Nvidia Knows This Sucks
Vložit
- čas přidán 24. 07. 2023
- Nvidia will not be releasing its own version of the 16GB RTX 4060 - or providing any review copies.
Watch the full WAN Show: • Sorry For The DDoS - W...
► GET MERCH: lttstore.com
► LTX 2023 TICKETS AVAILABLE NOW: lmg.gg/ltx23
► GET EXCLUSIVE CONTENT ON FLOATPLANE: lmg.gg/lttfloatplane
► SPONSORS, AFFILIATES, AND PARTNERS: lmg.gg/partners
► OUR WAN PODCAST GEAR: lmg.gg/wanset
FOLLOW US ON SOCIAL
---------------------------------------------------
Twitter: / linustech
Facebook: / linustech
Instagram: / linustech
TikTok: / linustech
TikTok (LMG Clips): www.tiktok.com/@_lmgclips_
Twitch: / linustech - Věda a technologie
Can't wait for a RTX 5060 8gb that gets the same FPS as the 4060 and 3060.
With DLSS4 that can add several frames between frames that were actually calculated and then scaled up :-D
@@jochenkraus7016RTX6060: renders 1 frame per second, invents a 60 frames movie with a neural networks to links each frames.
@@Tetemovies44gb of VRAM in the 7660TI, DLSS5 is offloaded onto the CPU.
Starting at $650
@jochenkraus7016 and they lock dlss4 to it even though the 40 series could run dlss 4 fine
Even if Nvidia lowered the price there's still the awful 128-bit memory bus which is why the 3060 Ti pulls ahead in some benchmarks with its 256-bit memory bus, 40 series is just all screwed up.
thats bwhy whenever i buy a card i ALWAYS look at the bus speed as much as cuda and ram, ive a 2060 6gb, and the 2060 12gb doesnt have any advantage because the bus speed is lower so kills any advantage the extra ram has, (maybe apart from textures)
the 3060 has a 192-bit bus not 256
@@rickgear2579 3060 ti
@@rickgear2579 he did say ti lol
40 series was made to be suck so they don't compete with the abundant 30 and even 20 series due to crypto crash.
Unfortunately, I don’t think Nvidia gives enough of a crap to have learned anything from the 40 series cards. And by Nvidia, I don’t mean the engineers or marketing team, I mean that one guy with the leather jacket who makes impulsive decisions that screws over all of their board partners and customers simultaneously.
for sure the marketing team is part of it, Jensen and the other execs have them held by the balls and they force the engineers to minimize cost and maximize profit
Marketing team is 100% a part of the problem. Who do you think came up with the early access 4060 ONLY ON CYBERPUNK review?
Linus is right that the only two SKU's worth looking at are the 4090 and 4070Ti though. The rest just aren't value for money at all
Buy AMD, Nvidia can care less about you
@@lordy1952 that doesn't mean AMD is doing good, mind you. Let's not play ball - AMD is just the lesser evil right now.
RTX 4060 is significantly cut down from 4060 Ti because the 4060 Ti is really a 4060 and the 4060 is really a 4050 Ti that they have just misbranded.
Misbranded implies a mistake. This is deliberately a scam.
And next generation will surely get a noticeable performance bump compared to this one. And that will make it look like a good price. But on the different price tier. So everyone will be willing to buy **60 card at the price of **70.
Strictly speaking, it's not even "cut down".
The RTX 4060 is a full die, but it's AD107, while the RTX 4060 Ti uses AD106. Until RTX 4000-series, Nvidia had *never* used a 107-die for above the 50 Ti GPU tier. The RTX 4060 is unambiguously a misnamed RTX 4050 Ti.
Explains the 1080p target
That's the thing. The 4060 TI architecture is weak. The 3060ti was based on the 3070. It was the same card with the same higher bus(256), but something on the card failed to meet the level to become a 3070 so it was disabled or it couldn't reach and hold the stable core frequency required of a 3070.
The 2060 Super (there was no 2060ti) was the same thing. It a 2070 with some things turned off.
At best the 4060ti should probably be a 4060 and what's being sold as a 4060 is probably a 4050(or even more of a mobile GPU). The whole product stack this generation is off by 1 or 2 levels per card.
EVGA saw the writing on the wall before the 40 series launch. I love their products, and I'll miss their step up program in particular, but they did the smart move and ejected from this Nvidia disaster.
I never had a EVGA,now I'm sorry didn't. Since their Smart Thumbing of NVIDIA, I took the time to REALLY dig-in learn all i could about them and have been mindful of Anything that NVIDIA does concerning how it treats its AIO Board Partners and stepping all over both them and the public. EVGA was like Kicking the Horse in the Teeth !
Yeah and now the company is close to dead.
I love their power supplies, and I have a watercooled 1080ti I'm using.
I remember when the GTX 970 (typically sold for about $300 before the 3.5GB scandal), also came with 2 AAA games, splinter cell blacklist, Arkham City, and one crappy non AAA game.
I got the Witcher 3 with mine in Sweden (and Batman) :P Got 400 hours of entertainment from just the game itself :P
Splinter cell blacklist is a crappy game tho, so it should be in the end with the "other crappy game".
@@andersbjorkman8666 Same, but I only got The Witcher 3. crazy how it varies depending on where you live. Thankfully got the Batman game free on Epic year later.
...and after the shitstorm on Batman they gave all the titles of the series on steam
@@shawnpitman876 splinter cell blacklist is fine
It almost feels like GPU manufacturers are trying to earn all their revenue from gamers now that mining is almost dead.
they can't exploit gamers when gamers are unable to keep their jobs in mcdonnalds due to all the ROBOTS
The sad thing is that, in the _current_ market (as of this writing), NVIDIA could afford to just abandon the personal computer market _entirely_ (granted with some amount of pain).
The big corps are paying giant money for AI stuff, buying reams of the already ludicrously overpriced high end workstation/server cards (like on the order of $9000 for the very high end; they make the 4080 and 4090 look like bargain deals). And the AI craze is letting them overcharge _even further_
This would let their GPU depts stay afloat on selling to "big AI" alone.
They don't need to care about consumers right now
@@arcadealchemist welp same with creators where art is involved lol
@@arcadealchemistthis what happens when you lobby for higher wages for a non-skilled job when the given market can’t handle the increase. case and point on what you said, companies will spend 3-4X times the cost of workers wages and benefits because the ROI is known in terms of time frame for robots. then they will employ a skilled technician at much higher wage to keep them going.
Gamers aren’t the ones fueling the margin on the GPU market and quite frankly NVIDIA has already basically all but confirmed this - it’s multimedia studios and other developers that basically necessitate having extremely powerful rendering stations as well segmenting NVIDIA GPUs in AI development that are keeping this price high.
Gamers have long been abandoned now since the pandemic.
Looking forward to the 50 series! Can’t wait to not upgrade from my 1070ti because of ridiculous pricing again!
10 series gang staying strong. My poor 1080 has had such a hard life but it’s still hanging on.
buy an AMD 6000 card in your budget and be happy
I got a 1660 Super and it is doing just fine with what I play. I imagine I will eventually buy a used 3060 maybe 3 to 4 years from now and be fine for another 5 years.
Well you can get a RX 6800xt for really cheap. Some RTX 4070 are also aviable for 550€. I got a 4070 and it was such a huge upgrade from a 1070 ti
@@talltale9760clean and repast!
In 1987 I bought a MacIntosh II. One year later I walked into the original Fry's Electronic in Silicon Valley and bought more VRAM to get 256-bit color ...
I mean the answer is very clear! Nividia doesn't care about gaming anymore, they've moved onto AI..
nvidia have been mining crypto for 3 years now directly to jensens wallet. regulators dont even know what a digital wallet is so they cant fine him
4060 is now the "50" class of Nvidia cards, which makes the 4050(is there one?) a Intel UHD graphics level card
There's no card called 4050 yet and I don't know if there are rumors.
But if it was a single slot, slot powered card with let's say still 8GB it would fill a niche for ITX and other small systems.
@@jochenkraus7016 AMD's APU would be the superior option in niche ITX systems, unless people want something very specific from NVIDIA.
@@tuanld91 itx will still got the PCIe slot for a LP graphics card anyway, presumably the soon releasing Gigabyte 4060 Lp
Imagine it being called the GT4050 with 6 GB of ram on a 98-bit bus.🤣🤣🤣
@@HarmonyEdge Still more than the RX6400
I just got rid of a 3070 with 8GB specifically because of the VRAM. It was at the point where the card had the performance to play games like RE4 for instance with ray tracing, but couldn't do so due to memory constraints. With that in mind, everything I've seen and heard about the 4060 makes it sound like a seriously bad time.
I get about 90 fps in resi 4 with high or ultra settings and ray tracing with my 3070. You must have a gimped set up or version
Sounds like you need a 4070 then...
I personally wouldn't even consider the 4060 ti 8gb compared to the 4060 ti 16gb because I'm simply not going to buy an 8gb card in 2023 no matter how cheap it is.
But I would probably just get an RTX 4070 and forget about the 4060 ti altogether.
@1O1O11 I just did that. I had to completely replace my computer and ended up getting a 4070 even though it was significantly more expensive than the budget I'd set myself. It was a good choice, that 8gb is already too low in some games. Even the 12gb on the 4070 is cutting it close in some of the more poorly optimized games I'm learning
@@1O1O11That's the Nvidia end game. If this convinced you to buy a 4070 then succeed in upselling.
@@slayerdwarfify Sometimes you just have to bite the bullet and overpay when you need you PC for work and such.
Kind of shitty on Nvidia to basically make having enough VRAM only available at $500 and up, but I've never been able to make any impact on stopping big corporations from being greedy.
Though when I develop indie games as a hobby, I always test them on a GTX 1060 laptop to make sure I always optimize my stuff.
Nvidia is making a killing with professional cards for AI. Those have much higher profit margin than the gamer products. It seems Nvidia is only doing the bare minimum not to lose the gaming market, but is not interested in shipping a lot of chips in that segment.
Why bother with gamers when they can have much higher AI margins. Nvidia should just announce their withdrawal from the consumer market.
Gaming is like 40 percent of their revenue though. They are plenty interested in that.
@@zealot00 The 4060 16gb is for both ai and the consumer market. They are positioning themselves because they have no competition.
@@illyaeater For the AI market they don't use consumer GPUs, they use GPUs designed for servers, like the H100 that has 80gb of VRAM and costs about USD 40k.
@@MarcusBuer I meant consumers who are into ai. You can comfortably run and train almost anything on 16gb of vram other than llms, and amd has nothing for windows
Considering how the barrier to entry for hardware required to run inference or train LLMs locally is dropping fast, I'm not surprised. Nvidia is protecting their AI card market by not giving a budget option for companies.
THIS, totally this.
That is no excuse for their non budget cards to suck this badly.
@@cidiracing7481of course, but it's not like nvidia cares about us
@FluorescentGreen5 Well if they want to give more market share to AMD so be it. At this rate I rather upgrade my 2070 to a Radeon or used 30 series nVidia.
Uh, the AI products are HIMARS, the GPUs are pop guns.
They didn't used higher density memory chips, they just soldered more memory on the board, on the other side(aka the backside) of the memory chip not on top of it or next to it. "Clam shell design" it's being called, so they don't have to increase the memory bandwidth this way. The 128 bit going to age so well!
It already feels aged and it just launched. It's still so wild to me that not only is the bus width even smaller than the previous generation 3060ti, but it's also smaller than even the 1060's 192-bit bus. You have to go back as far as the 900 series, 4 whole generations ago (and that's excluding the 1600 series), just to find a 60 series card that has the same memory bus width as the 4060ti.
This new generation is like taking a single step up a set of stairs before falling back down the previous ones, all while nvidia charges you extra for the fall.
@@rubetube7084 well actually, while writing your same point on another comment I saw that the 3060 8GB had only 128 bit bus (down from 192 of the 12GB version), so yeah they did this shit one year ago too 🤣
@@Daniel-yy3ty yes but that's the 3060, not the 3060ti. The 3060ti was 256-bit, which is pretty good and I think 128-bit for the non-ti 60 is not great but more acceptable than the ti
@@rubetube7084 gtx560 had 256 Bit bus :/
@@rubetube7084 The 4060's still beat the 3060's in real gaming performance while using 100W less despite the low bus.
The significantly lower power draw and less heat output is the best part of these cards.
Nvidia and AMD got complacent after the crypto mining era.
china got greedy they made the profits but didn't make the plans to increase the R&D
ughh china!
Wouldn't they just transition to Ai I know Nvidia is Selling Grace Hopper HPC cards for $100K a piece ATM.
@shadowarez1337 anyone still complaining about miners in regards to gaming gpus is still smoking crack. Miners, especially real miners, have dozens of better ways to mine coins. Using current GPUs to do so is dumb.
And it's not really miners faults, gamers showed Nvidia and AMD they would pay high end prices for mid range cards
I firmly believe if people just held out and didn't upgrade (I'm being a hypocrite saying this since I spent $900 on a 6700xt in 2021) we would be in a vastly different situation
My 3080 10GB has started showing its shortfalls over the past 6~ months. Really fortunate I went 1440p over 4K because I'm *_already_* hitting VRAM bottlenecks in select titles.
The VRAM issues are mostly Nvidias fault but I would also say that many AAA developers just lost it. There are games that are so terribly optimized with insane texture resolutions for no reason. I personally am happy with my 3070 ti at 1080p and sometimes DSR at 1440p, depending on the game.
But there are still some developers out there that make games look good by using clever tricks and artstyles to mask lower resolution textures or the absence of Raytracing. But there are many more out there that don't care. They have unlimited options and can go as far as they want, in the end the consumer has to pay to be able to run it at certain resolutions.
Really makes me nostalgic for older hardware, where the video game console was the limit, when developers had to be creative to make seemingly impossible things possible for their games. Today you just need a ton of money to be able to have the "best" experience.
I have a TUF 3080 10GB and I turn down graphics now lol, also, if you use 1440p DLDSR to 4k, you can lower or disable AA which will free up some VRAM and have minimal affect on graphics.
@@PinePizzamaybe AAA developers are building their games for consoles that generally have 16gb of Vram(-3gb for the OS). Imo it's a bigger problem of poor optimization, the notion that we may hit a point soon where 12gb isn't enough is insane.
The "VRAM means nothing" crowd is btfo
I feel like I'm beating a dead horse WITH a second dead horse. But, recently picked up a 6800 xt couple of months. If you don't care about ray tracing, computational ai or nvidia ecosystem, going team red is a actual competitive option. I upgraded from GTX 1080 just to bust the fanboy-ism. Fantastic 4k powerhouse at an decent (ish) price.
To be fair a horse corpse on a crane makes a formidable corpse bludgeoning device... 😅😅😅
considering going to a 6700XT from 1070 Ti for the same reasons. what kind of games do you play? how’s the difference?
@@SplosionMovies I went from 1060 to 6700xt... the difference is massive. Granted, this is moving up a good few generations, but I really didn't expect so much power from my 6700xt.
@@SplosionMovies I went from a 1070 Ti to a 3070 and I wouldn't recommend it. The 8gb limitation is showing and DLSS is not doing much to help here as it can take even more VRAM in some cases.
The 6700 XT is a phenomenal value card right now. It'd be my choice if this card would die tomorrow.
Did same. Got it brand new for $380. My side of the world a total bargain. No regrets. Running all my favorites in 1440p Ultra. Loving it.
% matter if the new generation of cards are not an improvement or adding new technologys then they should be models that take last products technology and scaled it down to make it more affordable
Yes. And/or use less power.
EVGA smiles from above with an "I told you so" face.
Nvidia is essentially saying, if you need a gpu now i guess buy it but you should really spend more money for a better gpu or wait until the next gen/ competitor to come out with a product
Nvidia doesn't want you to have easy 16gb models because of AI model training. 16GB is a bit of a standard there for a fair amount of AI models.
@nulian Then make a 12gb card, the 4060 and 4060 ti base version should have 12gb. Running out of vram on 8gb isn't some future scenario down the line in 2 years. It is already a present day issue. And retail prices for their non sucky 12gb cards is laughably high.
A card with 16GB of ram can be used for some serious Machine Learning - people in Academia like me could avoid buying a 4090 for 24GB or paying 13000 for A100 (unless you want to train a state-of-the-art many billions parameter model)
So I got the impression they are suppressing increasing ram on these cards to avoid losing money on the professional cards for ML
If 4060ti 16gb is not about sending a message by nvidia then I dont know what is. Btw i just bought 7900xt and it is awesome.
I can't wait until 5000 series comes out with marginally lower prices while everyone consooms all over GPU makers for slightly improving an awful situation they themselves created.
prices aren't going down based on the price trend from the last 5 years
Basically, all of the current gen cards were already planned ahead for during the mining crisis and price bubble. Their margins were factored in, as were the quantities and capacities. They could easily turn these cards into great sellers by cutting down a bit on margins and upping their marketing, but if you're in their shoes, you know the current stock will sell out eventually, and you don't have to produce more if the market can't tolerate.. realistically, what would keep them from simply just thinking "this is a burned generation, lets not waste any effort trying to min max this, lets just cut the spendings and just let the market roll however it wants to. Nvidia is fully focused on AI as their main priority anyways, and most likely they don't want to shift any hands away from that to salvage a current gen underselling gaming card product lineup. They made insane bank over the last 3 years of scalping and mining, they can afford to not care. They know that everyone's opinion can be changed by simply launching a single good product that performs well and has a good price point. They know they are still firmly in power here, so whatever doesn't hurt their bottom line, is kind of a toss. Eventually the ship does have to turn though, so that's all we as gamers and pc enjoyers can wait for.
Seeing how this gen of cards has been debuting & marketed makes me happy I just snagged a 3060ti for 300$ Canadian, essentially the same performance as a 4060ti but half the price after tax's & or shipping
Yeah I was not sure my timing was good a couple months ago but I'm so glad I got a 3060ti then
3060ti for 300 seems high to me. Just sold evga 3080 for 450!
I could be mistaken, but I believe other tech tubers have discussed it's a "clamshell" design where they didn't get double density chips, but instead put half the chips on the back of the PCB, and there's some additional cost in how that works, but not enough to account for the price difference between the MSRP of the cards
You are correct. Linus was confused.
AMD are doing different versions of Starfield depending on the product you bought. I bought 7900XTX and got Starfield Premium Edition. If you say get a 7600x you get a standard version of Starfield.
I got a standard Starfield steam key for buying a system with the Ryzen 7 7800X3D in it.
Same, got mine with the 6800XT. It's what sealed the deal for me after a long wait. Seeing as the game is 99, i got the GPU for 429.
I got the premium version with a $300(usd) 6700xt. Heck of a deal
@@JakeTaco83 yeah it is. I liked that even the cheaper products still got the game.
@@MrRez808 its amazing. How are Nvidia too blind to get behind this stuff?
Wanna slug us with over priced hardware? At least sprinkle some incentive for us to claw back a little value. I wanna be wined and dined first before getting F***'d. AMD are on to something.
The only reason I got a 4080 is because it was $900 open box with the protective film still on it.
That's a great deal, I would have bought the same.
Same bro, I got one for 757 pre tax amazon warehouse! Brand spankin new 0 issues to this day. It's so massive I think the previous owner returned due to not fitting in his/her case.
Finally, not a bad price
@@mewoozy2 true an 80 series card at $900 would've been completely unacceptable just 6 years ago and look at where we are now
what planet are u on where $900 is a good price for a GPU? And it's even an -80, the mid range offering? jfc
Where I live 6700XT is 200€ cheaper and comes with the 100€ starfield bundle on top of that. Haven't seen something this DOA in a long while
Last year I saw a lot of creed from corporations. And a dystopian future, with control from mega corps, is closer than ever.
What this means is that the connectors need to be upgraded to support less signal degradation. Perhaps something like a liquid metal interface or some kind of "bonding" contacts you'd need to replace during a switch/upgrade?
Also, the backside of motherboards are great places to add extra traces and such. Stick a new memory design back there with much denser chips and you can cut trace lengths well in half. What if RAM had LGA sockets?
780 ti stopped board partners from doing custom vram amounts because nvidia wanted their flagships to sell.
The 4060 ti 16gb has clamshell memory so there are extra modules on the back of the card, this allows 2 memory modules per 32bit bus.
@elcactuar3354 **NGREEDIA ALL CAPS
@elcactuar3354 I wish the dislike button on comments worked.
Yeah the custom 780 6gbs were really nice.
luke is indeed most likely correct that there are agreements preventing anyone from changing things like ram and etc on the card before sale. however, what linus speaks of used to also be true in the past, and we did indeed have some take the cards and double the vram or etc. like the 2gb gtx460, or the 1792mb gtx275, and many others. particularly at the lower end you will see hugely varying amounts of vram. things like the 8800gt had a 256mb version, all the way up to a 1gb version, with things like 896mb versions in between and etc.
You can get a second hand 3090 (TI) for 700 dollars. 500 bucks for a 4060 (TI) is insane.
I'm planning to do this upgrade from a 750ti to a 3090 ti and it will be insane, the 40-series is just fucked up
I bought my 3090 last fall used for 650.
I think Hardware Unboxed highlighted that the 4060 Ti 16 gig card had some limitation in terms of where the DRAM chips could be placed, so they ended up placing them on the front and back of the board to use the same traces. Possibly one of the reasons for a 100 dollar premium, but it still needs to be a lot cheaper.
There's no way that actually costs $100 to put together.
They said the "clamshell" process is more expensive than a chip swap, which makes sense
@@Collin_J They are literally wasting sand and all of this dumb situation could be avoided if Nvidia at the very least made 4060 Ti with 192 bit bandwidth then GPU could have 12GB VRAM
@@PocketDrummerThe problem might be to develop and produce a different PCB for a product that probably not many will buy.
I personally would either buy a 4070 where everything is faster or a 4060 that's at least much cheaper.
@@Collin_J Clamshell is more expensive, but about $20-30 more expensive, not $100.
Looks like it's not a bad idea to wait out the 5000 series next year. I guess my GTX 970 is still working well enough. It runs Battlebit pretty good.
2025 for 5xxx series mate
@@4ksandknives that far away?
Then that one has other issues, time to wait for the 6000 series ;P
@@Ali-cya ha. I get what you’re saying. That’s been a truism in the computer-world since the very beginning. This series, though, feels like a rip-off. If Nvidia gets sufficiently punished by bad sales for this series, they’re motivated to do better for the next one. I can only speak for myself, but I was ready to pull the trigger on getting a 4070, but all the negativity from sources I trust (Linus, among others), has 100% caused me to hold off for another year. Sure, it’d be better than what I have now, but that’s worked for me for over half a decade now, I can wait another year for a good deal. I ain’t a sucker, and I’m among many others.
NVIDIA is in it to make money. They’ve been spoiled by crypto which created an environment where people would rip up any gpu anywhere for any amount of money. That ship has sailed and now their expectations for what they can get for turning out unexciting things is unrealistic to say the least. While normally they would adjust to the market NVIDIA thinks they will slide into the AI market make a killing there and not give a damn about the gaming market. I believe their next game offering will be awol a bit to see if ai kicks off if it does I think they will for all intents and purposes abandon the gaming market
AI is such a ridiculously unstable market for them to go all in on, because all it'd take is someone coming out with a dedicated AI chip on a platform that is either faster, more efficient, or even both.
Nvidia is really banking on their performance advantage, and once that dries up, it's done and they'll be eating Ramen if they can't convince people to forgive them for 3 years of BS.
Dead on, Jensen Huang has clearly stated the direction NVidia is going to follow, but the remaining Nvidia fanbois and various CZcamsrs don’t seem to understand English and are thinking that pressuring NV using a small scale boycott of the 40series will change NV’s pricing practices.
It won’t even make a dent, they don’t care, period.
Look into the AI market, past the front page hype and you’ll see AI investors are in for the long haul, yes this is a bubble at the moment but it’s not going to pop the way crypto did.
Currently NVidia can’t make H100 GPUs fast enough to keep up with demand and even if their clients like Google make their own AI accelerators they still have a massive learning curve to keep up with NVidia’s lead.
For the next 2-3 years NVidia is it, the only game in town for AI.
They have billions of dollars in unfulfilled orders backlogged and don’t need gamers.
The message is you’ll take what we give you and you’ll like it.
@@glenyoung1809 Most of Nvidia's profits don't even come from the gaming market as it stands now. Sure, they are eating some humble pie as far as gaming hardware goes, with a bunch of cards they can't move, but since it's not their primary revenue stream, they probably don't care at all. In fact, since it's a company that likes control so much, they probably think their 4000 series products are great and us pc gamers are a bunch of ingrates for not buying their latest (poorly performing) product line. To Nvidia, this is probably a signal to leave the gaming space behind. Which, sure, whatever. If they aren't interested in giving consumers fair prices for good products, fuck 'em. I am fine giving my money to another company that cares about the hobby and the customers. Jensen sucks, anyway.
@@jakedemo4590 I think NVidia is playing the waiting game, they've put out the cards they're gonna put out, there won't be a 4050 because that's so low end the margins wouldn't be worth the bother.
NVidia probably thinks that they'll wait till the 30series stock is gone and see what happens, they're probably banking on NVidia fanbois not having much stamina to out wait them.
If the lower end 40 series begins to move they would be proven right with their pricing scheme, if those cards are sold but lose money(for the AIBs) NVidia won't care, it will tell them they're better off leaving the low-mid tier gaming market to AMD/Intel.
In both cases NVidia "wins" because they got billions of AI bucks flowing in, plus there is the added factor that their big shareholders don't care about the gaming market, they want results from AI sales, not gamer sales.
I've seen interviews with the Wall Street investor types who think gaming is an overrated sector and not worth investing in because the product development times are long and the profit margins are small.
AI will affect all of society, gamers in comparison don't count.
re: 4060 tu 16g memory - I thought I saw that they replaced the ram packages with clamshell packages (both sides of the pcb) and the controllers have to swap between ram banks (cannot access both sides at the same time)...don't recall where and may be mis-recalling.
Would it be feasible for a sort of hybrid approach, sort of like how we use cache? Have on-board memory for more important tasks (like for use by your p-cores), and then have SODIMM memory that you can expand for lower priority tasks? Since, isn’t the 64-bit address space **WAYY** wider than any system will feasible be using for a while, so just, take an empty chunk, and map it to the expandable memory? Sounds like the kind of architecture you’d see on a weird console.
I actually planned for a long time to finally upgrade my trusty 1070 to a 4060ti 16GB. And then they released this piece of crap that's literally a downgrade in some regards... Feels bad man.
The 3070ti are finally at a reasonable price here in Canada. I've seen some cards go for as "cheap" as $585, which is $35 more than my 1070 5 years ago. :D
@@marc-andrerenaud1394Yeah but 8 gb vram and no dlss3 really sucks. Either get annex 6800xt for more vram or a 4070 for dlss3, frame Generation und if you think 12 gb vram is enough
Buying nvidia is like on purpose step into a tourist trap.
hey at least im getting high frame rates with ray tracing.
@@KamekoTsurugaray tracing is always implemented as an afterthought and is completely a gimmick.
@@Michael-ft3vg that's your opinion, thanks for sharing.
The only reason to get Nvidia nowadays is if you are building/upgrading a workstation PC, or if you're interested in ML or AI.
I have been an Nvidia loyalist for years and unfortunately, even today, the Nvidia cards reign supreme in performance in blender, Maya, Davinci, and other software. My 1660s died and I got a 4070 to replace it. It's an amazing workstation PC, but I had to replace my CPU to see that performance increase.
1. at 12:30 those were DIPs (dual inline package). Your presumption that signal degradation is hardly the only issue. There's also cross talk from parallel trace, EMF/EMI interference depending on which lines are being signaled, and at frequencies nearing the low end of the microwave spectrum the phenolic plastic formulation of the socket strips themselves can become a liability. This is all due to the relationship between stray capacitance, inductance and resistance. Developing these systems requires high-end spectrum and Fourier analysis to eek out the bugs and try to keep the signals from interfering, turning tight square waves in to muddy harmonics. This is difficult enough on 32 bit busses running a few hundred megahertz (back in the day). Imagine trying to make modular memory for GPUs with 256 bit+ busses running at the low end of the microwave spectrum. For all the people who have successfully modded boards by replacing the memory BGAs with double density DIEs, kudos to them, but one wonders how many attempted the same feat and bricked their boards. PS. I've worked with hard state industrial electronics for over twenty years. I know what I'm talking about.
2. As to the question "what's the point of the 16GB 4060Ti?" I think the answer should be pretty obvious. Everybody wants to be a content creator these days, or use AI accelerated computation for, well, you name it. AI accelerated image generation, AI accelerated logarithms, AI accelerated video processing..not everyone can afford 3090s or 4090s, or their lower sku counterparts with 12-16GB+ VRAM, and the VRAM wall is the Achilles heel of AI acceleration. Even with a lousy 128bit memory bus, that extra eight gigs gives beginning content creators the little extra "oomph" to not wait X times as long for an AI workload to complete. You see a ridiculous gaming card, I see something intended to be in the reach of would be starter content creators. Your commentary smacks of elitism, I suppose you were born fresh with all knowledge and requisite tools fore-to-hand? And again, I speak as someone who is a (small time) content creator and just so happened to have the patience, curiosity, will to push to teach my self damn near everything I know with next to no help from ANYONE except the patience of research; and inculcated with the financial discipline to be able to save for all the big boy toys and learn how to build the machines I desired to my tastes with no help from elitists.
3. 8:15. This is just asinine. This shows a complete incomprehension of the complexities involved in establishing stable BUS clocking across multiple parallel traces at high frequencies. This has ZERO to do with nVidia proprietary requirements for any stable function whatsoever. Go read an engineering application datasheet for a lousy "Atmel 8051 Microcontroller" It's an 8 bit industrial microprocessor. THE ENGINEERING WHITE SHEET IS OVER A HUNDRED PAGES LONG. I wonder how many pages a modern GPU engineering application protocol white sheet has? But of course, these are all unpopular truths from a proud nobody, so no one will read any of this. It might accidentally make some one think a little too hard and not in step with their usual fanboy unconscious nodding.
The memory chips are the same density. They already use double density for 8GB, so they had to use Micron's clamshell config to achieve 16GB on 128-bit. The other half is on the other side of the board to couple both chips to the same lane. Nvidia claims that's expensive to do, of course.
Given the AI craze, I think Nvidia is focussing on AI chips over gaming cards, and will sacrifice its gaming arm if it comes to it.
AI and gaming will merge and then all will be perfect in the world.
I can understand if it was a 50 dollar increase, but 100 bucks bro, that's almost a whole tier difference in price... Like they're insane
I remember how back in the day I would buy Radeon 9550 that has more memory than it was supposed to, with twice the bits it was supposed to have and and faster ram dice at that (less ns latency) and with a simple super basic V-mod (just using a pencil!), a custom bios and unofficial drivers you could easily overclock it and unlock it with drivers so for the price just slightly higher than a reference Radeon 9550 I was getting a Radeon 9600XT. Similar thing could be done with some 9500 to make them into 9700 or with 9800SE to make it a 9800XT and so on, if you bought the right version from the right vendor. I miss these days it was so much fun just being on this mission to get the best bang for buck using all these tricks with manufacturers making their products upgradable for enthusiasts like that on purpose so those who know would pay a little extra to buy their version instead of a reference one.
Nvidia just looked at the iPhone's storage and thought "Hey, we can do that too."
Unfortunately there is no easy upgrade here. There are only four memory chips on the 4060Ti. The 16GB version uses clamshell chips on the back like the 3090 did to get that capacity. That’s more expensive than just higher density chips because 4GB VRAM modules do not yet exist.
You can get 4GB in a single package, though Samsung's GDDR6X memory does have a double-wide interface. Not sure if those allow wiring in a clamshell-equivalent manner. Even if they do, you would still need a custom PCB to accommodate them.
@@teardowndan5364 No, you can’t. GDDR6X is not produced in that density. They have to use Micron’s clamshell configuration with a custom PCB that allows two memory modules per each 32bit controller. Very pricey workaround made worse by the initial cost cutting. If the bus was 256bit with eight memory modules to begin with, swapping to 2GB modules would have been cheap and easy.
@@Aurummorituri GDDR6W can have up to four dies per package and is available in 32Gb format. That was one of its key selling points.
@@teardowndan5364 And? Is anyone actively using that in their products outside of data centers? Would any board partner plop bleeding edge memory on a -60ti card? No. Nvidia painted themselves into a corner by calling a 4070 with a -70 class die a 4080 and went down the line from there. Everything below the 4090 is extremely cut down and woefully overpriced. Their own chintzy designs caused this mess.
@@Aurummorituri GDDR6W isn't anything special, it is just up to four GDDR6 dies in one package with a proprietary pinout.
The PCB is different between the 4060 Ti 8 gb and 4060 Ti 16 gb if you follow the teardown. But it can’t be too expensive.
Probably so you couldn't just solder on more RAM
It means developing different parts for a product that probably won't sell well and doesn't add to prestige like e.g. if the 4090 was available with 16 or 32.
@@FM4AMGV oh you know every card produced from now on will have the extra RAM pads either removed, or traces broken to prevent anyone from making the card better without paying the Nvidia greed tax
@@yourfavoritelawnguy2722 they're not "removing" anything.
@@FM4AMGV That's a backwards way to put it. The 16GB version has extra solder pads on the back for the extra GDDR6 chips. The 8GB version doesn't need these pads so there's no reason for Nvidia to spend more to include them.
The 16GB version was an afterthought in response to complaints about VRAM capacity, most likely based on a PCB intended for workstation GPUs.
this is a place i think something like Optane could actually be used to huge benefit, local fast cache, thats not up to ram speeds, but is on card, and capable of giving the card a local cache it can directly access rather then resorting to system memory if vram limits are reached.. its not ideal but.. i can see huge benefits for alot of workloads and.. even games...if the games resources can be stored on card entirely... that would make a huge dif to perf and load times if done correctly..
Nvidia should move to a tick/tock pattern with their GPUs, where one year they release a new gen of desktop GPUs, then the next year, they refine the design and reduce power consumption and release only mobile GPUs. That is basically what happened with the 30 series anyway. The 40 series are not bad designs for mobile. Lower power consumption means laptops can comfortably keep them cool at max TGP. Raw rasterization is more than enough for most laptops which are capped at 1080p or 1440p (and will almost always be used that way). But you’re overpaying for every single card on the desktop, and really for no benefit unless you’re buying a 4080 or 4090.
Nvidia has to keep those GPU shortage era stock prices going somehow. Screwing their customers is a great way to do that.
that’s why they slow drip them from the warehouse to the distributors
They are the De Beers of hardware because a GPU is forever
When I bought my RTX 2080. I also received a code for CoD MW 2019, back in 2019. EDIT: It was about 700$ at the time.
I remember Dell prices for server RAM and storage when I was sizing up infrastructure. For DDR4 ECC, they were selling the exact same part number Crucial ram sticks for 4x the price. Needless to say I've always bought my Ram directly from the Crucial website. Storage was a complete rip off too.
The premiums on back then still existing fast spinning rust (10k-15k rpm SAS) was outrageous and those were much harder to procure independently. Nevermind the 'entreprise class' SSDs where they would be as vague as possible as to what exact brand/model it was they were selling so you couldn't go buy them on your own while still being somewhat sure there won't be any firmware issue or lock.
Wasn't so long ago 500 was top end 80 class it's outrageous how much it's inflated
SDXL come out this week and the AI model checkpoint is 13GB and it struggle on 12GB/8GB GPU. The people using stable diffusion will probably be enough to buy all of these GPU.
Here's hoping that model quantization comes to the rescue
Pruned models are much more manageable for practically no downside
If they cut the price by 100 I'll be interested
Put it differently: The 4060 Ti should have more RAM than 8GB. E.g. 10 or 12 which also means a wider memory bus that is faster than the current one.
Bigger question is artificial limitation of amount of supported VRAM on NVIDIA silicon? Only to protect far more lucrative AI business niche with A100 and H100.
When a 3060 beats the 4060 in some games makes me think the 4060 is really a 4050.
I think Nvidia is willing to take a loss this entire generation (which can be offset by gains in AI) in order to shift the market expectation for the following generations. Unfortunately, AMD seems in on this plan.
We need Intel to stay competitive, last hope.
Do you not remember how much a 2080 TI was? Like $500 for a graphics card was unheard of at this performance level four years ago.
2080 super was only about 10-15% slower for $700
My Microcenter isn't even selling the 16gb 4060ti
i would have to wonder the feasibility long the 8, 12, 16, 20 and 24gb platform speeding up ram to make up for the lack of ram is going to start causing operational issues if not today sometime in the future...
They don't care about consumer cards any more. They are now a business facing company selling AI professional equipment. The margins are way higher. I don't understand what is so hard to get about that.
20 years of marketing to gamers in the PC space, many gamers came to believe that their loyalty to the NVidia brand actually meant something to Nvidia, it didn't.
I don't think many gamers understand the Wall Street investor mentality and how NVidia has a fiduciary duty to maximize profits for shareholders.
AI specific cards have probably 5-10x the profit margins that consumer cards bring on a per mm2 of wafer used basis.
The amount of money poured into NVidia's stock is both a vote of confidence and a demand at the same time that NVidia needs to do better.
The gamer space is basically a saturated market, the margins are low and to be profitable requires large volumes of unit sales.
Plus AMD and now Intel are competing heavily in that space.
Whereas in the AI sector at the moment NVidia stands alone, there are no other products close to what NVidia has and available in the quantities that NVidia can make available.
But gamers are a stubborn and somewhat entitled bunch who demand constant performance upgrades while demanding prices don't rise, while the cost per wafer for the new process nodes continues to soar almost exponentially.
I'm not a fan of NVidia's business practices and attitude but I can see why they're doing what they're doing.
7:25 As a board manufacturer, you're only allowed to build cards to spec. Spec has set standards (pcie lanes, memory capacity, etc.), while others are minimums (eg. power delivery). So unfortunately, they are not allowed to interfere with the Nvidia Spec, which means no extra traces that are attached to the chip. Which is why that 1 company did those 4060s with the m.2. slots, which are NOT attached to the gpu die in any way.
The 3060 Ti is already somewhat hindered in performance with its 8GB VRAM. So while a 16GB card in the 60 class makes sense for the sake of longevity, considering most people buying a 60 series card will try to make good use of it, the price also has to be right. And indeed board partners could add memory. it's been done before. You can get a Radeon Rx 580 with both 8 and 16GB VRAM. So it should be very much possible. So indeed, why don't board partners just take the 8GB version of the card and add the other 8GB? It seems like a good strategy because you can definitely gain some performance.
I’ve got a 3060ti FE 8GB and it’s brilliant but I know I’ll need to upgrade for more vram soon and have no idea what’s worth getting - it certainly isn’t a 4060ti
16GB isn't going to do you much good without the VRAM bandwidth to actually put it to use. If assets get 50% bigger, you need ~50% more VRAM bandwidth to put them on screen and Nvidia likely squeezed almost every MBps out of the 4060Ti that it could.
Wonder what Nvidia would do if the board partners put 12 GB of vram on the card and charged 429.99 instead what are they going to do .
Laugh, because you can't put 12GB of vram on a 4060 Ti.
Nvidia would stop selling them the chips.
You can't do that, if it has 8 you can only go to 16 with Dram that is doubled the density. Ram is CHEAPER than the difference they are asking for; they are just overcharging for the ram difference.
This is the way i see it Nvidia can't call all the shots anymore and they are worried because Intel has now entered into this also you see all board partners have to do now is say sorry Charlie we are not making cards for you anymore that would catch Nvidia's attention and if the rest of the partners did the same they would see they have the power to control this.Anyway Intel in this changes everything and Nvidia hates that i say take the fight to Nvidia.
@@stevenanderson3205 Nvidia doesn't actually care that much. They will just drop their board partners if they complain most likely.
It's entirely possible they flat pull out of the mid/low range GPU market all together and only push out high end cards and ship the rest of their cores to AI compute GPUs and their professional cards that sell for 10x the cost of gaming GPU's.
They have the lead in AI by miles and that's all they actually need to make money hand over fist. We gamers are not needed any longer.
So how long before we start seeing motherboard manufactures introducing soldered, unchangeable memory modules directly integrated on desktop PC boards to achieve higher performance?
Mobile devices used to be PGA back in the socket G2 days... Wanna say up until 3rd gen (ivybridge). Currently have a machine running a mini ITX board (Jetway QM77) with a i7 3770 (Sr0ux).
Are you sure you tested this correctly?
I went from a 3080 ti to a 4090 and I can say there is a nice difference between the two but one hindsight I should have saved the money and just waited for a real leap. That being said building a entry level pc for a friend worked out buying used 30 series cards because the lower level 40 series cards are just not worth the prices nvidia is asking for. The whole pc market is sad right now and it seems like games are getting worse year after year. Another dark age is on the horizon 😢
I honestly think the only bad thing with the pc market right now is gpu prices, everything else seems to be value oriented
Man that pessimistic is honestly sad
The PC market is in a great state, if you don't fall for the marketing AMD and Nvidia shove that is "YOU NEED THE NEWEST AND GREATEST" AMDs 6000 series cards offer unprecedented value right now on the new market
And for the state of gaming being bad? Really?
Tears of the kingdom is probably the best game ever made imo
Last year seen Elden ring which is probably the most important game to come out in the past decade
Armor core 6 has gotten gameplay reveled and it looks breathtakingly good
Starfield (while I'm not getting my hopes to high because, well, Bethesda) does actually look promising
Sure gaming has trash, but it's never nir had trash, the only difference now is thanks to the rise of social media that trash gets attention, before if a game was bad it faded into obscurity, now bad games (like the lord of the rings gollem) get a bunch of media attention despite not deserving any, you know the old saying by now "any attention is good attention"
@@flamingscar5263 he said pc market not gaming in general
@elcactuar3354 **NGREEDIA
@elcactuar3354 don't you have anything better to do than correcting people in the comments? how sad
EVGA saw the writing on the wall and noped out at the perfect time
a lot of board partners used to make their own special edition cards with additional Vram, but i think nvidia put a stop to that years ago
Honestly I think the 4070 is a fine card... I came from a 1050ti and it is a giant upgrade. It can run most games like Battlefield (1080p monitor) on 150-200% renderscaling ultra settings over144fps (without dlss)... what is there to talk bad about? 12gb Vram is plenty last time I checked as can be seen by KreativEcke's testing.
I sure think it is not one of the "great value" items, but I do believe it is also not bad value compared to rtx 20/30 series... and that was everything between the 1080ti craze and today which is basicly one upgradecycle.
Just wanted to point out, GDDR6 chips are only made up to 2 gb each. So on a 128 bit bus (32 bit per package) on the 4060 ti 16gb, Nvidia did a process of soldering a chip to the front, then one in the exact same position on the back. This is called “clam-shelling”.
Yes, it was more expensive to manufacture it like this (more than just the $25 for the ram itself). However, Nvidia used this complicated process an excuse for the extra $100. It’s just scummy, it prolly would’ve been cheaper to just solder them all to the front and expand the memory bus to 256 bit. But Nvidia couldn’t do that because this 4060 ti would then have a wider bus than the 4070 at 192 bit. 🤦🏼♂️ they suck
Expanding the memory bus means adding memory controllers which have to be placed on the perimeter of the chip. Chip area is proportional to the square of the perimeter, and the controller circuits take up a disproportionately larger area because of the higher variability of cutting edge process analog parameters. In other words, it would be anything but cheap.
"Clamshell" refers to the side-by side arrangement like an open clamshell, the chips for one channel are on opposite sides of the PCB. The chips can be placed physically opposite but that requires even more layers
I think part of it to is that Nvidia might have to redesign their GPU cores memory controller to really address the issue. If they want they can double the memory controller width and when they cut down the GPUs to make smaller ones they can have 60 class cards with wider memory bus however this probably hurts yields and costs more as well as makes the core bigger. It would make it easier for Nvidia to just slap 16GB on a mid range card but it might increase cost and complexity. I'm sure Nvidia has ran the numbers and found that the way they are doing is not is the most ideal for what nvidia wants and that's likely maximum profits/ROI.
According to MSI website their 4060 video cards have a maximum memory throughput of 17Gbps which is 2gbps faster than the 3060 which is 15gps. But 2Gbps slower than the 3060Ti at 19gbps. Not sure if extra 2gbps make a difference in performance. MSI also lists the 16GB AMD 6800 gpu at 16gbps, which makes it 1gbps slower than 4060 but has double the memory.
I think the 4060 and 4060 Ti will not have higher memory. It’s the same reason why Apple sells base model non pro Macs with only 8GB of RAM. It’s so you buy the higher tier model.
They don't care - nVidia's agenda has changed to AI.
Wait doesn’t the 3060 have 12 gigs of ram? Why would the 4060 of 8?
Because the 4060 is a 128 bit bus card and the 3060 is a 192 bit bus card.
Ram chips come in 1GB 2GB and 4GB flavors. If you have 4 channels (128 bit bus) you can only have 4, 8, or 16GB.
If you have 6 channels you can only have 6, 12, or 24 GB.
insert mr krabs *money* meme
very informative comment@@JathraDH
Signal degradation due to pins or LGA or any other should resonably be overcome by increasing voltage slightly and including some voltage control on the the slottable gRAM unit to limit the final voltage digit the chips recieve.
You can't really increase the voltage though. Also, extra voltage won't overcome the inductance of sockets.
@@saucebosspl Ah, thanks for clearing that up, I assumed by increasing or making up for signal integrity by increasing voltage as and when needed, but I didn't think of the pads/pins themselves as a part of the issue.
AI is driving the specs for my next card purchase. I want at least 16gb with a 256-bit bus, the only way I can get that with NVIDIA is to shell out ~$1k. AMD and Intel both offer cards with those specs at $450 and $340 respectively, but I'm not confident in trying to deviate from NVIDIA, as a lot of the newer AI things tend to require NVIDIA or need to jump through hoops to get it working on other cards. Eventually the technologies mature and work on other cards, but it's been clear that running NVIDIA saves me a lot of trouble. I'm not sure I want to settle for 16gb 128-bit bus on an 4060 ti at $500 and I definitely don't want to jump all the way up to $1k. I guess for now I'll just hold out and see what happens.
just get a 7900xt
Still not that good a 6950xt is $600 and like 10% slower which is not noticeable
I would absolutely be willing to pay for extra vram on a gpu. 3070 with 20gb vram? Count me in! 4070 with 20+bg vram? No need for a 4090 anymore.
Never thought I would see a 107 die XX50 series card with 16gb of ram! What a time to be alive!
4060ti = ad106
@@basshead. hell you are right. so the 4060ti is a true 60 class card. The 4060 though is a 107 which should have been the 4050.
Part of me is glad I got the rtx 2080 super in late 2019.... but man The ram crunch is bothering me. I came from the gtx 970, and before that the 760, but I guess 4 years average is what I've gotten before, but man it feels shorter now
Here's hoping the 50 series will be good...
You're funny
😂
Every passing day I'm glad i got a 6700xt - we desperately need to increase intel and amd's market share of gpu's because if they don't start taking from nvidias insane profit margins, then Nvidia will just keep spewing out hot shit that could've come out years ago (and they sell it for the price of something new and innovative)
i was going to grab a 4080 for my future build, i can afford it but watching this vid should i still go for it? i want new modern specs to last me for awhile. im also impatient lol i dont want to wait
What I hate most about 3000 and 4000 series is that lower end cards can have potentially more vram than their higher end counterparts and this would certainly affect longevity of those cards. There should not be 12Gb 4070ti while a potential 16Gb 4060ti exists or a 12Gb 3060 while there is 10Gb 3080. Higher end skus should have had way more vram.
Sup notification gang
eyyy
I feel like these corporations are so far out of touch these days. I don't know the logistics behind a campaign like this, but they could offer monthly allotment of in game currency for certain partner companies. For example, $10-20 monthly in ea monies that is only usable for in game items. That would incentivize someone to play that companies games, it would encourage players to actually open the in game store (which we do not do) and it would lead to more people buying into micro transactions. Yet instead they would rather sit on piles of cards waiting for the next AI/crypto boom.
Anyone remember the GTX 960? It launched with 2GB of RAM. For some models, all the manufacturers swaped the RAM modules for double the density and a small price increase to create a 4GB model. Many people said it would be find with just 2GB, that there is no need for the extra money, that the core will become the bottle neck before the RAM capacity... I got the 4GB version and I am super happy I did. The little monster lasted me a lot longer than the 2GB would have for gaming, but also some 3D modeling and even some AI work. I upgraded it to a second hand 1080Ti for a solid price, right before the crazy price jumps a few years ago. Prices of those have recently dropped quite a bit as well, at least where I live.
Still rocking a 1060 gtx 6gig. Not looking for a upgrade any time soon. Just going to keep using Lossless Scaling off steam for my up scaling.
Since I keep seeing this commercial. Is the upcoming 5060 good enough to play Hero Wars?
Whats funny is that nvidia made a clamshell design with the 4060 ti meaning they have double the amount of ram chips but they only run at 16 bit instead of 32 bit to fit in the 128 bit memory bus, so it doesnt have any advantage but its more complicated to manifacture and they can sell it for more.
I know hardly anything about computers but I’ve been saving and getting parts to build my first pc, and I’m stuck on if I should buy a 3080 or a 4070/4070ti
So i was going to buy 4060 ti 16 gb and the store wants $700 and some change
so now i am really not sure what to get what is the best graphics card that wont
break the bank & get the most Vram and is good and be honest.