Intel Xeon 6980P vs. AMD EPYC Power Efficiency / Performance-Per-Watt Benchmarks

[phoronix]

Phoronix: Intel Xeon 6980P vs. AMD EPYC Power Efficiency / Performance-Per-Watt Benchmarks

Earlier this week in the launch-day Intel Xeon 6980P Granite Rapids review/benchmarks I unfortunately wasn't able to provide any CPU power consumption and performance-per-Watt benchmarks due a Linux kernel issue and the minimal time ahead of launch for testing. I've now repeated the Xeon 6980P benchmarking on the Linux 6.8 kernel of Ubuntu 24.04 LTS with power monitoring working and have those power efficiency numbers to share today for how Granite Rapids compares to prior Emerald Rapids / Sapphire Rapids / Ice Lake and against the current AMD EPYC Bergamo/Genoa(X) competition.

https://www.phoronix.com/review/intel-xeon-6980p-power

[geerge]

It's more comparable to the 9754, both at 128 cores with intel having roughly double the cache (instead of comparing to the 9684X with 96 cores with AMD having roughly double the cache). The intel part has AVX512 so also parity, the main differences are intel additionally having AMX, which shows as some clear wins in some benchmarks and should tip the scales slightly in intel's favour if anything.

The 6980P beating the 9754 by ~23% is good (geomean, statistic fudging be damned), doing it by consuming ~62% more power isn't. Especially when you consider that the node the 9754 is on is 5nm TSMC not the current gen 3nm TSMC. intel not being able to compete in efficiency using its "3" node is not good, against an older TSMC node to boot. Let's hope the 18A node is as good as they claim, they need it to be good and so do we if competition is to remain anywhere close. No one wants the lack of competition scenario we have with GPU's.

[eszlari]

Why no video encoding benchmarks?

[Michael]

Why no video encoding benchmarks?

There's one if going through to the OB result file page but most video encoders with 4K content don't effectively scale up to 512 threads with one instance encoder. Been working on some multi-instance "density" benchmarks but infrastructure not complete yet.

[pWe00Iri3e7Z9lHOX2Qx]

It's more comparable to the 9754, both at 128 cores with intel having roughly double the cache (instead of comparing to the 9684X with 96 cores with AMD having roughly double the cache). The intel part has AVX512 so also parity, the main differences are intel additionally having AMX, which shows as some clear wins in some benchmarks and should tip the scales slightly in intel's favour if anything.

The 6980P beating the 9754 by ~23% is good (geomean, statistic fudging be damned), doing it by consuming ~62% more power isn't. Especially when you consider that the node the 9754 is on is 5nm TSMC not the current gen 3nm TSMC. intel not being able to compete in efficiency using its "3" node is not good, against an older TSMC node to boot. Let's hope the 18A node is as good as they claim, they need it to be good and so do we if competition is to remain anywhere close. No one wants the lack of competition scenario we have with GPU's.

The node names are mostly marketing bullshit these days anyway, but you are putting to much weight in Intel's "3" being competitive with TSMC's "3". Intel renamed their old node targets to better align with TSMCs. Intel 4 = their old 7nm node. Intel 3 is an enhanced version of that. It isn't in any way efficiency competitive with TSMC's leading edge node.

[jeisom]

It's more comparable to the 9754, both at 128 cores with intel having roughly double the cache (instead of comparing to the 9684X with 96 cores with AMD having roughly double the cache). 's.

Yeah, but you are then comparing AMD's E-cores( Zen 4c) to Intel's P-cores. Neither is a great direct comparison of 1:1. Depending on the test all 3 had good/impressive showings though.

[Jorgp2]

It's more comparable to the 9754, both at 128 cores with intel having roughly double the cache (instead of comparing to the 9684X with 96 cores with AMD having roughly double the cache). The intel part has AVX512 so also parity, the main differences are intel additionally having AMX, which shows as some clear wins in some benchmarks and should tip the scales slightly in intel's favour if anything.

The 6980P beating the 9754 by ~23% is good (geomean, statistic fudging be damned), doing it by consuming ~62% more power isn't. Especially when you consider that the node the 9754 is on is 5nm TSMC not the current gen 3nm TSMC. intel not being able to compete in efficiency using its "3" node is not good, against an older TSMC node to boot. Let's hope the 18A node is as good as they claim, they need it to be good and so do we if competition is to remain anywhere close. No one wants the lack of competition scenario we have with GPU's.

I don't know what you're smoking, but the results show Xeon 6 using equal or less power than their AMD counterparts.

[chris200x9]

It's more comparable to the 9754, both at 128 cores with intel having roughly double the cache (instead of comparing to the 9684X with 96 cores with AMD having roughly double the cache). The intel part has AVX512 so also parity, the main differences are intel additionally having AMX, which shows as some clear wins in some benchmarks and should tip the scales slightly in intel's favour if anything.

The 6980P beating the 9754 by ~23% is good (geomean, statistic fudging be damned), doing it by consuming ~62% more power isn't. Especially when you consider that the node the 9754 is on is 5nm TSMC not the current gen 3nm TSMC. intel not being able to compete in efficiency using its "3" node is not good, against an older TSMC node to boot. Let's hope the 18A node is as good as they claim, they need it to be good and so do we if competition is to remain anywhere close. No one wants the lack of competition scenario we have with GPU's.

I don't know what you're smoking, but the results show Xeon 6 using equal or less power than their AMD counterparts.

I don't know what you both are smoking. Using 62% more power is for sure hyperbole but Xeon 6 using equal or less power is also plain wrong.

[fallingcats]

Those EPYC non-C cores are from 2 years ago, let's wait and see how those parts will compare to what's coming out this fall.

[Gamer1227]

30% more performace than Xeon 5 Emerald Rappids, but also 30% more power consumption. Intel 3 node is showing great efficiency here. /s

Seriously, performace per watt in geomean is similar to EPYC CPUs in TSMC 5nm node, wich is used in production since the Apple M1 in 2020, are we supposed to believe that Intel 3 is competing with TSMC 3nm, or is it just bad CPU design?