Info 64MB V-Cache on 5XXX Zen3 Average +15% in Games

[Kedas]

Well we know now how they will bridge the long wait to Zen4 on AM5 Q4 2022.
Production start for V-cache is end this year so too early for Zen4 so this is certainly coming to AM4.
+15% Lisa said is "like an entire architectural generation"

 

[DrMrLordX]

Funny how the tides have turned; Zen used to be particularly good at Cinebench, now with Golden Cove, Intel is disproportionately better at it.

Intel was likely keenly aware of this fact when Golden Cove was in development.

Iirc R20 added AVX2 support. Not sure if R15 used AVX or was just limited to SSE though.

Pretty sure R15 was SSE4.x .

 

[Asterox]

We can conclude, AMD and Intel CPU-s are the main actors.

 

[nicalandia]

View attachment 55761

Seems so.

I don't disagree with a lot of the other stuff you said in your post.

Funny how the tides have turned; Zen used to be particularly good at Cinebench, now with Golden Cove, Intel is disproportionately better at it.

Really?

Core i9 12900K processor review (Page 7)

Meet Intel's new flagship CPU, the Core i9 12900K. It is based on the Alder Lake architecture and is reviewed here. This time around, Intel was back at the drawing board, creating a completely new ar...

www.guru3d.com

 

[tamz_msc]

Really?

That's R15. R20 and R23 paint a different picture.

 

[Markfw]

Really?


View attachment 55779

Core i9 12900K processor review (Page 7)

Meet Intel's new flagship CPU, the Core i9 12900K. It is based on the Alder Lake architecture and is reviewed here. This time around, Intel was back at the drawing board, creating a completely new ar...

www.guru3d.com

Based on that, the only reason single core is faster for Intel, is the ludicrous voltage/wattage and resultant frequency of the 12900k. They all equal the 12600k

 

[nicalandia]

Yes but above chart is without clockspeed being normalized which is required, if you want to compare "IPC"

The Only time Anandtech has properly used the SPEC suit to measure IPC have been only "ONCE" where they normalized/ISO speed, was when they measured the IPC gains of a 10nm Skylake CPU Intel released in China. Any other SPEC results they have released after have been at stock speeds

 

[igor_kavinski]

Put down your pitchforks: Intel's 12th-gen CPUs aren't power hogs | PCWorld

Cinebench is possibly the worst case scenario for Alder Lake power consumption. In gaming and other common workloads, Alder Lake doesn't fare as bad as it does in Cinebench. The reason seems to be that Intel hasn't figured out how to design power efficient AVX units.

 

[igor_kavinski]

Intel was likely keenly aware of this fact when Golden Cove was in development.

If they DID optimize their architecture to be more favorable to Cinebench, it backfired spectacularly in terms of crazy power consumption.

 

[igor_kavinski]

Same CPU (12900K) compared to 5900X, has 43% more power usage in Cinebench but 11.4% more power efficient in games.

 

[nicalandia]

Based on that, the only reason single core is faster for Intel, is the ludicrous voltage/wattage and resultant frequency of the 12900k. They all equal the 12600k

Correct and if we normalized the SPEC benchmarks with ISO speed, the 5950X would match or beat Alder Lake but for reasons unknown to us anandtech has refused to do that, they test at stock speed so SPEC becomes just another performance test and not IPC test

 

[Hitman928]

Put down your pitchforks: Intel's 12th-gen CPUs aren't power hogs | PCWorld

Cinebench is possibly the worst case scenario for Alder Lake power consumption. In gaming and other common workloads, Alder Lake doesn't fare as bad as it does in Cinebench. The reason seems to be that Intel hasn't figured out how to design power efficient AVX units.

I think you mean in lightly threaded work loads.

it’s like that for most games because most games don’t use enough CPU cores or threads to push the Core i9 to the point where it will use more power.

Vast majority of games and many common 'consumer apps' are lightly threaded so it shouldn't be a surprise that a 16 core CPU doesn't use anywhere near its max power when running 4 cores. If you actually want to use all those cores though, at stock settings, it's going to use a ton of power. Intel did this because they have a performance core count disadvantage to AMD's best so they pushed the power like crazy to try and make up for it, performance wise. In lightly threaded apps, AMD's desktop (and server) CPUs take a power hit due to the chiplet architecture so the power use comes out more equal. As more and more cores are loaded, the chiplet power overhead becomes less and less a percentage of the overall power and so Zen 3's more efficient cores start to prove their efficiency.

 

[DrMrLordX]

If they DID optimize their architecture to be more favorable to Cinebench, it backfired spectacularly in terms of crazy power consumption.

They got the benchmark win. Not everyone puts the power consumption and performance on the same graph. Makes it easy for astroturfers to ignore that little detail.

 

[nicalandia]

This is the ONLY time Anandtech(website) has done a properly IPC test with SPEC. It was only used to see if there was any performance increase on a 10nm Skylake based CPU and there was none. But I wonder why they have stopped using that as a measurement of IPC when is about the best suit for doing just that? At ISO Speed the 5950X would match or beat Alder Lake

AnandTech Forums: Technology, Hardware, Software, and Deals

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

www.anandtech.com

 

[igor_kavinski]

I think you mean in lightly threaded work loads.

Intel 12th Gen 'Alder Lake' Review - Ipc-test - Tweakers



Question 1: 23.3% faster in R15 because AVX2 is not being used so 12900K gets to boost higher?

Question 2: Only 13.4% faster in R20 because AVX2 usage causes more power draw and limits boost clock, correct?

Please clarify. Thanks.

 

[nicalandia]

Intel 12th Gen 'Alder Lake' Review - Ipc-test - Tweakers

View attachment 55786

Question 1: 23.3% faster in R15 because AVX2 is not being used so 12900K gets to boost higher?


Please clarify. Thanks.

DDR4 With 2666 is the worst case scenario for Zen3

This is a better result without gimping or holding back Zen3

 

[igor_kavinski]

DDR4 With 2666 is the worst case scenario for Zen3

This is a better result without gimping or holding back Zen3

View attachment 55788

But here, 12900K suffers the worse latency of DDR5, no?

 

[nicalandia]

But here, 12900K suffers the worse latency of DDR5, no?

Not really, there are almost no difference between DDR4 and DDR5. So it's best to compare Alder Lake with Tight Timings DDR4 and Zen 3 with best RAM

 

[eek2121]

View attachment 55782
View attachment 55783

Same CPU (12900K) compared to 5900X, has 43% more power usage in Cinebench but 11.4% more power efficient in games.

That is system power usage, not CPU power usage. System power usage has too many variables to be useful for this type of test.

 

[eek2121]

Not really, there are almost no difference between DDR4 and DDR5. So it's best to compare Alder Lake with Tight Timings DDR4 and Zen 3 with best RAM

Well the best would be to use the recommended memory for both @ JEDEC speeds/timings.

 

[tamz_msc]

But here, 12900K suffers the worse latency of DDR5, no?

IPC should be measured at the intended operating frequency, not some arbitrary low frequency like 2.4 or 3.5 GHz. That is because when you artificially lower frequency the CPU is not stalling on memory any more. In essence the CPU is supposed to do "other things" while executing an instruction that is waiting for cache miss data; when you artificially lower the frequency, things like the size of caches come into play.

That's why at low frequencies, the larger caches of Golden Cove help in keeping the CPU stalls to a minimum. With 2666 MHz memory and smaller caches the effect is just more pronounced on Zen, and thus it performs worse.

 

[Hitman928]

Intel 12th Gen 'Alder Lake' Review - Ipc-test - Tweakers

View attachment 55786
View attachment 55787
Question 1: 23.3% faster in R15 because AVX2 is not being used so 12900K gets to boost higher?

Question 2: Only 13.4% faster in R20 because AVX2 usage causes more power draw and limits boost clock, correct?

Please clarify. Thanks.

Others already mentioned some issues but additionally on Zen, the infinity fabric is tied to the memory speed so running lower memory speeds has an additional performance penalty on Zen.

As far as your boosting theory, it looks to me like they are running at fixed clocks so no boosting going on.

 

[deasd]

Alright everybody please back to topic lol. Someone ran a EPYC 7773X which has 96mb L3 per CCD, @2.8Ghz max, and CPUZ score:

https://twitter.com/x/status/1480931095898386432

would be score ~700 ST @4.2Ghz?

edit: clocks still questionable since CPUID string read as 34/21 for this ES....

 

[nicalandia]

Alright everybody please back to topic lol. Someone ran a EPYC 7773X which has 96mb L3 per CCD, @2.8Ghz max, and CPUZ score:

https://twitter.com/x/status/1480931095898386432

would be score ~700 ST @4.2Ghz?

I was trying to find the correct thread to post Milan-X info you just posted, but could not find it. So this is a Milan-X Thread now right?

 

[deasd]

I was trying to find the correct thread to post Milan-X info you just posted, but could not find it. So this is a Milan-X Thread now right?

Well I think this belongs to Zen3D... isn't it? If this still off topic... alright sry😵

 

[nicalandia]

Can someone here with deep pockets get one of these for testing?

EPYC 7773X Milan-X QS CPU Hits Chinese Market at $2,800

Cheaper than retail

www.tomshardware.com