Discussion Intel current and future Lakes & Rapids thread

[naukkis]

Very lightly threaded (up to 2 cores) performance should go to ADL, but when more cores are loaded, I think RMB will match then pull away in perfromance. Reviews will be interesting.

But why? 15W Rembrandt boost clocks are match to 15W Alderlake and IPC is very similar. Rembrandt will have 16MB L3 against 12MB in Alderlake and Rembrandt very likely will sustain boost clocks better. I can't see any strong points for Alderlake.

 

[IntelUser2000]

They can.

That first slide is obviously a jab at Apple's M1 Max slide few months ago.

 

[insertcarehere]

Gonna be weird though. A lot of AMD U series laptops run at 25W, which is into P SKU territory, a battle that AMD will probably lose. They should look much better at 15W, but they need actual devices with that implementation. Hope we finally get some good AMD T&Ls this time.

Indeed, given that at the very least Dell is putting the 28w SKUs in their ultrabook XPS 13 (~1.2kg): https://www.notebookcheck.net/Dell-...-Forcepad-and-no-headphone-jack.585059.0.html

I doubt that there'd be a lot of Alder Lake-U laptops around given that they'd instantly be at a disadvantage performance wise, which would look bad for the laptop OEM even if there are positive trade-offs.

 

[naukkis]

Yes, but what you have to understand is that those 6 performance cores are faster than the 6 performance cores in Zen 3. Quite a bit faster than the laptop variants. We will need reviews, of course, but if the Intel parts can keep the clocks up, Intel will win the performance crown.

But they aren't. IPC is pretty much identical for Zen3 and Goldencove, and in those mobile variants boost clocks are pretty much identical. And those Goldencove cores burn power like there's no tomorrow so Rembrandt probably will sustain it clocks higher.

 

[eek2121]

But they aren't. IPC is pretty much identical for Zen3 and Goldencove, and in those mobile variants boost clocks are pretty much identical. And those Goldencove cores burn power like there's no tomorrow so Rembrandt probably will sustain it clocks higher.

Incorrect. CPU 2021 Benchmarks - Compare Products on AnandTech

EDIT: 12900k review: The Intel 12th Gen Core i9-12900K Review: Hybrid Performance Brings Hybrid Complexity (anandtech.com)

 

[Hitman928]

But why? 15W Rembrandt boost clocks are match to 15W Alderlake and IPC is very similar. Rembrandt will have 16MB L3 against 12MB in Alderlake and Rembrandt very likely will sustain boost clocks better. I can't see any strong points for Alderlake.

Alderlake has higher IPC. I didn't notice ADL-U cuts down L3 further than ADL-H/P. Still, it is close enough in L3 that I expect ADL will still have a small advantage. Is L3 a victim cache in ADL? It will be interesting to see how much performance is effected by reduced L3 in ADL versus Cezanne.

 

[IntelUser2000]

I doubt that there'd be a lot of Alder Lake-U laptops around given that they'd instantly be at a disadvantage performance wise, which would look bad for the laptop OEM even if there are positive trade-offs.

The U chips set at 9W will have a good use in the very thin and light market such as Lenovo Titanium X1 line.

Although seriously I'd like to see a proper Lakefield replacement.

 

[naukkis]

Incorrect. CPU 2021 Benchmarks - Compare Products on AnandTech

IPC means instructions per clock. Alderlake desktop is faster because it has faster clocks than desktop Zen3, seems that isn't the case with mobile variants.

 

[IntelUser2000]

IPC means instructions per clock. Alderlake desktop is faster because it has faster clocks than desktop Zen3, seems that isn't the case with mobile variants.

No it's not.

Intel Core i9-12900K, i7-12700K & i5-12600K im Test

Intel Core i9-12900K, i7-12700K und i5-12600K sind da. Die Alder-Lake-CPUs in der ausführlichen Test-Analyse und im Vergleich zu Ryzen 5000.

www.computerbase.de

Golden Cove is 11% faster than Zen 3 per clock. Also I remember while Intel chips perform within 2% between two platforms the AMD chips have a greater gap between desktop and mobile.

Still 2+8 is not an optimal 15W configuration.

 

[eek2121]

IPC means instructions per clock. Alderlake desktop is faster because it has faster clocks than desktop Zen3, seems that isn't the case with mobile variants.

We know what IPC is, but there is very little clock difference between the two chips. The 5950X is around 4.9-5.0 GHz vs ADL-S desktop being around 5.1-5.2 GHz. That is a 5-6% difference in clock speed. The difference in IPC is greater than 5-6%.

EDIT: Oh, and AMD halves the L3 cache on its mobile chips. That can cause as much as a 30% drop in performance depending on the workload.

 

[Exist50]

Indeed, given that at the very least Dell is putting the 28w SKUs in their ultrabook XPS 13 (~1.2kg): https://www.notebookcheck.net/Dell-...-Forcepad-and-no-headphone-jack.585059.0.html

I doubt that there'd be a lot of Alder Lake-U laptops around given that they'd instantly be at a disadvantage performance wise, which would look bad for the laptop OEM even if there are positive trade-offs.

Wonder if there will be any P SKUs with a 15W TDP-down option. But those have been pretty rare in prior gens.

 

[naukkis]

We know what IPC is, but there is very little clock difference between the two chips. The 5950X is around 4.9-5.0 GHz vs ADL-S desktop being around 5.1-5.2 GHz. That is a 5-6% difference in clock speed. The difference in IPC is greater than 5-6%.

Those are max single core boost clocks, sustained clocks are very different. Alderlake desktop sustain almost 1GHz higher clocks in multi-core tests than stock 5950x.

 

[IntelUser2000]

Wonder if there will be any P SKUs with a 15W TDP-down option. But those have been pretty rare in prior gens.

Might end up only being relevant in paper. You have the Tigerlake-Y chips set at 19W or something in some systems. That's way beyond the official limit as the Y doesn't have cTDPup, not officially. Also 19W doesn't exist anywhere.

Intel since few generations ago probably gave manufacturers leeway into setting TDP pretty much at any setting they wish.

 

[naukkis]

No it's not.

Intel Core i9-12900K, i7-12700K & i5-12600K im Test

Intel Core i9-12900K, i7-12700K und i5-12600K sind da. Die Alder-Lake-CPUs in der ausführlichen Test-Analyse und im Vergleich zu Ryzen 5000.

www.computerbase.de

Golden Cove is 11% faster than Zen 3 per clock. Also I remember while Intel chips perform within 2% between two platforms the AMD chips have a greater gap between desktop and mobile.

Goldencove has very good SIMD-IPC, but for normal integer code it's a tie with Zen3.

 

[insertcarehere]

Those are max single core boost clocks, sustained clocks are very different. Alderlake desktop sustain almost 1GHz higher clocks in multi-core tests than stock 5950x.

Alder Lake P cores sustain almost 1 ghz higher clocks, meanwhile the significantly weaker e-cores are clocking generally less than the Zen 3 cores in those same tests, and that's with no overall core count advantage to Intel (12900k vs 5950x).

Here, Intel has an overall core count advantage of (6P+8E) vs 8P for 28w and (2P+8E) vs 8P for 15w, and it must be mentioned that Cezanne has a clear IPC deficit vs desktop Zen 3 due to the reduced L3.

 

[eek2121]

Goldencove has very good SIMD-IPC, but for normal integer code it's a tie with Zen3.

Once again, incorrect. SPECint can be used to measure integer performance. See here: CPU Tests: SPEC ST Performance on P-Cores & E-Cores - The Intel 12th Gen Core i9-12900K Review: Hybrid Performance Brings Hybrid Complexity (anandtech.com)

 

[IntelUser2000]

The 2+8 could perform better but right now it's a wildcard as things can change significantly based on how it really scales and how well the power management of the whole system is under load.

 

[Hitman928]

Goldencove has very good SIMD-IPC, but for normal integer code it's a tie with Zen3.

That's true, on the integer side Zen3 and ADL are basically a tie for IPC. Still, at the H level, ADL will have a decent amount more L3 cache and obviously we're not talking strictly integer performance here. RMB should compete well at every level, but ADL should be the 45+W CPU leader. At ~30W it's less clear. At 15W, RMB should win. Obviously this is all pending reviews as real world clock frequencies will play a big role.

 

[naukkis]

Alder Lake P cores sustain almost 1 ghz higher clocks, meanwhile the significantly weaker e-cores are clocking generally less than the Zen 3 cores in those same tests, and that's with no overall core count advantage to Intel (12900k vs 5950x).

Here, Intel has an overall core count advantage of (6P+8E) vs 8P for 28w and (2P+8E) vs 8P for 15w, and it must be mentioned that Cezanne has a clear IPC deficit vs desktop Zen 3 due to the reduced L3.

In desktop you are comparing 241W Alderlake to 142W Zen3. Even in desktop situation changes when power usage is equalized.

 

[Hitman928]

Once again, incorrect. SPECint can be used to measure integer performance. See here: CPU Tests: SPEC ST Performance on P-Cores & E-Cores - The Intel 12th Gen Core i9-12900K Review: Hybrid Performance Brings Hybrid Complexity (anandtech.com)

That shows in SPECint that ADL is 4.7% faster with ~3% faster clock. Very little difference in IPC between the two.

 

[IntelUser2000]

That's true, on the integer side Zen3 and ADL are basically a tie for IPC. Still, at the H level, ADL will have a decent amount more L3 cache and obviously we're not talking strictly integer performance here.

L3 matters lot less for Intel. 16MB vs 12MB is only 33% difference and if you double it you might get 3-5%, so in this case it's 1-3%.

 

[Hitman928]

L3 matters lot less for Intel. 16MB vs 12MB is only 33% difference and if you double it you might get 3-5%, so in this case it's 1-3%.

Do we know that this holds true for ADL though?

 

[naukkis]

L3 matters lot less for Intel. 16MB vs 12MB is only 33% difference and if you double it you might get 3-5%, so in this case it's 1-3%.

Because of DDR5 memory subsystem - Rembrandt will have also have it.

 

[eek2121]

That's true, on the integer side Zen3 and ADL are basically a tie for IPC. Still, at the H level, ADL will have a decent amount more L3 cache and obviously we're not talking strictly integer performance here. RMB should compete well at every level, but ADL should be the 45+W CPU leader. At ~30W it's less clear. At 15W, RMB should win. Obviously this is all pending reviews as real world clock frequencies will play a big role.

How do you figure?

SPECint 2017 - Zen3 vs. Golden Cove (hopefully no typos)

500.perlbench_r: GC is 22.69% faster
502.gcc_r: GC is 17.18% faster
505.mcf_r - memory latency bound, both are tied.
520.omnetpp_r: GC is 6.47% faster
5.23.xalancbmk_r: GC is 10.21% faster
5.25.x264_r: GC is 13.07% faster
5.31.deepsjeng_r: GC is 6.1% faster
5.41.leela_r: Zen 3 is 3.63% faster
5.48.exchange2_r: GC is 4.58% faster
5.56.xz_r: Zen 3 is 9.54% faster

Zen 3 only wins in 2 of those tests above.

 

[IntelUser2000]

Do we know that this holds true for ADL though?

It shouldn't matter a whole lot. It's more than a victim cache for Intel chips, that's the whole point of the ring bus.

Also the mobile AMD CPUs perform few % less than on desktops. That's clear from SPEC comparisons.

I think Intel will lead significantly on the H, as Zen 3+ is few single digit % faster and it has similar clocks. It's the U segment where Remembrandt improves a lot in regards to clocks.

Because of DDR5 memory subsystem - Rembrandt will have also have it.

No. And this goes all the way back to Nehalem. Back when they had L2 as LLC such as Core 2 it mattered more. Even then it was 5-7% for doubling the size.