Discussion Intel current and future Lakes & Rapids thread

[nicalandia]

intel avx 512 support - Chaos Forums

Xeon skylake-sp supports avx. Will vray next use avx to improve performance?

forums.chaos.com

 

[msj10]

It should be easier to port around than RWC, but 3 entirely different nodes would be too much.

lion cove can still be on 3 different nodes
ARL-P on 20A
ARL-S on N3
LNL-M which we don't know what node it uses.

 

[Exist50]

lion cove can still be on 3 different nodes
ARL-P on 20A
ARL-S on N3
LNL-M which we don't know what node it uses.

I assume that both ARL and LNL use the same node, likely N3B or N3E. But even if there were slight differences (e.g. a LNC variant on 18A), that would still be substantially less effort than Intel 3, Intel 20A, and TSMC N3.

Obviously, there's nothing theoretically stopping them from doing that, but I'd have to imagine it would be stretching their team too thin.

 

[nicalandia]

Hm, perhaps. Shouldn't have any particular issues though. No big.little or anything.

Threadripper winning for an int workload by this kind of margin, iso core count, wouldn't be very unusual, but I thought V-Ray was basically a best case scenario for SPR. Heavily dependent on AVX throughput.

Also, remember that the W7-3455 Is a Locked CPU with low base clocks

 

[Harry_Wild]

Intel 14th Gen Meteor Lake CPUs Reportedly Target Over 50% Performance Per Watt Gain Over 13th Gen Raptor Lake

Intel is targeting some huge efficiency and graphics performance figures of over 50% with its 14th Gen Meteor Lake CPUs.

wccftech.com

Intel Reportedly Targeting Over 50% Performance Per Watt Improvement With 14th Gen Meteor Lake CPUs
'Based on information shared by a highly credible leaker and insider, OneRaichu, is stated that the 14th Gen Intel Meteor Lake CPUs are targetting some big performance and efficiency gains. Intel's 14th-Gen Meteor Lake CPUs are going to feature a brand new core architecture for both P-Cores & E-Cores and while the hybrid implementation is intact, the CPU itself will make use of several other IPs that will be fused together via multiple chiplets.

50% increased over Raptor Lake CPUs!

 

[DrMrLordX]

@Harry_Wild

Gee, sounds great until you remember the clock regressions that are likely to happen.

 

[Exist50]

@Harry_Wild

Gee, sounds great until you remember the clock regressions that are likely to happen.

I doubt there will be any notable clock regressions in mobile. If you're still referencing that hack who claims it'll lose a GHz without cobalt, don't waste your time.

The problem for MTL is reality vs targets.

 

[Markfw]

I doubt there will be any notable clock regressions in mobile. If you're still referencing that hack who claims it'll lose a GHz without cobalt, don't waste your time.

The problem for MTL is reality vs targets.

I will be glad to see them if thats the case (40% or more), I might even try one !

Edit: if they add avx-512 back in, I might try it. What I would like to see it 16 P-cores and avx-512. That should compete with Zen 4. But if Zen 5 is out by then ????

 

[deasd]

Intel 14th Gen Meteor Lake CPUs Reportedly Target Over 50% Performance Per Watt Gain Over 13th Gen Raptor Lake

Intel is targeting some huge efficiency and graphics performance figures of over 50% with its 14th Gen Meteor Lake CPUs.

wccftech.com

Intel Reportedly Targeting Over 50% Performance Per Watt Improvement With 14th Gen Meteor Lake CPUs
'Based on information shared by a highly credible leaker and insider, OneRaichu, is stated that the 14th Gen Intel Meteor Lake CPUs are targetting some big performance and efficiency gains. Intel's 14th-Gen Meteor Lake CPUs are going to feature a brand new core architecture for both P-Cores & E-Cores and while the hybrid implementation is intact, the CPU itself will make use of several other IPs that will be fused together via multiple chiplets.

50% increased over Raptor Lake CPUs!

it seems someone hinted it's Cannon-Lake story repeating.

https://twitter.com/x/status/1622535671025131521

 

[Exist50]

I will be glad to see them if thats the case (40% or more), I might even try one !

Edit: if they add avx-512 back in, I might try it. What I would like to see it 16 P-cores and avx-512. That should compete with Zen 4. But if Zen 5 is out by then ????

Well we're extremely unlikely to see such efficiency gains, at least across the VF curve. Either way, hybrid is here to stay, and we have no line of sight for AVX-512 to return. Certainly it won't in MTL. Though 2024 should be a Zen 5 vs Lion Cove + Skymont fight, which should be substantially more interesting.

it seems someone hinted it's Cannon-Lake story repeating.

Broadwell and Cannon Lake were very different things. MTL seems to be much more like the former.

 

[alcoholbob]

Looks like China is getting a 13790F that has 3MB more cache than the 13700F. I wonder if Raptor Lake refresh is going to be Intel looking for creative ways to increase cache…

 

[igor_kavinski]

Bad news for Harry_Wild? Is he going to wait for this Black Edition CPU now?

China-exclusive Intel Core i7-13790F tested in CPU-Z, costs 2999 RMB - VideoCardz.com

Intel Core i7-13790F tested in CPU-Z Two leakers have provided some new details on this brand-new Intel Core CPU. Just a day after first retailers mentioned this new processor, we now have the price, partial specs and first benchmark figures. As mentioned before, the Core i5-13490F and Core...

videocardz.com

 

[Joe NYC]

Looks like China is getting a 13790F that has 3MB more cache than the 13700F. I wonder if Raptor Lake refresh is going to be Intel looking for creative ways to increase cache…

I would not be surprised. Raptor Lake already increased L3 in 13900 to 36 MB, which is more than vanilla Zen 3 and Zen 4. It was 20% increase over Alder Lake, and no doubt, between 20% increase in cache and increase in clock speed, those 2 were majority of gains of Raptor vs. Alder Lake.

 

[Exist50]

Looks like China is getting a 13790F that has 3MB more cache than the 13700F. I wonder if Raptor Lake refresh is going to be Intel looking for creative ways to increase cache…

It'll probably be the same SKUs we have +100MHz or so, and that's it. I really doubt we'll see actual new silicon.

 

[nicalandia]

Looks like China is getting a 13790F that has 3MB more cache than the 13700F. I wonder if Raptor Lake refresh is going to be Intel looking for creative ways to increase cache…

Both the 13790F and the 13700F are the same silicon die as the 13900F but Intel disabled 8 e cores, in the 13970F model they enabled the additional cache found on the disabled e cores

 

[Hulk]

it seems someone hinted it's Cannon-Lake story repeating.

https://twitter.com/x/status/1622535671025131521

I wrote the same a few months ago. Between the huge amount of cores in the 13900K and the crazy high clocks it's highly unlikely Intel will have a desktop part to compete with it in the next generation. I'm betting 14th gen will be aimed at mobile. Then once the node is developed and higher clocks achieved it'll head for the desktop.

Every now and then Intel pushes themselves into a corner by going extreme with one generation either in terms of core count or frequency and they have difficultly beating previous gen performance until the node is really worked out for higher frequencies.

Broadwell is a good example as they had the Haswell based 4790K to contend with as far as high clocks for the time.

Then Rocket Lake had to contend with Comet, 8 cores vs 10, in many MT applications Comet was quite competitive.

And now with Raptor at 24 cores and 6GHz that's a high bar to reach for a new node out of the gate. If they had held back Raptor and stayed on Alder, 8+8 at Alder clocks would be been a much lower bar, but Zen 4 foiled that possibility!

 

[moinmoin]

I wrote the same a few months ago. Between the huge amount of cores in the 13900K and the crazy high clocks it's highly unlikely Intel will have a desktop part to compete with it in the next generation. I'm betting 14th gen will be aimed at mobile. Then once the node is developed and higher clocks achieved it'll head for the desktop.

Every now and then Intel pushes themselves into a corner by going extreme with one generation either in terms of core count or frequency and they have difficultly beating previous gen performance until the node is really worked out for higher frequencies.

Broadwell is a good example as they had the Haswell based 4790K to contend with as far as high clocks for the time.

Then Rocket Lake had to contend with Comet, 8 cores vs 10, in many MT applications Comet was quite competitive.

And now with Raptor at 24 cores and 6GHz that's a high bar to reach for a new node out of the gate. If they had held back Raptor and stayed on Alder, 8+8 at Alder clocks would be been a much lower bar, but Zen 4 foiled that possibility!

This may look like bad planning, but there is more to it.

When we look at AMD it gets interesting in that AMD originally warned that matching and increasing clocks on newer nodes would be increasingly harder. But then AMD managed to do just that twice now. The decisive trade off there is that in both cases AMD took the sweet time before eventually moving to a newer node, so far making them more of a latecomer respectively.

For Intel on the other hand the node switches so far have been an integral part of offering the density and efficiency improvements even with otherwise unchanged core designs (tick), with optimizations (including higher clocks) coming in the following gens (tock).

Both approaches have advantages and disadvantages. Currently AMD approach seems more sensible to me since node progress is slowing anyway and optimizing for specific nodes, while taking longer, allows it to make more out of it in an economical sense as well with the results still being perfectly competitive.

 

[nicalandia]

Intel Xeon Platinum 8480(56C/112T) vs Intel Xeon W9 3495X

HP HP Z8 Fury G5 Workstation vs Unknown - Geekbench

browser.geekbench.com

Both Windows version, The 3495X is faster in ST and about the same in MT as they are held back by the same TDP

 

[mikk]

Raptor mobile efficiency isn't great, AMD is more efficient with longer battery life. If they can fix the high uncore power draw and add a second voltage rail for the E cores beside the process node advantage they can improve with MTL big time when it comes to efficiency.

 

[IntelUser2000]

Raptor mobile efficiency isn't great, AMD is more efficient with longer battery life. If they can fix the high uncore power draw and add a second voltage rail for the E cores beside the process node advantage they can improve with MTL big time when it comes to efficiency.

The big gain on Meteorlake is going to be the SoC cores.

Alder/Raptor regressed and they can fix that too of course.

Broadwell and Cannon Lake were very different things. MTL seems to be much more like the former.

They are related somewhat. Both are results of delayed and neutered processes. Broadwell sucked because 14nm had serious issues, hence the 6-12 month delay. Of course 10nm and Cannonlake far surpassed both that's why almost no one cares about the 14nm delay. Both brought to you by Kraznich.

 

[Hulk]

Intel even gave the "desktop" Broadwell versions 128MB L4 probably in an effort to make them more competitive in some way with Haswell due to the frequency regression. I put desktop in quotes because those parts were as rare as hen's teeth.

 

[igor_kavinski]

I put desktop in quotes because those parts were as rare as hen's teeth.

Not now though. I think last time I checked, Ali Express had about 10,000 i7-5775Cs.

 

[IntelUser2000]

Intel even gave the "desktop" Broadwell versions 128MB L4 probably in an effort to make them more competitive in some way with Haswell due to the frequency regression. I put desktop in quotes because those parts were as rare as hen's teeth.

That's likely not the reason for doing so. They just decided to scale up some mobile parts and sell it as unlocked desktop. It's clock scaling was very poor. There were good indications that the chip was a mobile base.

The PC executive within Intel said they regretted not having full desktop line for Broadwell. They consciously never planned for desktop.

 

[jpiniero]

Intel starts End-of-Life process for 11th Gen Core "Rocket Lake" CPUs - VideoCardz.com

Intel Rocket Lake-S going EOL next year Intel has now started the product discontinuance process for its 11th Gen Core series. The 11th Gen Core “Rocket Lake-S” series are now on their final path. Intel has issued “Product Change Notification” earlier this week informing about the upcoming...

videocardz.com

As Videocardz notes, Intel is EOLing Rocket Lake. This isn't the end of Core on 14 nm though since they are still continuing to produce other generation "Embedded" models (read: You don't get the discounts anymore).

 

[AMDK11]

Are there any reliable sources that say ArrowLake will actually get LionCove cores?

This would be odd considering Intel officially claims LunarLake and LionCove is a brand new project from the ground up.

It's more fitting that Redwood Cove from Meteor Lake is a new x86 core and ArrowLake is based on RedwoodCove(+?).

Something doesn't add up here.