Discussion Intel current and future Lakes & Rapids thread

[AlltheWay LeeWhy]

Is it? 11th gen was extremely poor.

RKL was extremely poor, unless you grabbed an ROG Z590 Apex motherboard and a small number of very QVL specific high-speed mem kits.

Haters gonna hate, I guess.

5066 CL17, 5333 CL19, 5600 CL20, 5866 CL21 all possible with ambient air.

 

[AlltheWay LeeWhy]

Hint, look at their signature. Intel only and already going Raptor Lake. Intel could sell poo and some people would buy it. Like Rocket Lake.

I've ran AMD graphics cards for the last 10years. Only recently moved to an Nvidia workstation card because the AMD update from my AMD WX-4100 only uses 4lanes and 4GB of memory. And honestly, it's kinda ugly.



Whereas NVidia is offering an SFF workstation card with 12GB of memory and uses all 16PCIe lanes, also the Nvid card looks more professional to my eyes.



I was also kinda getting a bit tired of the AMD "way too blue" aesthetic after 5.5years of use, although it's an amazing workstation card, but Polaris architecture? Kinda long in the tooth.

Used AMD graphics for 10years and AMD workstation graphics for 5years.

Before and After pics:

 

[AlltheWay LeeWhy]

Should Intel skip 13 and call Raptor Lake 14th gen for good/better luck? My residential building does not have a 13th floor. At least, not that I know of. Haven't really taken the stairs to investigate.

But then the 14th floor is really the 13th floor. lolol

And what's really creepy is - this is the top of thread page 666.

 

[igor_kavinski]

And what's really creepy is - this is the top of thread page 666.

Ominous indeed. Much clouded the future of Intel Lakes is. Recommend professional exorcism.

 

[Thunder 57]

Perhaps I am wrong. It's happened. It will happen again. I can tell you that probably no one likes you taking up all that space with pictures.

 

[AlltheWay LeeWhy]

Perhaps I am wrong. It's happened. It will happen again. I can tell you that probably no one likes you taking up all that space with pictures.

Pictures that prove Rocket Lake was an amazing platform if you grabbed the right mobo and right mem kits - I realize for most enthusiasts RKL was a true dud, but it didn't have to be.

So many enthusiasts, like yourself don't know about the Hynix DJR DDR4 kits and how well they overclock with RKL's outstanding IMC (at least when using an ROG bios) - so I think it's a good thing moving from ignorance to understanding.

The Hynix DJR kits are still available for Alder Lake and will be there for Raptor Lake as well, so it's good to spread the word, helps out other enthusiasts running work builds who's apps will benefit greatly from high mem bandwidth. It's like moving from SATA to an NVMe PCIe 4.0 SSD - responsiveness is greatly improved.

Not everyone is using their PC for gaming. (Not that there's anything wrong with that)

I always post images when I feel like I need to prove my experiences actually do exist, and not simply fabricated. lol

This mem kit below is the fastest Hynix DJR DDR4 kit currently available from Gskill, and just take a look at the QVL.

F4-5333C22D-16GTES-(EOL) - Overview - G.SKILL International Enterprise Co., Ltd.

The Trident Z Royal Elite series is the upper echelon of DDR4 DRAM performance and design, featuring a meticulously sculpted crystalline pattern across the polished surface of the aluminum heatspreader, a patented full-length crystalline light bar, and customizable 8-zone RGB lighting. Created...

www.gskill.com

 

[AlltheWay LeeWhy]

Perhaps I am wrong. It's happened. It will happen again. I can tell you that probably no one likes you taking up all that space with pictures.

Dood you won't even come close to my "very old and ancient" RKL build for a million years, so why you keep talking?

 

[DrMrLordX]

View attachment 63972

Let me guess, Geardown 4? You had DDR4-5066 and all that got you was 46.5ns latency? Comet routinely did 40ns at much lower speeds.

The AMD part is a little fishy. Even if they were cutting APU and GPU orders, you would think they would be able to shift to more Milan and consoles.

If Milan suddenly became a lot more available, that would be real bad news for Intel's server business, you would think.

Isn't most of Intel's server market share still fulfilled by14nm Cascade Lake-AP? Even if server CPU demand contracts a bit, AMD could easily cut into that with Milan or Genoa. And some of those cancelled orders are probably N5-family orders.

 

[Det0x]

Dood you won't even come close to my "very old and ancient" RKL build for a million years, so why you keep talking?

Maybe i'm getting old and my memory is faulty, but didnt rocketlake gen11 lose handily to Zen3 both in gaming and almost everything else when both were maxed tweaked/overclocked ? (no i dont consider aida64 as a "benchmark") And when Alder Lake got released it got slaughtered in everything but corner case AVX512 ? (and when you disable e-cores on the early 12gen 12900k's we get same story even in avx512 workloads)

Anyway, think ive seen that picture in 3 different threads now.. in my decently large 4 pump ~7.5L (~2 gallion) watercooling rig i have a other solution with spillproof quickconnects on many of the major components -> its very easy to swap waterblocks/motherboard etc when i can just disconnect waterblock to both CPU and GPU without draining loop (also installed connects on my standalone mo-ra3 and extended the loop after pictures below were taken)

https://imgur.com/a/rlI8N18

 

[igor_kavinski]

HP Dev One With Ryzen 7 PRO 5850U Competes Well Against Intel's Core i7 1280P "Alder Lake P" On Linux - Phoronix


Woah. ADL mobile is barely faster than Zen 3 (plain Zen 3! Not even 3+!) in Linux and consumes considerably more power. Abysmal

 

[DrMrLordX]

Woah. ADL mobile is barely faster than Zen 3 (plain Zen 3! Not even 3+!) in Linux and consumes considerably more power. Abysmal

It's Phoronix, so be careful about drawing too many conclusions from their test suite. The geometric mean includes a lot of really niche stuff, and it also heavily favors ST performance which works quite well for the 12900k vs 5950X but not so well for 1280P vs 5850U.

Also, kind of the wrong thread since we have a dedicated Alder Lake thread (and dedicated Zen3 threads. In multiples).

 

[RTX]

RKL was extremely poor, unless you grabbed an ROG Z590 Apex motherboard and a small number of very QVL specific high-speed mem kits.

Haters gonna hate, I guess.

5066 CL17, 5333 CL19, 5600 CL20, 5866 CL21 all possible with ambient air.

View attachment 63966
View attachment 63967
View attachment 63972

Do you have the aida64 pic for 5866CL22 1T?

 

[Henry swagger]

Intel 14th Gen Meteor Lake Mobility CPU Details Leaked: Coming In 2H 2023 With Triple Hybrid-Core Design, Up To 14 Cores, Xe-LPG GPUs & LPDDR5X-7467 Support

The Intel 14th Gen Meteor Lake CPU lineup consisting of the U, P & H-series mobile chips will be launching in the second half of 2023.

wccftech.com

Maybe the third e-core is gracemont or a new one.. triple core design like arm ? 🤔

 

[HurleyBird]

Intel 14th Gen Meteor Lake Mobility CPU Details Leaked: Coming In 2H 2023 With Triple Hybrid-Core Design, Up To 14 Cores, Xe-LPG GPUs & LPDDR5X-7467 Support

The Intel 14th Gen Meteor Lake CPU lineup consisting of the U, P & H-series mobile chips will be launching in the second half of 2023.

wccftech.com

Maybe the third e-core is gracemont or a new one.. triple core design like arm ? 🤔

Quite different. ARM ecosystem has small cores, big cores, and for high end a single biggest core.

If accurate, seems like Intel is going for a mix that covers peak single threaded perf for one cluster, sheer compute density for another, and power efficiency for the new future cluster. ARM's small cores do both compute density and power efficiency, while Intel's current small cores only do the former.

 

[Exist50]

Intel 14th Gen Meteor Lake Mobility CPU Details Leaked: Coming In 2H 2023 With Triple Hybrid-Core Design, Up To 14 Cores, Xe-LPG GPUs & LPDDR5X-7467 Support

The Intel 14th Gen Meteor Lake CPU lineup consisting of the U, P & H-series mobile chips will be launching in the second half of 2023.

wccftech.com

Maybe the third e-core is gracemont or a new one.. triple core design like arm ? 🤔

It says "LP E-cores" and references "Low-power island CPU offload". I.e. these are the SOC die Atom cores (Gracemont or Crestmont) I've mentioned previously. @dullard, I recall you were skeptical of their existence. Hopefully this puts those doubts to rest.

Quite different. ARM ecosystem has small cores, big cores, and for high end a single biggest core.

For the latest devices, they're using three difference architectures with the X series, A7xx series, and A5xx series.

 

[DrMrLordX]

If accurate, seems like Intel is going for a mix that covers peak single threaded perf for one cluster, sheer compute density for another, and power efficiency for the new future cluster.

That sounds like a hot mess. They need to work on their e-cores to avoid that disaster.

 

[uzzi38]

Intel 14th Gen Meteor Lake Mobility CPU Details Leaked: Coming In 2H 2023 With Triple Hybrid-Core Design, Up To 14 Cores, Xe-LPG GPUs & LPDDR5X-7467 Support

The Intel 14th Gen Meteor Lake CPU lineup consisting of the U, P & H-series mobile chips will be launching in the second half of 2023.

wccftech.com

Maybe the third e-core is gracemont or a new one.. triple core design like arm ?

Huh, wonder if Intel uses Intel 7 for all their internal documents for Meteor Lake still or if this is an old one.

 

[mikk]

They went down from 192 to 128EUs? I wonder why. Maybe a 33% shader count increase with clock speed improvements is good enough for a 2x increase which should be the aim. Also seems like Xe LPG has some changes, reportedly there are no XMX units.

 

[dullard]

It says "LP E-cores" and references "Low-power island CPU offload". I.e. these are the SOC die Atom cores (Gracemont or Crestmont) I've mentioned previously. @dullard, I recall you were skeptical of their existence. Hopefully this puts those doubts to rest.

I am missing details on the reference to "Low-power island CPU offload" for the power supply. Basically, are these LP E-cores limited to the VPU or can it be used for general purpose computing? I'm confused by (A) the 6+8 tile, (B) two LP E cores, and (C) a maximum of 14 cores. (A) + (B) != (C)

I was a bit skeptical simply because I wanted something concrete to go off of. I always err on the side of a less powerful CPU until more details arise.

It does seem that you were correct that the extra size of the SOC is due to the VPU. I guessed a VPU was on Meteor Lake, but it might be the 4th tile. You did correct me saying the 4th tile was not a VPU but that the VPU might be in the SOC.

 

[igor_kavinski]

Basically, are these LP E-cores limited to the VPU or can it be used for general purpose computing?

Intel VPU For AI Inferencing With Atom Cores

My guess is that these are specialized Atom cores acting like accelerators so it may not be possible to run general purpose workloads on them.

What is the purpose of this VPU, anyway? Will it just be an enhancement of the Windows Hello feature to allow face login or will it be something more sinister, like giving your webcam access to Windows and let it keep watch on you and your surroundings for whatever software requires that feature?

 

[repoman27]

I am missing details on the reference to "Low-power island CPU offload" for the power supply. Basically, are these LP E-cores limited to the VPU or can it be used for general purpose computing? I'm confused by (A) the 6+8 tile, (B) two LP E cores, and (C) a maximum of 14 cores. (A) + (B) != (C)

I was a bit skeptical simply because I wanted something concrete to go off of. I always err on the side of a less powerful CPU until more details arise.

It does seem that you were correct that the extra size of the SOC is due to the VPU. I guessed a VPU was on Meteor Lake, but it might be the 4th tile. You did correct me saying the 4th tile was not a VPU but that the VPU might be in the SOC.

It's mentioned on the "continued..." slide, further down the page. Also, here's a link to the original Igor'sLAB post.

I don't think the LP E-cores have anything to do with the VPU, but they probably cannot be used in conjunction with the cores on the compute tile, so they don't include them in the total core count. I reckon they're only used when the system doesn't want to wake the compute tile. The U series maxing out at 12 cores is odd though. Either there is a 4+8 compute tile, or some SKUs will use a 6+8 with two P-cores disabled.

 

[dullard]

What is the purpose of this VPU, anyway? Will it just be an enhancement of the Windows Hello feature to allow face login or will it be something more sinister, like giving your webcam access to Windows and let it keep watch on you and your surroundings for whatever software requires that feature?

I can imagine possibilities. I assume the uses will be limited at first until more developers start to use them.

Things like: automatic image recognition (what if you want only photos of your ex to show up on a slideshow or vise-versa), background elimination in video chats (much better than the extreme oddities that show up now for those of us who tried working from home), video upscaling/downscaling without putting strain on the GPU, webcam color correction, noise reduction in video or audio, robotic controls (or anything with lots of sensors like driving a car), computer aided diagnosis (image processing for medicine), real-time speech translation, etc. Anything where AI can help.

 

[repoman27]

My guess is that these are specialized Atom cores acting like accelerators so it may not be possible to run general purpose workloads on them.

What is the purpose of this VPU, anyway? Will it just be an enhancement of the Windows Hello feature to allow face login or will it be something more sinister, like giving your webcam access to Windows and let it keep watch on you and your surroundings for whatever software requires that feature?

AFAICT, only WCCF Tech / Hassan Mujtaba have said that the VPU is related in any way to the Atom cores. That's not in the slides leaked by Igor'sLAB, and it doesn't make much sense. Why would you use general purpose x86 cores for an AI accelerator? Do previous Movidius Myriad VPUs employ Atom cores?

Here's the Intel product brief for Movidius Myriad X VPUs:

Intel® Movidius™ Myriad™ X VPU Product Brief

Take your imaging, computer vision and machine intelligence applications into network edge devices with the Intel® Movidius™ Myriad™ X VPU.

www.intel.com

 

[coercitiv]

I always err on the side of a less powerful CPU until more details arise.

Indeed, you were also skeptical about a higher core count Meteor Lake die for the desktop in the absence of concrete details.

 

[Exist50]

I am missing details on the reference to "Low-power island CPU offload" for the power supply. Basically, are these LP E-cores limited to the VPU or can it be used for general purpose computing? I'm confused by (A) the 6+8 tile, (B) two LP E cores, and (C) a maximum of 14 cores. (A) + (B) != (C)

I was a bit skeptical simply because I wanted something concrete to go off of. I always err on the side of a less powerful CPU until more details arise.

It does seem that you were correct that the extra size of the SOC is due to the VPU. I guessed a VPU was on Meteor Lake, but it might be the 4th tile. You did correct me saying the 4th tile was not a VPU but that the VPU might be in the SOC.

These cores have absolutely nothing to do with the VPU. They're just normal CPU cores that live on the SoC die.

In the MTL timeframe, at least, I think their existence will be beneficial, but not revolutionary. The use case is that you could shut down the entire compute tile in low-load scenarios where the die-to-die interface, LLC, etc constitute significant overhead. But remains to be seen how often the SoC is able to achieve that state.

Meanwhile, Crestmont (working assumption) on N6 will generally be less efficient and powerful than Crestmont on I4 at mid to high loads. So there's going to be a pretty narrow range where it makes sense.