Question Raptor Lake - Official Thread

[Hulk]

Since we already have the first Raptor Lake leak I'm thinking it should have it's own thread.
What do we know so far?
From Anandtech's Intel Process Roadmap articles from July:

Built on Intel 7 with upgraded FinFET
10-15% PPW (performance-per-watt)
Last non-tiled consumer CPU as Meteor Lake will be tiled

I'm guessing this will be a minor update to ADL with just a few microarchitecture changes to the cores. The larger change will be the new process refinement allowing 8+16 at the top of the stack.

Will it work with current z690 motherboards? If yes then that could be a major selling point for people to move to ADL rather than wait.

 

[nicalandia]

In addition, due the the characteristics of the e-cores and the loss of avx512, I get better performance with no e-cores at all for what I do on my 12700F.

As far as I am aware only the early release Alder Lake CPUs were able to unlock the AVX-512 by disabling the e cores, but on later releases that can't be done because Intel is now Fusing them by laser. Raptor Lake should suffer the same fate.

 

[Markfw]

As far as I am aware only the early release Alder Lake CPUs were able to unlock the AVX-512 by disabling the e cores, but on later releases that can't be done because Intel is now Fusing them by laser. Raptor Lake should suffer the same fate.

But I have seen mine do it, so I know it must be early. Its in a DC log file, one of my team mates noticed it when I said it was taking 300 watts for 8 cores.

Edit: I have just about the largest heatsink you can buy for a LGA1700.

 

[nicalandia]

But I have seen mine do it, so I know it must be early.

Yeah, your CPU clearly has it, I was talking about later release CPUs and Upcoming Raptor Lake CPUs will not enjoy the AVX512

 

[Markfw]

Here is the text from Stefan:
I clicked through your current results list:
All of the small "proof tasks" were calculated with the FMA3 transform.
However, of the three currently visible results of "main tasks", two used the AVX-512 transform right away (llr321_406188011_0, llr321_406188013_0), and one started with the FMA3 transform, was suspended, resumed at 71.57 % progress, and switched to AVX-512 (llr321_406209826_0).

AVX-512 is often pictured as a power hog, but at the same time it can be very power efficient if the workload is suitable, in terms of energy expended for a task. Similar to how GPGPU computing can be a lot more efficient than CPU computing for some types of workloads.

Here: https://forums.anandtech.com/threads/primegrid-challenges-2022.2600827/post-40720971

Edit: the task logs are no longer there, this was in March.

 

[Markfw]

And not to derail, but to answer exist50's desire for proof of the 5950x blowing away the 12700F, even with avx512, its not even close. I realize that I have no e-cores, and the 12900k/ks do, but 300 watts for 8 cores, and 16 cores are blowing it away at 300 watts ???? 80 seconds vs 200 seconds ? and twice the number of tasks running ? Thats like 4 1/2 times the processing power and half the wattage for the 5950x.

Page 6 - PrimeGrid Challenges 2022

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Exist50]

And not to derail, but to answer exist50's desire for proof of the 5950x blowing away the 12700F, even with avx512, its not even close. I realize that I have no e-cores, and the 12900k/ks do, but 300 watts for 8 cores, and 16 cores are blowing it away at 300 watts ???? 80 seconds vs 200 seconds ? and twice the number of tasks running ? Thats like 4 1/2 times the processing power and half the wattage for the 5950x.

Page 6 - PrimeGrid Challenges 2022

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Links are dead, so how can we validate any of your claims?

And you realize what you're claiming, right? That Zen 3 outperforms Golden Cove (with AVX-512 no less) at less than half the power per core. Your 12700 doesn't even have a 200W PL2, much less the 300W you claim. Do you realize how absurd this all sounds? Not a single reviewer has even claimed Zen 3 to be stronger on a per core basis, never mind with half the power budget. And if your task uses AVX-512, it shouldn't even be close.

 

[Markfw]

Links are dead, so how can we validate any of your claims?

And you realize what you're claiming, right? That Zen 3 outperforms Golden Cove (with AVX-512 no less) at less than half the power per core. Your 12700 doesn't even have a 200W PL2, much less the 300W you claim. Do you realize how absurd this all sounds? Not a single reviewer has even claimed Zen 3 to be stronger on a per core basis, never mind with half the power budget. And if your task uses AVX-512, it shouldn't even be close.

Well, it was was discovered by team mates. So you are calling all of the DC crowd liars ? And the term "embarrassingly parallel" most of the time is wrong, at least for DC. As you see, the tasks are one per thread, so 32 threads for the 5950x vs 16 threads for the 12700F. If you don't like it, join DC and run stuff, and find out how CPUs and video cards do work. As far as the 300 watt vs the 200w pl2, I am sure that's not including avx512, and most likely the reason for disabling it in ADL and Raptor lake. I was using a kill-a-watt, so its not my imagination. It was using 170 watts for 8 cores before I saw it jump. The 5950x uses 142 watts, ALWAYS. Both systems are totally stock, both have huge heatsinks.

And for those wondering why ADL talk in Raptor lake, its relevant, especially the avx512 info. I hope the power usage discussion becomes irrelevant in Raptor Lake. They are very closely related though.

 

[Exist50]

Well, it was was discovered by team mates

You're the one making the claims right here, so I'm addressing you. Again, you certainly didn't leave any way to validate the ones you made, and they directly contradict every published review available. Combine that with a history of similar false claims, and...

As far as the 300 watt vs the 200w pl2, I am sure that's not including avx512, and most likely the reason for disabling it in ADL and Raptor lake.

That's not how power limits work. Higher AVX-512 power consumption would be compensated for by lower clocks. Power limits don't get ignored by feature toggles.

 

[nicalandia]

Links are dead, so how can we validate any of your claims?

And you realize what you're claiming, right? That Zen 3 outperforms Golden Cove (with AVX-512 no less) at less than half the power per core.

It would not be the first time AMD Beats the beaks out of Intel on AVX-512 Optimized apps due to sheer core numbers and over all higher speed(AVX lowers the all core speed)

 

[Markfw]

I am not going to derail this thread anymore. The proof was there in March, and many witnesses saw it, and the results. The job numbers and run times are there. If exist50 does not believe it, that is his loss. I am done with this particular discussion. Trying to discuss performance and issues with facts just don't work on some people.

 

[Exist50]

It would not be the first time AMD Beats the beaks out of Intel on AVX-512 Optimized apps due to sheer core numbers and over all higher speed(AVX lowers the all core speed)

He's claimed that the 5950x is more than twice as fast as his 12700f, while consuming half the power. For that to be true, a Zen 3 core has to beat Golden Cove while using a quarter of the per-core power, and that's outright ignoring the ISA advantage from AVX-512. Not a single review shows anything close to that, nor has any verifiable data been provided to substantiate the claim.

Edit: Accidentally said Zen 3 would be using half the power. Actually, it would have to be using 1/4. Even more absurd.

 

[Markfw]

He's claimed that the 5950x is more than twice as fast as his 12700f, while consuming half the power. For that to be true, a Zen 3 core has to beat Golden Cove while using half the per-core power, and that's outright ignoring the ISA advantage from AVX-512. Not a single review shows anything close to that, nor has any verifiable data been provided to substantiate the claim.

I will add this... All those things you are reading are NOT running primegrid tasks. There is the chance that this one app just runs better on AMD. And as I said, the data WAS available in March, they purge the logs after so long.

 

[Exist50]

I will add this... All those things you are reading are NOT running primegrid tasks. There is the chance that this one app just runs better on AMD. And as I said, the data WAS available in March, they purge the logs after so long.

And does that data include the system, settings, power limits, etc? Surely primegrid isn't so unique a task that there's no public data for it or something equivalent.

Numerous reviews have tested Golden Cove in compute workloads, with or without AVX. I can't think of a single review that's shown it as weaker than Zen 3, much less Zen 3 with a quarter the power.

 

[nicalandia]

He's claimed that the 5950x is more than twice as fast as his 12700f, while consuming half the power.

Why is that so hard to believe? While AVX-512 bring HUGE gains(shown on perfect scaling apps) the gains will be dependent on the App and it's AVX-512 implementations.

For example here an 8C/16T 12900k With AVX-515 is actually slower than a stock 12900K(that can't use the AVX-15 instructions) and consumes much more power.

 

[Exist50]

Why is that so hard to believe? While AVX-512 bring HUGE gains(shown on perfect scaling apps) the gains will be dependent on the App and it's AVX-512 implementations.

For example here an 8C/16T 12900k With AVX-515 is actually slower than a stock 12900K(that can't use the AVX-15 instructions) and consumes much more power.

View attachment 63659

View attachment 63660

What exactly were you trying to show here? From your own example, vs P-cores only (i.e. without AVX-512), it gives +92% performance at the cost of +6% more power. That is an enormous efficiency boost, and yet Mark is claiming that even with that ISA advantage, Golden Cove is <1/4 the efficiency of Zen 3. Find me a review that even shows half that gap.

 

[nicalandia]

What exactly were you trying to show here? From your own example, vs P-cores only (i.e. without AVX-512), it gives +92% performance at the cost of +6% more power.

Yet it slower than a 8P + 8E without AVX-512. Just saying that 16 cores can and are faster than 8 cores even if those 8 cores get AVX-512. How much faster? Well it depends on the app used. Not all apps take full advantage of the AVX-512

 

[Exist50]

Yet it slower than a 8P + 8E without AVX-512.

Ok? Yeah, the Gracemont cores actually add quite a bit of MT performance. But that has nothing to do with the claims being made. Actually, he was insisting that they're worthless, so...

 

[nicalandia]

Intel Raptor Lake 13900K Die Shot Interpretation by Locuza





Do you guys find interesting that Raptor Cove with Raptor lake will be releasing before Golden Cove in Sapphire Rapids?

 

[IntelUser2000]

Alder Lake it's on its way out. Raptor Lake has a daunting task ahead of them, Redwood cove with it's smaller(7nm) process might take advantage of the process node and be more efficient(might not be required to be pushed to the limit to compete)

It should be closer in the Raptorlake generation.

There's no Meteorlake-S for K. It'll be Arrowlake-S on 20A.

 

[Markfw]

Intel Raptor Lake 13900K Die Shot Interpretation by Locuza

View attachment 63668



Do you guys find interesting that Raptor Cove with Raptor lake will be releasing before Golden Cove in Sapphire Rapids?

Where is Raptor cove core number 8 ? Wasn't it 8+8 ? Or is it the upper left one that details parts of the CPU core ? (If thats what they are)

 

[Hitman928]

Where is Raptor cove core number 8 ? Wasn't it 8+8 ?

To the right of Display Control Logic. It is just not labeled as a Raptor Cove core but instead is shown broken down into it's sub-units.

 

[nicalandia]

There's no Meteorlake-S for K. It'll be Arrowlake-S on 20A.

Meteor Lake-P should be a powerhouse with 8+16, but that will be out by first quarter of 2024?




Also Any info on how much Performance Gains are we expecting on the Gracemont Cores on Raptor Lake? Or do they remain the same Skylake level? Any info on the performance of Crestmont core IPC? Will they be Ice Lake level?

 

[IntelUser2000]

Meteor Lake-P should be a powerhouse with 8+16, but that will be out by first quarter of 2024?

There's no 8+16 Meteorlake. The highest is rumored to be 6+8. They should be out by early Q1 just like Alderlake mobile.

Also Any info on how much Performance Gains are we expecting on the Gracemont Cores on Raptor Lake? Or do they remain the same Skylake level? Any info on the performance of Crestmont core IPC? Will they be Ice Lake level?

Nothing about changes on Gracemont for RPL. Bit of clocks expected.

 

[TheELF]

Overclocking is DEAD..

Today's Chips(Specially Intel) are pushed to their limits. Which brings us back to the subject. You are getting a Pushed to the Limits CPU for the performance, but it's way power inefficient so what you do about it? Nothing, since lowering the voltages/power will be like a downgrade(from 12900K to 12700K)

You live in the PAST...
Overclocking from bios to a rigid all core is DEAD.

The things you state are the reasons intel released the IXTU and AMD released ryzen master.
The mobo makers send you a mobo that pushes everything to the limit? You run ixtu or ryzen master and tweak all the workloads you are unhappy with until you get the best results, it applies a separate profile for each app you train on it.
You reduce the power for each software as much as you want, until you see a performance degradation or get the temps you want.
It will still be much more powerful than the 12700k because you won't limit it to a rigid 125W you can run some things at 65W if it still has the same performance and others at 80 or 120 or 190 or whatever limit you want.
As the benches at 125W have shown the 12900k is still the most efficient CPU with up to 24 threads, only the 5950x beats it and only at some things, it has 30% more threads though so if it can't beat the 12900k with 30% in everything then what's the point.
The 24thread 12900k at 125W demolishes the 24thread 5900x ,with the 5900x even needing about 8% more power.

DC is pretty much the only thing the 5950x makes any sense for because there it does get better performance and better consumption.

 

[coercitiv]

Then why are you getting an overclocking-oriented chip just to power limit it? You can do that with the 12900 just as well.

Not interested to get in whatever argument is unfolding here, but I'd like to remind everyone about a very specific quirk of Alder Lake.

In the past going with locked parts was a way to save some money or bypass availability issues without losing much performance. I did that with my i7 8700 and it only got better with Comet Lake. However, AFAIK with Alder Lake something changed, locked parts also have locked SA voltage (no official source, but multiple reports from locked parts buyers official source). This can severely limit memory OC potential, to the point where an unlucky user may have trouble running XMP settings. (and I'm talking mild memory speeds, not the really aggressive ones)

I hope Raptor Lake gets rid of this limitation, but I'm not holding my breath. To me it seems they found a way to better enforce their segmentation.