the name of the case I have. It's definitely not a top tier one by any means. Probably not even like a mid range by today's standards. I have the chip in a b650 board with updated bios. I run everything default with updated bios, chipset etc. Obviously expo and rebar and all that on. 32gb ddr5 6000 corsair vengeance ram. 2x16gb dimm. I have a Phantom Spirit 120 air cooler. Idle temps 40ish. Benching never got hotter then 85° or so. Granted I haven't preformed any extra long tests or anything. I have done quite a bit of gaming with it so far though. Usually 65-70° ballpark. I have all that paired with a 4070ti super. Alot of games I'm able to run 4k60 with the ray tracing and everything, but usually 1440p and everything turned up to max for higher fps. (4k 60 on the newer Sony game Horizon is gorgeous and an incredible experience) Highest temp ive seen during gaming was about 75°. The cpu preforms as expected in all scenarios so far. I haven't under volted, I haven't touched any cores in BIOS, it's completely default on the newest bios and chipset drivers. With all the uncertainty around zen5 and the new CPUs, and it being unclear if the wins the 7800x3d was having against the new chips is still going to be a thing...lol buy the 7800x3d dude. Just my opinion. If you can hold out another year to be sure, maybe wait, I'm confident they will end up having a good product there, I just don't know when. But the 7800x3d is a fantastic cpu man. I mean actually impressed me. Especially if you're primarily gaming, this thing is insane for that
Fine Wine has always been irrelevant. It ends up getting somewhat closer to Nvidia's performance 3-4 years later, right about when you're going to throw it out anyway.
@@marceldiezasch6192it's closer to 1-2 years but lately it seems you have to find the right driver version for your specific game. To be fair tho, even Nvidia's drivers gave me problems recently on a 3080 randomly running high clocks and power at idle.
why the hell didnt amd just wait and pull back the launch like 2 months? They were in no position where they had to rush anything out. The 7000 series and hell even the 5000 series are still good cpus, why rush it out It makes no sense, Lisa needs to fucking get control of her ship and heads need to roll.
Thats not the Issue. Its a good Feature if it Works. IF IT WORKS. The Issue is it doesnt work. Either AMD fucked up or Windows. I dont know and i dont care. But it should work by now. Its insane it still doesnt 4 Years or how long its been. That is. the Issue.
Doing a refresh of a hardware generation while refusing to use the existing code base of said generation would be absolutely wild. This is something I'd expect out of the Radeon division.
Things are different when looking at the Linux side of things. Expectations were generally met there with a ~15% IPC increase. The question is whose responsibility it is to fix things inside the windows kernel and scheduler. Part of it certainly is AMD but all of it? There are couple of other oddities creeping up. Anand tech’s testing shows something interesting in how multiple CCD’s communicate with each other. Half the cores in the remote CCD have an extra 20 be of latency which doesn’t exist going core to core on the local CCD. Gamers Nexus efficiency testing has some voltage anomalies which make me think things are ~50 mV higher than they should be. It doesn’t make sense that Zen 5 at the same clocks and a slightly better node is at that much higher voltage. This is also all in the context of a delay to their original launch date. AMD needed to push it back a bit but still with some extra time this so the mess at launch? And officially AMD hasn’t given a specific reason for the launch delay. Ultimately I think Zen 5 will be fine in the end as the issues are software based. The key components of the hardware appear to be OK if underwhelming right now. AMD has an event next week so hopefully things will be addressed then.
It's 100% Windows responsibility, It's their code, and they're the only people with access to it, and scheduler code is extremely complicated code that takes lots of planning and consideration. On the other side AMD directly writes code for Linux and submits it to Linus and Co for review, this is one of the reasons why Linux is so attractive to big corps, there's no bureaucracy and being put on the wait-list, you just write the patches, justify your code, have engineers tell you ways to fix it, and they accept it. There's no middle managers and finance department getting in the way saying "this has no monetary value, so we won't assign engineers to it". Linux also has market advantage of being used in Routers and IBM super computers, and ARM Android phones, They've had spent a long time writing flexible code that works in most situations. The new EEVDF CPU scheduler is probably why Linux is doing so well in these benchmarks.
@@Cognitoman Given Phoronix and Wendell’s testing, yeah. There is some nuance there to jump to a modern kernel that has Zen 5 support but that is straight forward.
@@rodrigorras It means that they started with Zen 2 and redesigned from there. It's hard to say exactly what effect that would have without more information but I don't think it's the most common practice.
This is how AMD has been treating the software and code for their GPU's since forever. Release it now - fix it with patches and better drivers in 3 to 6 months Sad to see this leak into CPU releases
For a while? Every gpu since the ati Radeon 9700pro (yes I'm old) has been a rough launch. And every CPU from the athlon 64 (again... I'm old) to zen 2 has been a rough launch. Their whole company is founded on rough launches with crap software and crap drivers. Which is why Nvidia dominates and always will until they sort their crap out
Dual V-cache is significantly better than nothing at all. The Infinity Fabric is fundamentally flawed, part of the massive IPC increases AMD did is by changing the cache structure. Intel only has to worry about overvoltage problems, but the Ring Bus design proved itself to be reliable since Sandy Bridge. When they changed it, it proved to be a flop (watch Meteor Lake and HEDTs).
It’s extremely unfortunate that this launch was botched. I was looking to get a 9950x, but now I’m going to need to wait and see if this issues can be resolved with updates to the microcode and get the gains at least within throwing distance of what AMD initially claimed. As always, thanks for the great reporting.
zen+ was good, fantastic even. you could get 2700x's for $150 consistently not long after launch, which was absolutely killer. and on the other end of the spectrum, 1600AF's were only $85, sometimes less offering competitive performance with the 7700k in some games, and even if it was usually a little slower, an extra 2 cores at less than half of the price. it was a fantastic gen... this one is more expensive than Zen 4 while offering no meaningful improvement in most cases
@@zuckdaddy1596zen+ did not start that cheap at all. Prices dropped months after launch. Ryzen 9000 could drop prices to be really low and be a recreation of Zen+, but the chance of AMD doing this is 0.
@@zuckdaddy1596 yes that's exactly what I meant Zen plus was great value and still strong for Mt performance In my opinion the only way and could have saved zen 5 was make the CCD 10 cores instead of 8 So 9600x is 8c 9700 10c 9900 16 9950x is 20 cores And launch them at closer to street price of zen 4
@@dex6316 the 1600af was a later addition, but $150 2700x's were a feature of Christmas sales that year. a couple months after launch is not long at all, and even before those insane deals it was still arguably a more sensible purchase than Zen 4 is now
AMD marketing is dropping the ball quite a lot... they are responsible for all self owns from the last 10 years and they are still not being reorganized... failing leadership imho...
I find it funny you saying they're responsible for all the "self owns" in the past decade. Aren't they the only owns able to "self own" themselves, you know otherwise it wouldn't be a "self" own?
No this isn't a marketing thing, this falls on the CEO, it wasn't ready and they knew it, the marketing did their best to make a turd look golden, it isn't their fault that AMD decided to launch a buggy product before it was ready.
The 9950X3D NEEDS dual 3D-Vcache CCD's its not optional anymore for AMD. Gamers want to buy the 9950X3D specifically if it has that feature, and full overclocking support. The Ryzen 7 9800X3D will sell ten times more than the Ryzen 9 X3D product line, if the 9950X3D has only one 3D-Vcache CCD.
I cant help but feel disappointed Amd never misses to miss an opportunity They had intel on the ground and instead of delivering the final blow they strike their own feet
Why do people root for final blows to competition? I’m fine if you are rooting for one company over another, but without each other we all would suffer.. do people already forget what things were like when AMD wasn’t putting out competitive products?
@@trailblazinnbruh. How did you get that. Think more like “two top companies pretending to compete to fck over customers” instead of “make the best product and win” it’s that type of thing where AMD could have delivered value and stability where Intel had the opposite
With new releases it really can be nice to just sit back and wait for a couple 2-3 months before you upgrade. Let a couple bios updates come out, lots of reviews. I mean its your hard earned money. These are hundreds of dollars. It really won't kill you to wait a few months.
I'm overdue for a rebuild, but I'm kind of stuck with the 5800x3D as the best I can get on AM4 without any clear CPU of the future that is blowing me away right now. I'm going to be monitoring the next 3D chips but may just wait to see what Zen 6 looks like down the road. The 7800x3D would clearly be an upgrade, but one that would require a full rebuild with a new motherboard, RAM, ect., which is not worth it to me right now. I'll wait until they start to get this new architecture and software sorted.
@@AKSBSUIt's already clear that teams didn't have enough time to work on Zen5 & changed frequently, this isn't going to improve significantly. Any hope for more than 5-10% improvement in anyway is futile. Wait for November/December & it's likely that the cheaper Zen4 X3D will be more appealing than zen5, there is only so much you can fix a born deformed baby, the fixing should have been done when it was still developing, now it's mostly hardcoded in the architecture & scope of improvement is mostly locked up.
@@AKSBSU The whole ram thing slowed me down too. I'm a dev mostly and ddr5 is better but it's hard to get motivated. 10% can sound like a lot, but it's only 10% faster on the edge when you need it and more modest benefit in my workloads. It's always a function of cost; when it's a big jump like that usually you want things to be a lot better. It makes complete sense.
2:37 Gamers Nexus covered this today in their review. Running a 9900x or 9950x requires you to do a bunch of garbage to get core parking working, or something. It's a mess on Windows, and you need to follow their review guide, or you get really weird results.
Windows is a mess did you know that Win 11 has code from ALL previous versions in it? you can literally run Vista or Win2k native programs from Win 11 itself. Windows became a clumps of code glued together pretending to be a new thing
This is exaggerated by CZcamsrs to get views. Install the drivers from Amd, and it just works. Core parking is key in two main scenarios: game core ccd allocation and power saving.
Also I have another question, why do we need to park NON-3d v-cache cores with Zen 5...when we didn't with Zen 4? What is going on with that? Nobody answered this question in any reviews
I guess instead of binning chiplets into different tiers, they can mix a good chiplet and a lesser quality chiplet into each cpu, getting higher gaming and single core performance without impacting all core performance. But it needs correct scheduling...
I really wanted to jump onto zen 5 as im currently into a 1600x, but man this really messed up. AMD had the market in its hands with 13th and 14th gen intels problems. Now Im waiting for arrow lake to even see whether I would take AMD over intel. Heads are going to roll at AMD, or they should.
My guess is that the infinity fabric is a bottleneck. Better scheduling might help to mitigate this by reducing inter-ccd communication but it will not resolve the performance penalty when inter-ccd communication is unavoidable. It’s about time AMD start working on a generational improvement to the infinity fabric of their CPUs.
Tom, your whole concept of dual V-cache R9's continuously misses why its not an option. AMD released new chipset drivers that introduced core parking to non X-3D Zen 5. The reason for that is because it has always been a problem for a certain number of games where inter-ccd communication hurts performance going all the way back to Zen 2. Nothing about dual X3D fixes that until game naturally need to address more than 16 threads. The only reason to get an R9 in general is if you lean toward productivity. That doesn't mean you can't be a gamer as well that can benefit from 1 CCD of Vcache. The performance gained by X3D generally out weighs the 8-10% loss from the clock decrease of the 1 stacked CCD. If you are losing 16-20% by stacking both it makes the product more pointless. Your all or nothing take on gaming vs productivity is really missing that point. I'll explain it this way. I'm nerd with a sysadmin background. I do a number of things that includes Linux and virtual machines, and I play flight sims and various games that very much appreciate the cache. I've only seen star citizen using 60% of a 5950X which is approaching 20 threads with a 6800XT. The 7950X3D is so much faster that it may kiss 50% with an even stronger GPU to feed. That is not anywhere close to being most games. With the 1 stacked CCD approach, you can main Linux, pin the stacked CCD to a Windows VM which will cover almost all of your gaming needs, and still have cores that reach 5.7Ghz for everything else.
Your last sentence there is exactly the main use case I was thinking of for the 1 stacked CCD approach. Having Windows run on a VM specifically configured to only use the stacked CCD is the most guaranteed way to get around all the awkwardness of that design for gaming
@@spralwers I had started doing this way back with a 3900X on Proxmox, and I had other Linux VMs/containers handling various services, while using Windows just for gaming.
@@MaxIronsThird ~18% geometric mean improvement over the 7950x on close to 400 (GNU/Linux) benchmarks sounds fine to me. I can upgrade my 5950x rig to a 9950x and my LLVM compilation/linking times will be cut in half.
And THAT is the reason I don't jump on new products right from launch. I always wait about 3 months (the sweet spot IMO) to let the problems or issue ironed out.
Unfortnately as an engineer I can't agree with most of you. I beleive the guy who forced zen 2 team to design zen 5 with zen 4 code base should be fired. Zen 2 team had a very talented chief architect: AMD fellow David Suggs. Those people who blaimed AMD not punishing him, who are you? You are far lack of knowlege to comment on this. If Dr. Suggs is the chief architect, that means he should make the decison on his own. Zen2 created a new chapter for AMD, why wouldn't his team reuse their code base for Zen5? Let me tell those people who asking zen2 team to fxxxoff, David Suggs left AMD and retired since then, this was AMD loss you pathtic
I actually find this concerning. AMD just got handed the biggest CPU win in history with Intel flagships falling on their face with voltage instability issues and they pull this abortion? Zen5 was supposed to be the new architecture the next few generations Ryzen would be built on. If it can't even match f*cking Zen4 on some cases that's a massive issue. The vast majority of users don't use AVX 512 threads. Intel handed AMD an easy win and they just opened the door for Arrow Lake to stomp Ryzen into the curb.
This seems very much like the incompetence epidemic that has hit most industries after they start replacing vital employees with Indian hires to cut costs for the stockholders..
@@ddelin100 Possibly, yes. Just look up what they were testing. A huge part of their performance tests included AVX512 benchmarks (I'd say about 10-20%of benchmarks show significant scaling with AVX512). Given their changes to AVX512 (first full AVX512 implementation), this drastically impacts average performance. AVX512 is fine in general but doesn't really apply to the real world. Rocket Lake had up to 2.7x speed up in some AVX512 workloads over Comet Lake. In benchmarks that are both used on windows and Linux (like blender), performance is in line with what others found. There the 9600x is only about 14% faster than 7600 and the 9700X about 8% faster than the 7700 (both 65W parts). That's in line with what GN etc found on windows.
He's more of a hardware architect no? I would imagine he mostly designs the chips. But then again, he said in an interview that often his job is to make the tough calls and nuke an architecture when it becomes a tangled enough mess and to set teams on the right path to go forward from. Sounds like they sure need something like this at AMD.
Jim Keller only worked on the first generation of Zen. The Zen 2, 3, and 4 architectures have all been decent uplifts without Keller being involved. Zen 5 doesn’t look like a big uplift in gaming but its expanded AVX-512 instructions support and wider front end will very much benefit the datacenter.
Keller designs the team, not the chips, Apples chips are still good and that's because he got a blank slate to create a top team of engineers. You don't bring in an Engineer like Keller to build a chip and then leave you hanging after he leaves, he comes in, restructures the Engineering team, so they can keep churning out products, Keller isn't a magic silicon wizard who can magically manifest a uarch that's better than the previous generation. Zen 1 wasn't designed by Keller, When you make a CPU design you have to predict the future of software design and how people write code and also run simulations on existing code and see where the bottlenecks are, this isn't a person problem, it's a "how do we structure the team so it's really good at finding bottlenecks and building around them" problem, Zen 1 was already in the mind of someone at AMD, the Issue was trashing the dogshit philosophy they had when producing Bulldozer, AMD actually produced a Bulldozer chip with 40% IPC uplift near the end of it's life, but no one noticed. Zen 5 being a clean sheet, also means it's a longer projection, higher risk gamble, Zen 1 was pretty bad compared to Intel at the time as well, but getting lead times down from 5 years to 2 means they can be more align and flexible and adapt to new bottlenecks, and silicon issues, this is why Zen 4 is so good, because they had a good foundation to work off, Zen 4 was designed by the team, on top of the foundations of Zen 1, you can either stagnate like Intel did with Skylake, or AMD with Bulldozer, or take a slight risk by doing a clean sheet design that lay out the foundation for the next 4 generations, over 2-3 node shrinkages.
@@samuelrodgers2742 Jim Keller's designs were used up until Zen3, and Zen4 is mostly just a frequency boost and introduction of AVX512 instructions, Zen5 was when the Zen team had to go on their own without his guide for the first time. Same seems to be the case for Intel, but it seems like Keller left before everything was fully sketched out, that's why MTL underperformed, and why ARL doesn't have adamantium or Rentable Units, only thing that seems to be going as planned in Foveros.
In the olden times, people knew not to buy the new chip the day or week it was released. Traditional broadcast and media outlets took their time and were careful not to burn their bridges and the people building systems valued stability over performance. These days, it appears that we expect everything to be immediately flawless, overclocked, liquid-cooled, and RGB'd. It's a different set of values. I'm not saying either set is right or wrong, but the changes are worth consideration.
Tbf zen 3 launched fine and zen 4's issues were due to overzealous board partners. We havent seen this level of fuckery since zen 2's launch and x570's usb dropout kerfuffle.
I got to upgrade anyways, so I will buy the 9900X. If you are on an older Intel Gen (Coffee Lake-S) anything recent is an upgrade. Also I am using Linux, so don't really care if the issues are isolated to Windows. Got to research the Linux performance a bit though. Edit: Per "phoronix", Linux performance seems excellent!
I got my 7950x3D back in January and after a certain Windows update i've noticed that core parking was basically disabled in the Balanced power plan profile, no matter what value i was setting the "Processor performance core parking min cores" setting to, (which is normally hidden) it would ignore it, except for the Power Save power plan. It made no, sense, until... ...After doing what Gamer Nexus suggested with that command line, core parking settings are no longer ignored, that's with the latest chipset driver and having set preferred cores to Auto in the BIOS. Normally i don't use core parking in almost any power plan and while that's the opposite of what people have been having issues with these specific CPUs i thought i'd share this.
Zen5 was made to do in the server/workstation market what x3d chips did in gaming. Phoronix not just runs their tests on linux, but also has a more diverse test suite than just running cinebench as a "productivity test". With low cost am4 x3d chips on the market, gamers should essentially ignore everything else. Just buy the best x3d you can afford, and ignore the marketing slides.
Once again petty office politics has derailed a product launch and this time it's worse than marketing slides.... no doubt heads are gonna roll on this one. *I'm referring to both AMD and Microsoft* .
Unless the huge latency between CCDs can be fixed, having v-cache on both CCDs sounds like it would provide very limited utility. You're going to want games to stick to one CCX anyway, so why put v-cache on both CCDs when it probably has to come with at least some sacrifice to clocks?
The only reason that core parking is needed is because of the huge latencies. Using zen2 code base instead of zen 4 is probably why the latencies are so huge; they're doing things the dumb old way. So, once you tune the code to properly shuttle data between CCDs having Vcache on both CCDs would make a 9950X3D the undisputed king of code compiles and (some) games.
Thats what i am saying since forever. V Cache on both CCDs will not change anything. It might even get worse (because you also would have less Frequency and you will cut that away too if your Game would need that). Its totally fine to have only one if it works. Issue is it doesnt. AT ALL. Its like a Chance if it works or not. Like… Well. As long as we talk about Games that is obviously. I dont know about other Apps that need more V Cache. I understand thats a Thing too and i am not against AMD making Chips for them with 2 CCDs V Cache. But for Gaming it would be fine if it would work correctly.
@@JeffGraw And that’s an Ok Idea but you will still loose (a little bit) Performance with that Approach. And it will be more Expensive. It will show better Results in Charts but the single CCD CPU will still be better imo. I don’t want to sound like an Asshole. Really, I don’t. I just don’t see the improvement. I also don’t see how it would help with Energie. If both are activ, there is a Chance it will be worse. Even if not and Coreparking is still a Thing (which it should) it won’t do anything either. I would love to see it so we can test it. But I just think most ppl are thinking V Cache on both is a silver Bullet for Gaming and it will outperform the 7800X3D by another 15% and that’s just not true at all. It may still be worse than the 7800X3D. Hope you understand. It’s not against you personally.
I want more than 8 homogenous cores on one CCX within a single CCD.. Shame no such CPU exists. Intel maybe in Bartlett Lake has one. But Intel come on make 12 Lion Cove cores on Arrow Lake please. That way 12 cores on same tile and good latency. Not the stupid Big.Little crap.
Given that Zen5 with the same microcode performs so much better using Linux and a pretty new 6.10 kernel than using Windows I tend to say that either Zen5 is either optimized for server workloads (hint: Epyc) or Windows is just not ready for Zen5. I lean towards the latter. Yes, it was a bad launch but most of the audience here is waiting for X3D and maybe lower prices anyways. So give them some time. All of our CZcamsrs will tell us regularly about the progress, I'm sure.
It has nothing to do with windows. Just have a look at the chosen applications: They testet: Web performance (mostly js compiling) Code Compiling DAW+3D Rendering (in total one of 14 pages) CAD Synthetic HPC benchmarks (benefit a lot from AX512, there are more AVX512 benchmarks than DAW+3D Rendering+Gaming combined) Database performance Library tests (Python, numpy, php etc) AI Benchmarks (again strongly scale with AVX512) Phoronix didn't find better performance on Linux than Windows, they found better perormance in a certain set of apps.
@@lukas_lsyou don't understand how scheduling works. It doesn't matter the type of test, if windows scheduling sucks, it means all tasks using more than one ccd will have increased latency which means decreased perf. Even some non multi ccd tasks can be worse. And then you say "phoronix results are the same except when they are different" Like bro.
@@putneg97 Then proof it by running the same tests both on Windows and Linux. All tests by phoronix that also run on Windows aren't signifantly better than tests on windows.
@@putneg97 Average fan boy comment. Just look at benchmars done both on Windows and Linux. There is literally no benchmarks that shows signifantly better improvements on Linux than Windows. There are lots of benchmarks done both on Linux and Windows. Examples: Geekench and Blender. Just look at their open databases, you'll find no significant difference in performance improvements. Please explain how I don't understand scheduling and how you'd explain this.
I wonder if the inter-core latency test that are flying around the internet are the same if they are ran in Linux. AMD's next gen HAS TO work on lowering latency/speeding up the IF
The core parking on the 9950X and 9900X is strange, and the wattage on the 9700X is odd. For the X3D, they better not fumble with the 2 CCD V-Cache, no core parking🤞 , and have overclocking. It's unfortunate that AMD has an opportunity with Intel down, yet they seem to be messing it up.
I have to wonder what Lisa is actually doing daily at AMD. There is such a clear disconnect between engineering and marketing. I don't know if engineering is handing marketing legit numbers and marketing then makes up new numbers or vice versa. We saw the terrible marketing with the pointless 5900XT launch, which proved to be 1% faster than a 5950X. Lisa needs to get both parties in the same room and have a very uncomfortable conversation. Is she respected?
Wait a minute. There should actually be no core parking with 9900X or 9950X because the feature was only there for the X3D variants, so that the game uses only the X3D CCD. But the regular 9900X and 9950X don't have X3D so core parking shouldn't be necessary, right?
No. Core parking has been a thing much longer than x3D. It’s pretty simple: if you don’t have enough demand to keep N cores busy, you can save energy and heat by parking unused cores. Chiplets and x3D complicate this practice because it matters which cores you park.
@@AbeDillon But aren't cores going to low power state anyway, if they just do nothing? I always thought that core parking was invented for prefering the 3D Cache chiplet. I never heard about core parking problems back when I had a 3900X or later 5900X.
The cross CCD latency is a very suspicious regression. I’ve seen HWINFO screenshots where the second CCD is clocked much lower. The whole things is really sketchy
Dude I’m sorry but you need to admit you were wrong here. Don’t hide behind someone else’s work in order to redirect scrutiny to them (like Wendell). You were wrong about the performance uplifts, and then you were wrong in suggesting these parts were simply underpowered or underprepared (by not shipping faster RAM, for example, which has absolutely no effect on performance for Zen 5), or that PBO wasnt tested, or that there’s some mysterious bug at the root of it… yes-there are minor discrepancies such as the ones Wendell has pointed out. But no, these parts aren’t going to change drastically after a microcode update. Admit you were wrong and move on.
It's the underlying foundation that is relied by the software stack. That should have been at least Zen4... not Zen damn 2 .... petty office politics. I've seen the b.s that crap causes and it derails everything.
Architecture. Ie the logical blocks of transistors that make up part of a core. Eg the fpu is a logical block of transistors. There can be different versions of an fpu which achieve the same results in different ways with different speeds. In theory, you could take an fpu from the original amd x64 and slot it into a current gen design. You wouldn't do it, because it has surely evolved long past that, and the performance would be terrible, but it should theoretically work. The implication in the claims is that instead of evolving from the Zen5 blocks, they rewound to Zen2 and evolved that. It's kinda like your grandparents having a child when you're already an adult. You're older than them, but they're technically an aunt/uncle. It's awkward, confusing, and there ain't nobody who wants that crap.
Having two VCache CCDs is pointless unless you can make sure that threads are not bouncing between the two CCDs, since the latency benefit you get from storing stuff in the cache would be seriously reduced if not entirely eliminated by the latency of moving across the Infinity Fabric connection to the other CCD.
Exactly, I feel like amd said screw the prior generations of their cpus on purpose to force people to upgrade when they really don't have to just so they can make bank from it.
It's a shame because I need a new PC and work in a lot of "productivity" apps that benefit from a chip like the 9950X, but the game bar and core parking and just general nonsense is such a turnoff. It may not be AMD's _fault,_ but it _is_ their problem.
AMD needs to put 3d v-cache on both dies of the 9950x3d. All is forgiven if they do that. Edit: I wrote this before I finished the video. Tom is right on with how he ends the video!
I have a 7950X3D and I don't understand why people want V-Cache on both CCDs. Hear me out: Tie games to the V-Cache CCD: 7800X3D performance Tie games to the vanilla CCD: 7700X performance Tie games to 4 cores from each CCD: AVG FPS goes 20-ish % below the 7700X/Vanilla CCD and the 1% Lows get laughably bad. What you DO NOT WANT is cross-CCD communication. Slapping V-Cache on both CCDs WILL NOT fix that. AMD needs to bribe Valve and other game stores into forcing the CCD lock right from the launcher or some shit.
[SPON: Ugreen Nexode Power Bank 20000mAh 130W 30% OFF: amzn.to/4ciEoHi ]
[SPON: Ugreen Nexode Power Bank 12000mAh 100W 20% OFF: amzn.to/4cpUUVS ]
[SPON: Check Out Ugreen Nexode Power Bank lineup: amzn.to/4dDZ79F ]
#Ugreen #powerbank
Why are almost none of the reviewers talking about the windows issues?
@@skilletpan5674because it is not their job to make AMDs products work when they cannot do it themselves.
Hi there, I was wondering if you have an affiliate link to support for the Amazon UK store for the UGreen 100w charger.
Thanks
I had to check, if "Moore's Law" is the name of the theory stating that the number of transistors in ICCs doubles in a certain interval.
Then I had to check again, if this channel's name is "Moore's Law is dead".
And I must admit I found myself confounded by this complaint video, demanding nothing less than a relaunch of an entire processor family, predicated on the assessment that doing more with much less power isn't good enough.
I had to unsubscribe, because there are a dozen YT-channels where I can listen to the same narrow point of view without the confusion about where we are with this technology.
The name "Moore's Law is dead" at least suggests that the premise of the channel is that assessments are based on the realization that the semi-conductor technology is inevitably approaching a plateau and thus "generational uplifts" will become smaller and smaller as time goes on.
P.S.: Windows 11 is rubbish.
@@skilletpan5674 Because most reviewers don't have the technological understanding to separate hardware- from software-issues.
AMD being a good sport and shooting its own foot so Intel doesn’t go out of business
This is why I bought Intel stock yesterday on the cheap. AMD will always massively fumble at some point and so it's only a matter of time before Intel is back in the game.
@@Kryptic1046you didn’t by any chance buy $700k worth of stock did you?
@Kryptic1046 yeah but MLD already indicated that Intel's next generation CPUs might be delayed until 2025 already so let's see.
Plau they are late to AI which the bubble might already be busted. Let's see if they have anything for ARM.
@@Kryptic1046 Except if Intel ends up declaring bankruptcy and reorganizes the stock, you'll receive a pittance of shares of the new stock in exchange for your existing stock and basically be left holding nothing. Don't go too crazy on buying cheap stocks that might not be long for this world.
This happened to me with a large number of shares I head in Weatherford years ago when they went bankrupt.
@@idontneedthis66 I appreciate the warning. Nah I didn't do anything too crazy, 50 shares to start. If it rebounds, great, but if not, I'll still be fine.
Never thought I'd see 'ship it early access, we'll finish it with patches' philosophy applied to a processor before...
They are starting to think like a game development company! 😉
Great so we get super computing by accedent as the processors starts identifys as non binary soon?
You know the way game companys went down...
Everybody this is AMD's ARC style launch
It‘s the next logical step after intel tried pushing subscription cores for XEON
@@skyvenrazgriz8226 oh calm down with that crap. Leave the social politics at the door
AMD fumbled under no pressure 😭
They fumbled when Intel is down as down they've been in a while
Oh I'm sure there was pressure.....but it was internally and from marketing. Never attribute to incompetence that which is adequately explained by an overly aggressive marketing team.
When Intel fumbled the microcode, AMD followed suit. It just shows how Intel will always be the superior of the two.
My tinfoil hat theory is that they either intentionally neutered the standard SKUs to make their upcoming X3D look better, or they don't care because they assume anyone waiting for X3D was never going to buy the nonX3D parts anyway.
Whatever gaming performance uplift they offer with these new 9000 series would need to be significantly outperformed by the 9000X3D chips in order for AMD to charge a meaningful premium for said X3D parts.
Put differently, if any of the non-X3D parts outperformed the 7800X3D, that just makes "life" harder for the 9000X3D chips.
@@TMAN-lk2dd Assumingly you prefer microcode that damages the processor?
Amd never misses an opportunity to miss an opportunity
On point
It’s actually incredible, their motto should be something like „swing and a miss“
So original comment
They are masters at snatching defeat from the jaws of victory.
Gee where did you hear that? Every 3rd comment maybe.
how the hell didnt Project management tell the zen 2 team to fuck off.
Who ever is leading that team should absofuckinglutely be fired, can't play nice with the rest of the company and seems like they turned zen 5 into a disaster.
I see this all the time in my company. A dev team who wants to do things their way, and where the project manager came from the same team and gives them preferential treatment.
Welcome to Big Company 101. I've worked at many big (10,000+ employee) companies. This is, sadly, very common. Someone gets the CEO/VP/Board seat, demands 'kiss the ring' from underlings, and all the good people nope out or get laid off for refusing to tongue-punch fartboxes. Also, in such large companies, getting teams to work together is discouraged because silos drive efficiency (in the addled minds of leadership that can't lead, and managers that can't manage), so tribalism isn't just unfortunate, it's a direct result of...just...garbage tier leadership.
@@spiralout112 Don't assume its only project management. I've also seen how much power the marketing team can and frequently does have over a dev team. Think crap movies you've seen. Sometimes the director is being forced to make decisions because of studio interference....WB I'm looking right. at. you.
It's the management that should be told to fuck off. It's always these people rushing the engineers and not doing proper testing.
What am I missing from the Level1Tech’s CS2 data, at 1080p it’s more CPU limited, at 4K you’re more GPU limited. That said it’s really odd that all CPUs can throughput well over 300 fps at 1080p, but at 4K the 7950X is 5% slower than the 9950X, I can’t say I’ve ever seen results like that.
AMD did some weird shenanigans with Zen 5. So much so that in so many cases Zen 5 is slower than Zen 4. Maybe it's a software issue, we hope it's one, cuz this just isn't it...
Could just be another mistake from wendell like the incorrect 7950x cienebench score
It's so weird, especially if they could've worked more on the microcode before launch. AMD should learn with Gaben about how to not do anything and still win, they could've let Intel catching strays in the press and kept quiet, instead they outted themselves with this garbage release for no reason at all lol.
The thing that interests me is the .1% lows - much better on the 9950 at 4k and that might make me think that it would have the edge over the 14900 as resolution drops.
Zen 5 about to become the next 7900 GRE with microcode updates. 💀
AMD treating Zen like how they treat their GPUs is NOT a good thing.
Well pointed out. Massive RDNA 3 vibes here. And it should be the other way around, Radeon should learn from Zen.... sigh
Bro they made a completely new design and they got a few things wrong, (lets not forget that it was meant to be on TSMC's 3nm, but was back ported to 4nm), but it's by no means implies we should panic, Zen 1 was pretty crap, and 4th gen Bulldozer got a 40% IPC improvement on it's last generation that no one seems to remember, and Intel is still on 10nm, everyone is having issues, that's just a fact of trying to put trillions of transitions on a chip.
@@woobilicious. I got a 7900XT and couldnt run FFXIV for 7 months after I got it because the drivers were shit. I literally put my 3060 back in my PC because of it.
Don't give me excuses for a multi million dollar company. I also plan to go to AM5 when either my 5900X shits the bed or zen 6 x3d comes out. I buy AMD products, I know their products and I know that they CAN NOT afford to do to their CPUs what they have done to their GPUs.
@@woobilicious. Excavator was 20% IPC at most and it came with a clock speed regression from Steamroller. In many cases, performance was barely improved.
@@Azureskies01 thank you.
I'm so tired of people volunteering to lick the boots of billion dollar companies. It's disgusting. You should always have adversarial relationships with these companies. They are not there to do what's best for you, they only want your money. They are happy to take it from you while doing less and less for it. We all need to hold them accountable for everything everytime. No matter what company it is
CZcams comments never misses an opportunity to state AMD never misses an opportunity to miss an opportunity.
CZcams comments missed the opportunity to call this channel out for flopping too. From months of claiming "AMD is sandbagging with Zen 5, it'll be great!", to this video where he says "I knew they won't deliver. AMD, do this and that!!".
A decent person would do a bit of self-reflection and apologize.
As a security researcher, launching CPUs with incomplete microcode is horrifying.
See that slide at 7:31. We were expecting FCLK of 2400MHz, and DDR5 7200 to be the sweet spot and here we are with same old 2000MHz FCLK and DDDR5 6000 the sweetspot.
Or better, have IF use QDR at 3000Mhz, and support 6000MT/s DDR5 memory.
@@erkinalp What do you mean QDR at 3000Mhz? 3000MHz FCLK would be 1:1 memory divider like Zen 3...
@@woobilicious. infinity fabric currently transfers data in a DDR fashion, i.e. in both edges of every clock cycle. QDR is double that. So, 3000Mhz QDR would let you transfer at 12GT/s.
@@woobilicious. Yeah, exactly. 1:1 should be much better for latency than Zen 4's weird 2:3 ratio
true, but it was 2400mhz and DDR 8000
Apparently, AMD was feeling sorry for all the bad press Intel has been getting lately.
Misery needs company!
😂😂😂
Thanks Steve.
Windows is the best advertisement for Linux
Until you use Linux
@@FireF1y644 It really depends on the distro you chose. There are stable-beginner friendly distros out there
@@andreaspatsalides1914 i've heard otherwise from a programmer friend, but even if it is true, what's the point if Linux doesn't have all the software I need and games too, and using VM or emulators will cut performance and convenience
Does Zen 5 has better performance on Linux?
@@henson2k seems like it
The person behind y-cruncher did a long blog with observations on the zen5 core, relevant to programmers. While most things are good, sse/avx/avx2/avx512 latency has doubled due to a 'core hazard'. I occasionally write this kind of code - 2c latency is far harder to optimize for, and existing code will hit penalties. Opinion: The design team can't have wanted this. Something either interacted badly, or they had to chicken-bit something important.
I think its penny-pinching to save area. They implemented vpintersect and vpconflictd but made and,or,add,xor,nand... half speed. It makes no sense to me.
To be fair, avx512 also has doubled throughput. Perhaps latency will go down with a revision. I'm excited to have real avx-512 though.
Lisa Su needs to get ALL of her employees on the same page, they have the ability to make great hardware but seem to be lacking true leadership and communication.
The fish rots from the head. Lisa IS the problem.
I think issues like that are symptom of AMD going back to being a large company. During Zen 1 design they were pretty much skeleton crew, so the company was easy to manage, now with many teams working on many projects in parallel and competing for budget, promotions etc. they are likely to start making mistakes like Intel did for many years.
@@mrmarecki1 ...and like Intel continue to do. Corporations tend to get uglier and nastier as they grow up, and have to work harder to mask it with public appeal.
@@henryvaneyk3769 lol, dude throwing around claims based on what data/info?
Its defo a wake up call for the whole company and Lisa Su. Hope she cuts some slack and all the idiots who are responsible for this mess.
This should have been a slam dunk launch, instead it’s an early access release. 😂
Accroding to the guy, Nobody even knows if later it will truly improve significantly.
We're getting the release now, patch later mentality we see in the gaming sphere in the CPU release world? My oh my. Intel and AMD both out here taking Ls.
Perfect cut at the end. You could just keep getting angrier and angrier at the nonsensicality of it all.
AMD not only had no reason to release this early, they had every reason _not_ to release this early. Quick, look at Jessie rolling on the floor to avoid getting too wound up. x)
AMD is the only company right now who's "trying" to get bad reviews and impressions.
The last couple of seconds of this video sums it up perfectly. 😂 Well done, Tom.
Indeed! 🤣 I had just finished drinking a glass of water otherwise.... 🤣
I mean Intel was down, all they needed was either a good generation or "do nothing" and ride the wave of 7000 series and launch zen 5 when it was ready. Its like AMD wanted to kick Intel while they were down but forgot to watch its own back.
Zen5 is now radioactive and nobody is going to want to touch it. WTF indeed.
I was going to grab an X870E with a 9800x3d. Instead, I will upgrade my 3800x to 5800x3d, and I will see you all for Zen7 on AM6.
Only question is how much ram we need
@@victorrazvan8480 At this point in time prepare to have 128gb of ram and sit back relax and enjoy.
Zen 5-6-7-8 will be in AM5,
just like Zen 1, 2, 3, 4 is in AM4
GN had the frequency bouncing all over while gaming. The frequency was higher than Zen4 parts, but the average performance was lower.
Power consumption was also higher because of the unproductive higher frequency. It's like it's boosting at the wrong times, plus other weirdness.
Yes! Something are way off..... Maybe software/AGESA/Bios etc
Whiskey Tango Foxtrot indeed 😐I can't imagine working on a newer generation of a product and refusing to look at the last generation because "I know my code/product better". That just seems insane to me and now I'm wondering if holding out for the 9000X3D in my new build is actually something I should do vs just going with a Ryzen 7 7800X3D.
I really hope AMD can fix this with an update and get Zen5 back on track to where it should be.
Lol I couldn't be happier with my 7800x3d
@@d8l835 What kind of build do you have and how do you find the heat?
I'm looking at a Fractal Ridge build and it seems like everyone that goes with a higher end AMD X3D in that case needs to undervolt the CPU. My hope is that the 9000X3D chips run much cooler and I can avoid undervolting altogether.
@@stedavid13 I forget the name of the case I have. It's definitely not a top tier one by any means. Probably not even like a mid range by today's standards.
I have the chip in a b650 board with updated bios. I run everything default with updated bios, chipset etc. Obviously expo and rebar and all that on. 32gb ddr5 6000 corsair vengeance ram. 2x16gb dimm.
I have a Phantom Spirit 120 air cooler. Idle temps 40ish. Benching never got hotter then 85° or so. Granted I haven't preformed any extra long tests or anything. I have done quite a bit of gaming with it so far though. Usually 65-70° ballpark.
I have all that paired with a 4070ti super. Alot of games I'm able to run 4k60 with the ray tracing and everything, but usually 1440p and everything turned up to max for higher fps. (4k 60 on the newer Sony game Horizon is gorgeous and an incredible experience) Highest temp ive seen during gaming was about 75°.
The cpu preforms as expected in all scenarios so far. I haven't under volted, I haven't touched any cores in BIOS, it's completely default on the newest bios and chipset drivers.
With all the uncertainty around zen5 and the new CPUs, and it being unclear if the wins the 7800x3d was having against the new chips is still going to be a thing...lol buy the 7800x3d dude. Just my opinion. If you can hold out another year to be sure, maybe wait, I'm confident they will end up having a good product there, I just don't know when. But the 7800x3d is a fantastic cpu man. I mean actually impressed me. Especially if you're primarily gaming, this thing is insane for that
@@d8l835 and you should be, that should last you a long time.
Seems more and more like they should've released in september alongside the 800 series motherboards, no idea why they greenlit the launch
TFW now AMD CPU also have that "Fine Wine" effect (aka not optimized well at the start)
AMDone
Is copium... He said the same with RDNA 3... Somehow a software patch might be the answer to everything...
Fine Wine has always been irrelevant. It ends up getting somewhat closer to Nvidia's performance 3-4 years later, right about when you're going to throw it out anyway.
@@marceldiezasch6192it's closer to 1-2 years but lately it seems you have to find the right driver version for your specific game. To be fair tho, even Nvidia's drivers gave me problems recently on a 3080 randomly running high clocks and power at idle.
@@marceldiezasch6192 somebody doesn't look at benchmarks.
why the hell didnt amd just wait and pull back the launch like 2 months? They were in no position where they had to rush anything out. The 7000 series and hell even the 5000 series are still good cpus, why rush it out It makes no sense, Lisa needs to fucking get control of her ship and heads need to roll.
16:41 should of started the video with this. Lol
Agreed 😂
What the fuck happened with zen5? This core parking shit is ridiculous. That shouldn't be needed on a non x3d chip!
Honestly it shouldn't be needed for any CPU
Look at AnandTech, it's because the cross-CCD latency is fucked.
Yes. That is basically NUMA on one "CPU" which is bullshit.
@@BillSherry Which leads to the question why Windows is not able to keep the threads for one process together on one CCD?
Thats not the Issue. Its a good Feature if it Works. IF IT WORKS. The Issue is it doesnt work. Either AMD fucked up or Windows. I dont know and i dont care. But it should work by now. Its insane it still doesnt 4 Years or how long its been.
That is. the Issue.
Doing a refresh of a hardware generation while refusing to use the existing code base of said generation would be absolutely wild. This is something I'd expect out of the Radeon division.
Nowadays, new hardware launches are more fun to watch than TV shows 🤣🤣🤣
Things are different when looking at the Linux side of things. Expectations were generally met there with a ~15% IPC increase. The question is whose responsibility it is to fix things inside the windows kernel and scheduler. Part of it certainly is AMD but all of it?
There are couple of other oddities creeping up. Anand tech’s testing shows something interesting in how multiple CCD’s communicate with each other. Half the cores in the remote CCD have an extra 20 be of latency which doesn’t exist going core to core on the local CCD. Gamers Nexus efficiency testing has some voltage anomalies which make me think things are ~50 mV higher than they should be. It doesn’t make sense that Zen 5 at the same clocks and a slightly better node is at that much higher voltage.
This is also all in the context of a delay to their original launch date. AMD needed to push it back a bit but still with some extra time this so the mess at launch? And officially AMD hasn’t given a specific reason for the launch delay.
Ultimately I think Zen 5 will be fine in the end as the issues are software based. The key components of the hardware appear to be OK if underwhelming right now. AMD has an event next week so hopefully things will be addressed then.
It's 100% Windows responsibility, It's their code, and they're the only people with access to it, and scheduler code is extremely complicated code that takes lots of planning and consideration.
On the other side AMD directly writes code for Linux and submits it to Linus and Co for review, this is one of the reasons why Linux is so attractive to big corps, there's no bureaucracy and being put on the wait-list, you just write the patches, justify your code, have engineers tell you ways to fix it, and they accept it. There's no middle managers and finance department getting in the way saying "this has no monetary value, so we won't assign engineers to it".
Linux also has market advantage of being used in Routers and IBM super computers, and ARM Android phones, They've had spent a long time writing flexible code that works in most situations. The new EEVDF CPU scheduler is probably why Linux is doing so well in these benchmarks.
So if you use Linux it’s alright to get this CPU?
@@Cognitoman Given Phoronix and Wendell’s testing, yeah. There is some nuance there to jump to a modern kernel that has Zen 5 support but that is straight forward.
That zen2 shit is shocking no wonder zen 5 is fucked.
Does it means the chip design is as old as Zen2? or is it related to software/libraries/microcode?
@rodrigorras all good questions!
@@rodrigorras microcode basically. they used old libraries.
I dont know why this was acceptable? The design has changed a lot since then.
hey PS5 pro will be zen 2 so it fits
@@rodrigorras It means that they started with Zen 2 and redesigned from there. It's hard to say exactly what effect that would have without more information but I don't think it's the most common practice.
This is how AMD has been treating the software and code for their GPU's since forever.
Release it now - fix it with patches and better drivers in 3 to 6 months
Sad to see this leak into CPU releases
This is one of the rougher launches amds had in awhile.
Remember RDNA3?
@@jackleville546 Nah 'we' don't.
It was that forgettable
For a while? Every gpu since the ati Radeon 9700pro (yes I'm old) has been a rough launch. And every CPU from the athlon 64 (again... I'm old) to zen 2 has been a rough launch.
Their whole company is founded on rough launches with crap software and crap drivers. Which is why Nvidia dominates and always will until they sort their crap out
It's the worst new generation Ryzen launch they ever had.
Worst since Bulldozer!
V-cache on both CCDs is pointless. The fabric latency is higher on zen 5. Dual v-cache won't work until they improve the fabric.
Dual V-cache is significantly better than nothing at all. The Infinity Fabric is fundamentally flawed, part of the massive IPC increases AMD did is by changing the cache structure.
Intel only has to worry about overvoltage problems, but the Ring Bus design proved itself to be reliable since Sandy Bridge. When they changed it, it proved to be a flop (watch Meteor Lake and HEDTs).
Probably explains they they delayed it 2 weeks but that wasn’t enough. One of those scenarios lets launch it and fix it as long as it’s stable.
It’s extremely unfortunate that this launch was botched. I was looking to get a 9950x, but now I’m going to need to wait and see if this issues can be resolved with updates to the microcode and get the gains at least within throwing distance of what AMD initially claimed. As always, thanks for the great reporting.
The last time I was this early amd CPUs actually had generational improvements
This make zen+ look good
zen+ was good, fantastic even. you could get 2700x's for $150 consistently not long after launch, which was absolutely killer. and on the other end of the spectrum, 1600AF's were only $85, sometimes less offering competitive performance with the 7700k in some games, and even if it was usually a little slower, an extra 2 cores at less than half of the price. it was a fantastic gen... this one is more expensive than Zen 4 while offering no meaningful improvement in most cases
@@zuckdaddy1596zen+ did not start that cheap at all. Prices dropped months after launch. Ryzen 9000 could drop prices to be really low and be a recreation of Zen+, but the chance of AMD doing this is 0.
@@zuckdaddy1596 yes that's exactly what I meant
Zen plus was great value and still strong for Mt performance
In my opinion the only way and could have saved zen 5 was make the CCD 10 cores instead of 8
So 9600x is 8c
9700 10c
9900 16
9950x is 20 cores
And launch them at closer to street price of zen 4
@@dex6316 the 1600af was a later addition, but $150 2700x's were a feature of Christmas sales that year. a couple months after launch is not long at all, and even before those insane deals it was still arguably a more sensible purchase than Zen 4 is now
Well, I use i5-6500 to this day and thinking about upgrade. I wonder if getting AM4 board is good move, to avoid buying new DDR5.
AMD marketing is dropping the ball quite a lot... they are responsible for all self owns from the last 10 years and they are still not being reorganized... failing leadership imho...
Marketing must have something on management.
I find it funny you saying they're responsible for all the "self owns" in the past decade.
Aren't they the only owns able to "self own" themselves, you know otherwise it wouldn't be a "self" own?
No this isn't a marketing thing, this falls on the CEO, it wasn't ready and they knew it, the marketing did their best to make a turd look golden, it isn't their fault that AMD decided to launch a buggy product before it was ready.
MLID even said they were 'sandbagging' the performance numbers 😄😄
@@arbpaninken6719yeah they are "sandbagging" the performance so much you'll only get the performance after half a year 😅
"AMD does tons of dumb stuff all the time.."
We all knew this but its good to hear you say it Tom.
The 9950X3D NEEDS dual 3D-Vcache CCD's its not optional anymore for AMD. Gamers want to buy the 9950X3D specifically if it has that feature, and full overclocking support. The Ryzen 7 9800X3D will sell ten times more than the Ryzen 9 X3D product line, if the 9950X3D has only one 3D-Vcache CCD.
I cant help but feel disappointed
Amd never misses to miss an opportunity
They had intel on the ground and instead of delivering the final blow they strike their own feet
Why do people root for final blows to competition? I’m fine if you are rooting for one company over another, but without each other we all would suffer.. do people already forget what things were like when AMD wasn’t putting out competitive products?
@@trailblazinn We would have had a live instead of gazing to a computer monitor?
@@trailblazinnbruh. How did you get that. Think more like “two top companies pretending to compete to fck over customers” instead of “make the best product and win” it’s that type of thing where AMD could have delivered value and stability where Intel had the opposite
Ah well, so long as it gets fixed, all the bad reviews will only make them cheaper.
Glad someone make some sense...
THIS!!! F##ck yeah, bad reviews help our wallets, prices will fall like meteorites. Man of wisdom.
Life is too short, why not buy now?
@@jorgevaladas788 Uh, because there's nothing wrong with holding onto your current system? That's a wild reason to buy a CPU of all things.
AMD about to make this their 28-3 moment if Intel can launch a competent lineup that doesn't eat itself while heating your house in the winter.......
With new releases it really can be nice to just sit back and wait for a couple 2-3 months before you upgrade. Let a couple bios updates come out, lots of reviews. I mean its your hard earned money. These are hundreds of dollars. It really won't kill you to wait a few months.
I'm overdue for a rebuild, but I'm kind of stuck with the 5800x3D as the best I can get on AM4 without any clear CPU of the future that is blowing me away right now. I'm going to be monitoring the next 3D chips but may just wait to see what Zen 6 looks like down the road. The 7800x3D would clearly be an upgrade, but one that would require a full rebuild with a new motherboard, RAM, ect., which is not worth it to me right now. I'll wait until they start to get this new architecture and software sorted.
@@AKSBSUIt's already clear that teams didn't have enough time to work on Zen5 & changed frequently, this isn't going to improve significantly.
Any hope for more than 5-10% improvement in anyway is futile. Wait for November/December & it's likely that the cheaper Zen4 X3D will be more appealing than zen5,
there is only so much you can fix a born deformed baby, the fixing should have been done when it was still developing,
now it's mostly hardcoded in the architecture & scope of improvement is mostly locked up.
@@AKSBSU The whole ram thing slowed me down too. I'm a dev mostly and ddr5 is better but it's hard to get motivated. 10% can sound like a lot, but it's only 10% faster on the edge when you need it and more modest benefit in my workloads.
It's always a function of cost; when it's a big jump like that usually you want things to be a lot better. It makes complete sense.
2:37 Gamers Nexus covered this today in their review. Running a 9900x or 9950x requires you to do a bunch of garbage to get core parking working, or something. It's a mess on Windows, and you need to follow their review guide, or you get really weird results.
Windows is a mess
did you know that Win 11 has code from ALL previous versions in it?
you can literally run Vista or Win2k native programs from Win 11 itself.
Windows became a clumps of code glued together pretending to be a new thing
This is exaggerated by CZcamsrs to get views. Install the drivers from Amd, and it just works.
Core parking is key in two main scenarios: game core ccd allocation and power saving.
@@SickPrid3 yep indeed it crazy too. Win11 base code is the original windows tweaked and updated. That's why it's straight shit.
@SickPrid3 Perhaps, to your surprise, I worked seven years at Microsoft 😊. And you are correct. There are a lot of legacies here.
@@imad1996 apparently some things make a 50% difference. Might be more than just core parking. Maybe something else changes.
Also I have another question, why do we need to park NON-3d v-cache cores with Zen 5...when we didn't with Zen 4? What is going on with that? Nobody answered this question in any reviews
I guess instead of binning chiplets into different tiers, they can mix a good chiplet and a lesser quality chiplet into each cpu, getting higher gaming and single core performance without impacting all core performance. But it needs correct scheduling...
so we went from "AMD is hiding peak Zen5 performance"
to "AMD to blame for bad reviews" & "AMD must reboot Ryzen"
2kliksphilip was right.
I really wanted to jump onto zen 5 as im currently into a 1600x, but man this really messed up. AMD had the market in its hands with 13th and 14th gen intels problems. Now Im waiting for arrow lake to even see whether I would take AMD over intel. Heads are going to roll at AMD, or they should.
I'd take a 5700x3d as a stopgap to Zen 6
Simple solution (IMHO): Get Jim Keller back and in charge of the CPU teams.
When an AMD product disappoints we resort to good old “don’t worry, finewine!”, what a classic 👏
We are the beta testers for AMD.
Only if you sign up for the early beta test. My personal rule: always wait at least 3-6 months before buying new AMD products.
Intel has problems with their prev gen while AMD has problems with their new gen. If Arrow Lake delivers what was promised, AMD will be in trouble.
My guess is that the infinity fabric is a bottleneck. Better scheduling might help to mitigate this by reducing inter-ccd communication but it will not resolve the performance penalty when inter-ccd communication is unavoidable.
It’s about time AMD start working on a generational improvement to the infinity fabric of their CPUs.
But it works fine with linux
@@srinathshettigar379 No, it doesn't. Linux performance isn't significantly higher for Zen 5.
Jufles said that they had to change the IO die.
I like the new intro music
Reserved for when I am not happy...
I guessed it by the music.....Tom is not happy.
LLOOLL! MAKE SENSE NOW 😂😂
@@MooresLawIsDead Like that electric guitar rocking hard. keep it
Tom, your whole concept of dual V-cache R9's continuously misses why its not an option. AMD released new chipset drivers that introduced core parking to non X-3D Zen 5. The reason for that is because it has always been a problem for a certain number of games where inter-ccd communication hurts performance going all the way back to Zen 2. Nothing about dual X3D fixes that until game naturally need to address more than 16 threads. The only reason to get an R9 in general is if you lean toward productivity. That doesn't mean you can't be a gamer as well that can benefit from 1 CCD of Vcache. The performance gained by X3D generally out weighs the 8-10% loss from the clock decrease of the 1 stacked CCD. If you are losing 16-20% by stacking both it makes the product more pointless. Your all or nothing take on gaming vs productivity is really missing that point.
I'll explain it this way. I'm nerd with a sysadmin background. I do a number of things that includes Linux and virtual machines, and I play flight sims and various games that very much appreciate the cache. I've only seen star citizen using 60% of a 5950X which is approaching 20 threads with a 6800XT. The 7950X3D is so much faster that it may kiss 50% with an even stronger GPU to feed. That is not anywhere close to being most games. With the 1 stacked CCD approach, you can main Linux, pin the stacked CCD to a Windows VM which will cover almost all of your gaming needs, and still have cores that reach 5.7Ghz for everything else.
Yep, people who want 2 3d ccds are cIueIess
Your last sentence there is exactly the main use case I was thinking of for the 1 stacked CCD approach. Having Windows run on a VM specifically configured to only use the stacked CCD is the most guaranteed way to get around all the awkwardness of that design for gaming
@@spralwers I had started doing this way back with a 3900X on Proxmox, and I had other Linux VMs/containers handling various services, while using Windows just for gaming.
AMD with a masterclass in kicking yourself while your opponent's down
Code compilation performance is excellent it’s fine for devs. Microsoft may share some blame here since it works fine out of the box in Linux.
Or maybe they have different teams handling Windows and Linux.
it doesn't work fine in Linux, it's just works slightly better, but that's not saying much, if it's not a server workload/avx task
@@MaxIronsThird Phoronix's results look good to me all around.
@@MaxIronsThird ~18% geometric mean improvement over the 7950x on close to 400 (GNU/Linux) benchmarks sounds fine to me. I can upgrade my 5950x rig to a 9950x and my LLVM compilation/linking times will be cut in half.
@@oceanusdigital3621 if you want the CPU for server, ai or avx tasks then it's better, for all the rest it isn't.
And THAT is the reason I don't jump on new products right from launch.
I always wait about 3 months (the sweet spot IMO) to let the problems or issue ironed out.
or 2 years if its intel
@n8spL8or never if it's intel gpus
Hell I wait at least a year unless it's on a really good sale or something
I would say 6 months after the 9550X3D release, once all the major issues are convincingly worked out. So maybe 2 years from now...
I buy prev gen to save money. Always with GPUs
Unfortnately as an engineer I can't agree with most of you. I beleive the guy who forced zen 2 team to design zen 5 with zen 4 code base should be fired. Zen 2 team had a very talented chief architect: AMD fellow David Suggs. Those people who blaimed AMD not punishing him, who are you? You are far lack of knowlege to comment on this. If Dr. Suggs is the chief architect, that means he should make the decison on his own. Zen2 created a new chapter for AMD, why wouldn't his team reuse their code base for Zen5? Let me tell those people who asking zen2 team to fxxxoff, David Suggs left AMD and retired since then, this was AMD loss you pathtic
I would accept the microcode/software fix not being ready if this was released 12 months after Zen 4...but 23 months?!?! Come on...
I actually find this concerning. AMD just got handed the biggest CPU win in history with Intel flagships falling on their face with voltage instability issues and they pull this abortion? Zen5 was supposed to be the new architecture the next few generations Ryzen would be built on. If it can't even match f*cking Zen4 on some cases that's a massive issue. The vast majority of users don't use AVX 512 threads.
Intel handed AMD an easy win and they just opened the door for Arrow Lake to stomp Ryzen into the curb.
This seems very much like the incompetence epidemic that has hit most industries after they start replacing vital employees with Indian hires to cut costs for the stockholders..
Maybe they were losing Zen4 sales to people who were holding out for Zen5 so they went "ok here it is you can buy Zen4 now"
Lmao, that somehow makes the most sense to me
Phoronix got an 17.8% uplift average over all their tests in Linux, so this can´t be a hardware problem.
Happy I switched from Windows many years ago.
the CCX latency issue is CRAZY
Ever had a look at the chosen benchmarks? They just tested different stuff than all other reviews
@@lukas_ls Of course, but do you really think that the dozens of different workloads Phoronix runs would magically be the ones that performs well ?
@@ddelin100 Possibly, yes. Just look up what they were testing. A huge part of their performance tests included AVX512 benchmarks (I'd say about 10-20%of benchmarks show significant scaling with AVX512). Given their changes to AVX512 (first full AVX512 implementation), this drastically impacts average performance. AVX512 is fine in general but doesn't really apply to the real world. Rocket Lake had up to 2.7x speed up in some AVX512 workloads over Comet Lake.
In benchmarks that are both used on windows and Linux (like blender), performance is in line with what others found. There the 9600x is only about 14% faster than 7600 and the 9700X about 8% faster than the 7700 (both 65W parts).
That's in line with what GN etc found on windows.
That is what happened when you try to change the micro architecture without Jim Keller.
He's more of a hardware architect no? I would imagine he mostly designs the chips. But then again, he said in an interview that often his job is to make the tough calls and nuke an architecture when it becomes a tangled enough mess and to set teams on the right path to go forward from. Sounds like they sure need something like this at AMD.
Jim Keller only worked on the first generation of Zen. The Zen 2, 3, and 4 architectures have all been decent uplifts without Keller being involved. Zen 5 doesn’t look like a big uplift in gaming but its expanded AVX-512 instructions support and wider front end will very much benefit the datacenter.
Keller designs the team, not the chips, Apples chips are still good and that's because he got a blank slate to create a top team of engineers.
You don't bring in an Engineer like Keller to build a chip and then leave you hanging after he leaves, he comes in, restructures the Engineering team, so they can keep churning out products, Keller isn't a magic silicon wizard who can magically manifest a uarch that's better than the previous generation.
Zen 1 wasn't designed by Keller, When you make a CPU design you have to predict the future of software design and how people write code and also run simulations on existing code and see where the bottlenecks are, this isn't a person problem, it's a "how do we structure the team so it's really good at finding bottlenecks and building around them" problem, Zen 1 was already in the mind of someone at AMD, the Issue was trashing the dogshit philosophy they had when producing Bulldozer, AMD actually produced a Bulldozer chip with 40% IPC uplift near the end of it's life, but no one noticed.
Zen 5 being a clean sheet, also means it's a longer projection, higher risk gamble, Zen 1 was pretty bad compared to Intel at the time as well, but getting lead times down from 5 years to 2 means they can be more align and flexible and adapt to new bottlenecks, and silicon issues, this is why Zen 4 is so good, because they had a good foundation to work off, Zen 4 was designed by the team, on top of the foundations of Zen 1, you can either stagnate like Intel did with Skylake, or AMD with Bulldozer, or take a slight risk by doing a clean sheet design that lay out the foundation for the next 4 generations, over 2-3 node shrinkages.
@@woobilicious.Finaly someone who knows without pretending to drink leaks.
@@samuelrodgers2742 Jim Keller's designs were used up until Zen3, and Zen4 is mostly just a frequency boost and introduction of AVX512 instructions, Zen5 was when the Zen team had to go on their own without his guide for the first time.
Same seems to be the case for Intel, but it seems like Keller left before everything was fully sketched out, that's why MTL underperformed, and why ARL doesn't have adamantium or Rentable Units, only thing that seems to be going as planned in Foveros.
In the olden times, people knew not to buy the new chip the day or week it was released. Traditional broadcast and media outlets took their time and were careful not to burn their bridges and the people building systems valued stability over performance. These days, it appears that we expect everything to be immediately flawless, overclocked, liquid-cooled, and RGB'd. It's a different set of values. I'm not saying either set is right or wrong, but the changes are worth consideration.
Tbf zen 3 launched fine and zen 4's issues were due to overzealous board partners. We havent seen this level of fuckery since zen 2's launch and x570's usb dropout kerfuffle.
in the olden days new generations were coming out every 15 months.
I'm not paying hundreds of dollars to beta test AMD's processors. 7000 series it is.
I got to upgrade anyways, so I will buy the 9900X. If you are on an older Intel Gen (Coffee Lake-S) anything recent is an upgrade. Also I am using Linux, so don't really care if the issues are isolated to Windows. Got to research the Linux performance a bit though. Edit: Per "phoronix", Linux performance seems excellent!
New intro is absolutely sick and that cut was perfect
you feed us those fake rumor for last 2 years
Also that snap cut at the ending lol
The Alex Jones of tech hardware. But the discount version.
I got my 7950x3D back in January and after a certain Windows update i've noticed that core parking was basically disabled in the Balanced power plan profile, no matter what value i was setting the "Processor performance core parking min cores" setting to, (which is normally hidden) it would ignore it, except for the Power Save power plan.
It made no, sense, until...
...After doing what Gamer Nexus suggested with that command line, core parking settings are no longer ignored, that's with the latest chipset driver and having set preferred cores to Auto in the BIOS.
Normally i don't use core parking in almost any power plan and while that's the opposite of what people have been having issues with these specific CPUs i thought i'd share this.
AMD's Rocket Lake.
Nah, isn't nowhere near as bad as Rocket Lake, it would be a Kaby Lake moment.
Zen5 was made to do in the server/workstation market what x3d chips did in gaming. Phoronix not just runs their tests on linux, but also has a more diverse test suite than just running cinebench as a "productivity test". With low cost am4 x3d chips on the market, gamers should essentially ignore everything else. Just buy the best x3d you can afford, and ignore the marketing slides.
Once again petty office politics has derailed a product launch and this time it's worse than marketing slides.... no doubt heads are gonna roll on this one. *I'm referring to both AMD and Microsoft* .
Unless the huge latency between CCDs can be fixed, having v-cache on both CCDs sounds like it would provide very limited utility. You're going to want games to stick to one CCX anyway, so why put v-cache on both CCDs when it probably has to come with at least some sacrifice to clocks?
The only reason that core parking is needed is because of the huge latencies. Using zen2 code base instead of zen 4 is probably why the latencies are so huge; they're doing things the dumb old way. So, once you tune the code to properly shuttle data between CCDs having Vcache on both CCDs would make a 9950X3D the undisputed king of code compiles and (some) games.
Thats what i am saying since forever. V Cache on both CCDs will not change anything. It might even get worse (because you also would have less Frequency and you will cut that away too if your Game would need that).
Its totally fine to have only one if it works. Issue is it doesnt. AT ALL. Its like a Chance if it works or not. Like…
Well. As long as we talk about Games that is obviously. I dont know about other Apps that need more V Cache. I understand thats a Thing too and i am not against AMD making Chips for them with 2 CCDs V Cache. But for Gaming it would be fine if it would work correctly.
@@flimermithrandir For me, if the penalty is
@@JeffGraw And that’s an Ok Idea but you will still loose (a little bit) Performance with that Approach. And it will be more Expensive. It will show better Results in Charts but the single CCD CPU will still be better imo.
I don’t want to sound like an Asshole. Really, I don’t.
I just don’t see the improvement. I also don’t see how it would help with Energie. If both are activ, there is a Chance it will be worse. Even if not and Coreparking is still a Thing (which it should) it won’t do anything either.
I would love to see it so we can test it. But I just think most ppl are thinking V Cache on both is a silver Bullet for Gaming and it will outperform the 7800X3D by another 15% and that’s just not true at all. It may still be worse than the 7800X3D.
Hope you understand. It’s not against you personally.
I want more than 8 homogenous cores on one CCX within a single CCD.. Shame no such CPU exists. Intel maybe in Bartlett Lake has one. But Intel come on make 12 Lion Cove cores on Arrow Lake please. That way 12 cores on same tile and good latency. Not the stupid Big.Little crap.
Given that Zen5 with the same microcode performs so much better using Linux and a pretty new 6.10 kernel than using Windows I tend to say that either Zen5 is either optimized for server workloads (hint: Epyc) or Windows is just not ready for Zen5. I lean towards the latter. Yes, it was a bad launch but most of the audience here is waiting for X3D and maybe lower prices anyways. So give them some time. All of our CZcamsrs will tell us regularly about the progress, I'm sure.
It has nothing to do with windows. Just have a look at the chosen applications:
They testet:
Web performance (mostly js compiling)
Code Compiling
DAW+3D Rendering (in total one of 14 pages)
CAD
Synthetic HPC benchmarks (benefit a lot from AX512, there are more AVX512 benchmarks than DAW+3D Rendering+Gaming combined)
Database performance
Library tests (Python, numpy, php etc)
AI Benchmarks (again strongly scale with AVX512)
Phoronix didn't find better performance on Linux than Windows, they found better perormance in a certain set of apps.
@@lukas_lsyou don't understand how scheduling works. It doesn't matter the type of test, if windows scheduling sucks, it means all tasks using more than one ccd will have increased latency which means decreased perf. Even some non multi ccd tasks can be worse.
And then you say "phoronix results are the same except when they are different"
Like bro.
@@putneg97 Then proof it by running the same tests both on Windows and Linux.
All tests by phoronix that also run on Windows aren't signifantly better than tests on windows.
@@putneg97 Average fan boy comment. Just look at benchmars done both on Windows and Linux. There is literally no benchmarks that shows signifantly better improvements on Linux than Windows.
There are lots of benchmarks done both on Linux and Windows. Examples: Geekench and Blender.
Just look at their open databases, you'll find no significant difference in performance improvements.
Please explain how I don't understand scheduling and how you'd explain this.
What I hear is, skip Zen 5 completely. The whole microcode base is a disaster than cannot be patched.
This feels so similar to the RDNA3 launch that it's as if AMD learnt nothing from it.
I wonder if the inter-core latency test that are flying around the internet are the same if they are ran in Linux. AMD's next gen HAS TO work on lowering latency/speeding up the IF
This is FSR 3 in CPU form. Late, still unfinished, and not worth the wait.
The core parking on the 9950X and 9900X is strange, and the wattage on the 9700X is odd. For the X3D, they better not fumble with the 2 CCD V-Cache, no core parking🤞 , and have overclocking. It's unfortunate that AMD has an opportunity with Intel down, yet they seem to be messing it up.
I have to wonder what Lisa is actually doing daily at AMD. There is such a clear disconnect between engineering and marketing. I don't know if engineering is handing marketing legit numbers and marketing then makes up new numbers or vice versa. We saw the terrible marketing with the pointless 5900XT launch, which proved to be 1% faster than a 5950X. Lisa needs to get both parties in the same room and have a very uncomfortable conversation. Is she respected?
AMD's new slogan: At Least We're Not Intel!
Wait a minute. There should actually be no core parking with 9900X or 9950X because the feature was only there for the X3D variants, so that the game uses only the X3D CCD. But the regular 9900X and 9950X don't have X3D so core parking shouldn't be necessary, right?
No. Core parking has been a thing much longer than x3D. It’s pretty simple: if you don’t have enough demand to keep N cores busy, you can save energy and heat by parking unused cores. Chiplets and x3D complicate this practice because it matters which cores you park.
@@AbeDillon But aren't cores going to low power state anyway, if they just do nothing? I always thought that core parking was invented for prefering the 3D Cache chiplet. I never heard about core parking problems back when I had a 3900X or later 5900X.
It feels to me that if this was always going to be the case for launch, they should have changed the core counts meaningfully.
The cross CCD latency is a very suspicious regression. I’ve seen HWINFO screenshots where the second CCD is clocked much lower. The whole things is really sketchy
Also inter core, that is a bigger issue
The clockspeed difference is probably intentional, AMD is pairing a higher silicon quality CCD with a lower quality one.
Second core complex is purposely chosen from lower clocking dies.
@@lharsay And they have been doing that since Zen 2.
@@lharsayOMG, they should be the same quality at least
Dude I’m sorry but you need to admit you were wrong here. Don’t hide behind someone else’s work in order to redirect scrutiny to them (like Wendell). You were wrong about the performance uplifts, and then you were wrong in suggesting these parts were simply underpowered or underprepared (by not shipping faster RAM, for example, which has absolutely no effect on performance for Zen 5), or that PBO wasnt tested, or that there’s some mysterious bug at the root of it… yes-there are minor discrepancies such as the ones Wendell has pointed out.
But no, these parts aren’t going to change drastically after a microcode update. Admit you were wrong and move on.
What is zen2 code base? Is that microcode or drivers or something else?
It's the underlying foundation that is relied by the software stack. That should have been at least Zen4... not Zen damn 2 .... petty office politics. I've seen the b.s that crap causes and it derails everything.
Architecture.
Ie the logical blocks of transistors that make up part of a core.
Eg the fpu is a logical block of transistors.
There can be different versions of an fpu which achieve the same results in different ways with different speeds.
In theory, you could take an fpu from the original amd x64 and slot it into a current gen design.
You wouldn't do it, because it has surely evolved long past that, and the performance would be terrible, but it should theoretically work.
The implication in the claims is that instead of evolving from the Zen5 blocks, they rewound to Zen2 and evolved that.
It's kinda like your grandparents having a child when you're already an adult.
You're older than them, but they're technically an aunt/uncle.
It's awkward, confusing, and there ain't nobody who wants that crap.
Having two VCache CCDs is pointless unless you can make sure that threads are not bouncing between the two CCDs, since the latency benefit you get from storing stuff in the cache would be seriously reduced if not entirely eliminated by the latency of moving across the Infinity Fabric connection to the other CCD.
The way things are going with amd and intel, prob will be using my 5800x3d for ten years.
It's like my 2600k, rocked that for 10 years. Gaming was garbage towards the end though, as was the memory bandwidth on DDR2.
That arrow lake is suddenly starting to look like a good option huh
@3:26 a CPU benchmark at 4k? 🤨 Looks GPU bound to me.
And one more bridge AMD is burning people with is not patching a critical vulnerability in Ryzen 3000 parts that are still very much in use.
Exactly, I feel like amd said screw the prior generations of their cpus on purpose to force people to upgrade when they really don't have to just so they can make bank from it.
They walked that back and are patching 3000
03:04 Doesn't it just mean that Windows kernel and drivers are comparatively worse than the Linux ones?
It's a shame because I need a new PC and work in a lot of "productivity" apps that benefit from a chip like the 9950X, but the game bar and core parking and just general nonsense is such a turnoff. It may not be AMD's _fault,_ but it _is_ their problem.
AMD needs to put 3d v-cache on both dies of the 9950x3d. All is forgiven if they do that.
Edit: I wrote this before I finished the video. Tom is right on with how he ends the video!
and on the 9900x3d
Would be funni if Zen 5Cs with 3D v-cache lol
But that wn't happen
@@samson7294 💯
I have a 7950X3D and I don't understand why people want V-Cache on both CCDs. Hear me out:
Tie games to the V-Cache CCD: 7800X3D performance
Tie games to the vanilla CCD: 7700X performance
Tie games to 4 cores from each CCD: AVG FPS goes 20-ish % below the 7700X/Vanilla CCD and the 1% Lows get laughably bad.
What you DO NOT WANT is cross-CCD communication. Slapping V-Cache on both CCDs WILL NOT fix that. AMD needs to bribe Valve and other game stores into forcing the CCD lock right from the launcher or some shit.