Architecture All Access: Live at Lunar Lake ITT: Next Gen P-core Lion Cove
Vložit
- čas přidán 9. 09. 2024
- Welcome to Season 4 of Architecture All Access!
Architecture All Access is a master class technology series, featuring Senior Intel Technical Leaders taking an educational approach in explaining the historical impact and future innovations in their technical domains.
For Season 4, we are diversifying the coverage of Architecture All Access by focusing on three separate content tracks:
1) Architecture All Access: IP Insights, where we will provide architectural animations to explain our main IPs
2) Architecture All Access: Live Sessions, where we provide presentations from our Lunar Lake Intel Technology Tour
2) Architecture All Access: Master classes. This is our usual educational content that helps connect architectures with the usage models they benefit in an easy to understand manner.
In this episode Ori Lempel, Senior Principal Engineer in the P-core architecture team, provides an in-depth look at the newest P-core microarchitecture codename Lion Cove and he covers the key goals of the microarchitecture. The episode also unpacks the changes in both front and back-end of the CPU Core, optimizations for ST performance and area efficiency, as well as the generational improvement in IPC (Instructions Per Cycle) and PnP (Power and Performance) of the microarchitecture.
Missed seasons 1,2,3? Over the past 3 seasons, we have covered topics like CPU and FPGA architectures, Quantum and Neuromorphic computing, Transistor technology, Artificial Intelligence, Meteor Lake architecture and much more. Check them out here:
Season 1 Playlist: • Architecture All Acces...
Season 2 Playlist: • Architecture All Acces...
Season 3 Playlist: • Architecture All Acces...
Subscribe now to Intel Technology on CZcams: intel.ly/3P9BA7x
Follow Intel Technology on TWITTER: / inteltech
Intel been cooking for this one, looking forward to it
Really looking forward to laptops powerd by Lunar Lake! Good job Intel.
Lunar Lake FTW
Interesting stuff. I've been wondering about the need of hyperthreading already now when the core counts are getting so high - but here it was stated that the P/E core structure makes it not that useful.
Ditching HT should net them some gains in power efficiency and Single core performance
@@BonusCrook it clearly doesn't though, even in their own slides.
@@vyor8837 disabling HT yields more single threaded performance on all current intel atchitectures
@@BonusCrook Because intel is incompetent, yes. You don't see such on AMD.
@@vyor8837 lol ok dude... You know intel and AMD have different architectures... Right? Their chips benefit from different things.
Nice, I'm looking forward to seeing the differences in Lion Cove between LNL and ARL
Looking forward to AMD
Lion Cove is gonna be lit in MTL-P and ARL-S!!!
😃Awesome Presentation 😃. It will be actually interesting to see the level of enhancements that the Lion Cove based P-Cores bring with the Lunar Lake and the Arrow Lake iterations. Looking forward for a really heated up race for the AI PC's/Laptops
Requesting updates for 12th Gen - 14th Gen Thread Director to improve performance gains on threadless workloads, similar to what's expected from the upcoming Lion Cove cores processor.
Wow, I thought meteor lake had a lot of advancements. This is next level. Great to see efficiency improvements!
MTL's core was more or less the same as their current desktop ones.
@vyor8837 The advances of MTL aren't all in the cores. It moves to a tiled design, introduces the low power island concept, introduces ARC iGPU architecture, inplements a new E-first Thread Director, and gets designs rolling for both in-house and external fabs in the same product.
There's a lot in MTL that isn't a CPU core, and all of those are still solid advances over RPL-P. MTL makes changes to the SoC and keeps a relatively safe uArch change as time was clearly spent elsewhere making changes that will benefit future atchitectures.
@@DigitalJedi and none of those will make it to the desktop.
@vyor8837 Incorrect. Arrow Lake-S has dual-sourced chips (20A & N3), arc graphics (Xe LGP+), a tiled architecture, and a yet unconfirmed thread director that will trace its heritage up from RPL, through MTL, and into the SoC tile.
Low power island is the only one unlikely to come to desktop because there's no need for it. In its place, the E-cores are put right on the ring bus next to the P-cores and L3 cache. A suitable substitution when performance takes priority over single-digit TDPs.
And even if none of those innovations came to desktop, so what? Like desktop is the only segment that matters? The laptop market is massive, and even if MTL only spawned a new wave of chip design there, it would still be responsible for test driving changes that would improve the quality of 10s of millions of devices in each generation to follow.
I don't want to come across as a fanboy. I worked on the process that stacked those tiles. I've seen how the sausage is made, and even if the performance is unimpressive, the tech supporting it is undeniably here to stay.
@@DigitalJedi desktop and server are indeed the only ones that actually matter because intel already has a near monopoly on laptops. They could, and have, release a new laptop gen that is worse in every way than its predecessor and only lose maybe 10-20% of total sale volume.
It's neat tech, but also not exactly new overall. Like... AMD kinda already did parts of it with zen4.
I have seen Intel reduce the price of its products. I want to heavily discount products that cannot be overclocked and overcloclable cpu don't have to discount, because users often do not want to buy a chip that is too many limited, even overclocking like Intel. AMD all cpu don't have any limit to overclock
Must have a nula nake overclokable, or same cpu with powerful igpu overclockable like luna lek. I will buy it to overclock it immediate. ALL IS NEED OVERCLOCKABLE LAPTOP CPU WITH MOST POWERFUL IGPU LIKE ARC OR IRIS ON LAPTOP FOR GAMING LAPTOP NOT POWER EFFICENT LAPTOP
software people need to rethink all the important algorithms to not just build around multi-threading, but also unequal cores.
Lunar Lake is impressive. The time is now, Intel, strike!
Gonna be buying a new laptop soon, I hope lunar lake will handle some light to heavy gaming with the xe2 architecture,
Thanks for the 32 gb ram as well tho!
looked at P-core slide and i so believing that Intel P-core paradigm is changing from P6 into E-core design
just like centrino era when P6-design called as E-core /small-core while Netburst is called as P-core / big-core and years later, Core 2 come with P6 design that used until 2024 and since that, P6 is a base of Intel Big core (until 2024) and Netburst abandoned.
and now, Intel designing new P-core with the E-core design (one of the notably design is splitted scheduler between INT and VECT)
Arroe Lake When
1080p60fps video nice intel 👍🏻
🥰🥰🥰🥰
Intel is back! Buy the dip
Mobile computing looks like its about to change!
All I want is better IPC and power efficiency
Do you have a counter to AMD’s 3d vcache?
What are the micro architectural roadblocks ?
And what are the differences between the LNL version of lion cove and the ARL version ? Very impressive on the process agnostic design……
Make integrated GPUs good...otherwise no buy....for the 12th years consecutively
Meteor Lakes iGPU already surpasses AMD's iGPU in quite a few games, they just need to continue optimizing the Arc drivers and then Intel Arc iGPU will be ahead in most games.
@@__aceofspades that's a lie, msi claw sucks big time vs Rog or Legion products. Core Ultra was just hype, various updates couldn't fix the lag and stutters, MSI panicked and now releasing lunar lake version.
@@StartreckTuClaw is a really bad example of MTL. It fares much better in a traditional laptop setting, where it is competitive with AMD's phoenix APUs. The 780M is usually a little ahead still, but the race is much close than it was with Rembrandt vs RPL-P.
now we need better discrete and AI class GPUs... fight back intel.
I know I am old school but when you do a presentation for the public please try and dress for the occasion instead of looking like you just woke up. Especially when you are behind other companies in technology.
What drugs are you on?
Come on Intel! Please deliver at least 20% IPC increase from Raptor Cove. And please much lower power consumption than Raptor Lake. And please a 12 P core only Arrow Lake Core Ultra for world class gaming and no hybrid scheduling gimmicks for ultimate best set and forget gaming CPU for todays games and all games past 15 years, And super great core to core latency being on same tile. I so badly want a homogenous CPU with 10-12 Big cores on a single tile/ring bus/CCX-CCD/die. Come on Intel deliver it.
You don't need more than 8P cores. No game is gonna benefit from that. And for workloads that do actually properly utilize more than 8 cores, E cores are better.
@@lycanthoss Well Cyberpunk and Last of Us Part 1 and Spiderman Remastered have seen reports of them in some situations using more than 8 P cores. More P cores better for gaming as no scheduling quirks and easy set and forget solution. No Star Citizen or Elden Ring issues. A 12 P core gaming only chip and then the 8+16 regular ones Intel should release both.
@Wolverine607 just because they utilize more than 8 cores does not mean they benefit from them. In basically every game, when you go from 8 core to 16 core CPUs like from 5800X to 5950X, the performance gains are mostly from higher clocks and cache. Games (and most apps honestly) mostly care about single thread performance.
Just look at previous CPUs. The 7600X easily beats the 5800X and X3D (in some games the X3D is better, but those games are more cache dependent), 5900X and 5950X. And if you think the CCD part matters, look at the 12400 vs 10900K.
@@lycanthoss Well I am sure CCD part matters dual CCD/CCX cross latency penalty 12400 vs 10900K. Well 12400 has far better IPC on Golden Cove than 10900K on Comet Lake. We have no hypothetical 10 P core Golden Cove/Raptor Cove which would be the best.
And Xeon please. Xeons have worse IPC than client Golden /Raptor Cove. Maybe if Intel HEDT was like past iterations without overloaded glamorous enterprise features detrimental to gaming, than also maybe. But no ring bus or client more core equivalent after Broadwell-E. Instead the horrible mesh latency snice Skylake X lake and yes that includes Sapphire Rapids but worse.
@@lycanthoss Star Citizen hello? Its just a matter of time before the new games uses 10-12 cores.
Please stop throwing more electricity at your problems, and give us performant & high energy-efficient CPUs, thanks.
That is literally the entire point of Lunar Lake. IPC gains from 14-68% depending on the core and int/fp. Power at iso performance is down and performance at iso power is up.
Decade with weak chip, ditch by apple and then microsoft let's see if intel survive.
I can't forget the situation when HT was first released more than 20 years ago, it brought an 80% performance gain in certain supported software. Of course, this refers to the performance gain that can only be brought about when the application software itself is cache-friendly. Intel is giving up its own advantages. Intel gave up its own advantages but was picked up by AMD. Intel's advantages are likely to get worse. AVX512's success on the Zen4 architecture is an example.
You couldnt be more wrong. HT used to be extremely important when core counts were tiny, back when core counts were 1-4 cores. Now with Intel having 24 core consumer CPUs, it no longer is relevant, there are more upsides to removing it then keeping it.
As for AVX512, you have to be joking. For consumer use it never gained traction, you can count the number of applications that use it on one hand. When Intel discontinued support, it became effectively obsolete, as nobody is going to develop consumer applications for AMD's fraction of marketshare where only Zen 4 and beyond support it. Also Intel has already created a better successor, AVX10 which is already rolling out in servers and should hit client next year.