Question Zen 6 Speculation Thread

[Abwx]

>1.7x. it can be 2.2x perf for all you care.

No, it cant, laws of physics cant be trolled contrary tp forums.

They bench sockets at rated power, which is 500W vs 600W here.
600W 256c Venice has >1.7x PPW delta relative to 500W 192c 9965.

You have trouble using basic maths, if perf/W is 1.7x at 1.7x the perfs it means that the comparison is made at 500W, laws of physics state that power is increasing as a square of frequency, at least at medium frequencies.

The consequence is that PPW decrease linearly with frequency increasement, if you increase power by 1.2x then perf increase by sqrt(1.2) and PPW decrease by sqrt(1.2).

 

[Joe NYC]

With 1.7x better efficiency this would mean 1.85x the perf, period.



Lol, the pot that call the kettle, it s rather you who should get out of basic arithmetics irrationality, sure that if 1.7 x 1.2 = 1.7 was to be right then everything would be possible,
but for sure that i have more confidence on AMD that on someone who s left downplaying their numbers because of his ego and unability to accept their numbers as being way closer to reality than his random speculations.

I would not go too deep into the weeds nitpicking between 1.7 and 1.85 from one vague slide.

 

[adroc_thurston]

No, it cant, laws of physics cant be trolled contrary tp forums

It's mildly amusing that you can't even read vendor pptware.
Idk what the > sign is for.
Maybe for moar?

if perf/W is 1.7x at 1.7x the perfs it means that the comparison is made at 500W

What's the > sign I wonder.
It means uhhhhhhhhh, something

 

[Abwx]

I would not go too deep into the weeds nitpicking between 1.7 and 1.85 from one vague slide.

The thing is that AMD made a comparison with an existing SKU, higher TDP with Zen 5 is supposed to be 500W, that s the basis of their number and 1.7x better PPW is in line with the expected process PPW improvement.

 

[StefanR5R]

@Abwx, AMD presented a new projection at the "Financial Analyst Day" in November which is no longer the same as their projection presented at their "Advancing AI" event in June. See post #7,441. (Furthermore, the old projection was given without additional detail, while the new projection, or estimate, was qualified with an endnote.) Make of these vendor estimates what you will; they were both said in public, but one was made months closer to the launch than the other.

 

[Abwx]

@Abwx, AMD presented a new projection at the "Financial Analyst Day" in November which is no longer the same as their projection presented at their "Advancing AI" event in June. See post #7,441. (Furthermore, the old projection was given without additional detail, while the new projection, or estimate, was qualified with an endnote.) Make of these vendor estimates what you will; they were both said in public, but one was made months closer to the launch than the other.

That s precisely the latest infos that are discussed at this point.

 

[Markfw]

The thing is that AMD made a comparison with an existing SKU, higher TDP with Zen 5 is supposed to be 500W, that s the basis of their number and 1.7x better PPW is in line with the expected process PPW improvement.

I have a 500w Zen 5 9755, in fact 2 of them. They rarely go over 370 watt or so, except when fully utilized on all cores with very heavy avx-512 code running, THEN they hit 480 watts or so @2.7 ghz.

The point being that AMD used MAX values for load to come up with the 500, not normal mixed usage. I know this for a fact. and the poor VRMs are what is slowing it down at that. If my motherboard has better VRM heatsinks it might push the 500 watt limit and the ghz run speed.

 

[StefanR5R]

That s precisely the latest infos that are discussed at this point.

Everybody else talks about >, you keep talking about =.

 

[StefanR5R]

I have a 500w Zen 5 9755, in fact 2 of them. They rarely go over 370 watt or so, except when fully utilized on all cores with very heavy avx-512 code running, THEN they hit 480 watts or so @2.7 ghz.

Your samples (100-000001245-08, 100-000001535-05 ?) may not have the same boosting algorithm as the final production samples (100-000001443). With 4.1 GHz all-core boost limit of the 9755, I would expect even a lighter software workload to pull the whole 500 W when all hardware threads are loaded.

edit: product IDs added

edit 2: Meanwhile I was looking around a little if I could find something about power draw of SPEC CPU 2017 Integer Rate on high core count CPUs, but haven't found any.

 

[branch_suggestion]

You don't need to go far back for an example.

Page 148 - Discussion - Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 148 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Markfw]

Your samples (100-000001245-08, 100-000001535-05 ?) may not have the same boosting algorithm as the final production samples (100-000001443). With 4.1 GHz all-core boost limit of the 9755, I would expect even a lighter software workload to pull the whole 500 W when all hardware threads are loaded.

edit: product IDs added

edit 2: Meanwhile I was looking around a little if I could find something about power draw of SPEC CPU 2017 Integer Rate on high core count CPUs, but haven't found any.

the 1245-08 is ES, the 1535-05 is a QS, should be the same as production/retail. If I load 256 jobs of WCG , it uses 370 watts on kill-a-watt. With 128 cores load with no SMT and heavy avx-512 (primegrid) it uses 480.

That all I am sure of.