Amazon Graviton3 vs. Intel Xeon vs. AMD EPYC Performance

[phoronix]

Phoronix: Amazon Graviton3 vs. Intel Xeon vs. AMD EPYC Performance

Earlier this week AWS announced general availability on their new Arm Neoverse-V1 based processors, Graviton3. Right after that I posted some initial Graviton3 benchmarks against prior-generation Graviton2 for showing the very sizable generational improvement with Amazon's new in-house Arm server processors. Since then I have been carrying out a more robust set of around 100 benchmarks across the original Graviton instance, Graviton2, Graviton3, and then up again Intel Xeon and AMD EPYC competing instances. Here is that much larger collection of Graviton3 performance benchmarks carried out on Ubuntu 22.04 LTS.

https://www.phoronix.com/vr.php?view=31151

[microcode]

Michael you can normalize the "seconds (fewer is better)" charts by cost, by multiplying the time by the cost per time. The scale becomes "dollars (fewer is better)". I think this is a very important type of comparison when it comes to bulk compute like this.

[jjmcwill2003]

Interesting to see how far their Arm architecture has come in terms of performance.

It's too bad there's nothing comparable in the desktop market. (Honeycomb LX seems comparatively ancient and lacks single thread performance).

Maybe one of the upcoming RockChip RK3588 based boards will turn out to be interesting, depending on how quickly Linux support arrives.

[eoerl]

I think that there are really two ways to look at the charts:
- cpu IP wise: clearly the graviton have progressed a lot (not a surprise), but the cores are still not super duper impressive, the comparison points are 8 x86 cores + SMT (half the cores then), and the next gen of both AMD and Intel is just around the corner, yet the x86 cores get a fair number of wins. Would be interesting, IP wise, to get perf/transistor for instance

- price wise, then at least for AWS the Gravitons are clearly a steal. It's really something else though, could well be that it's at a loss for Amazon, that they see it as an investment, or that AMD and Intel are just inherently overpriced. It's not really possible to answer the question with this article

[tambre]

Interesting to see how far their Arm architecture has come in terms of performance.

It's too bad there's nothing comparable in the desktop market. (Honeycomb LX seems comparatively ancient and lacks single thread performance).

Maybe one of the upcoming RockChip RK3588 based boards will turn out to be interesting, depending on how quickly Linux support arrives.

NVIDIA's Jetson Orin might be the best option once they release the cheaper variants in the autumn. Of course they aren't very open source friendly.

[linuxgeex]

ServeTheHome did a narrower comparison but they involved Intel engineers to bring x86 acceleration features into the picture and the Graviton won very few, by small margins, but Intel spanked Graviton by a factor of 10 in several tests and in some the margin was so broad STH didn't even bother publishing it because the results were too embarassing for AWS... "there's no comparison."

So it's interesting to see how Michael's results without acceleration enabled worked out.

I recommend that you read both. Patrick at STH also reminds the reader to consider the total balanced application performance instead of specific benchmarks because even if SSL is 4.3x faster on Intel, SSL is a tiny portion of for example a Wordpress site's workload and disabling that acceleration only lowered performance by 20%. Despite big differences in a few key areas which may be important to a few researchers or businesses selling specific data processing services, overall both platforms remain close when it comes to costs and peak performance.

[smitty3268]

I'm not sure how useful it really is to compare processors with the same threadcount when some have SMT and some don't.

Ultimately I think performance per $ is what primarily matters for the cloud.

But if you were attempting to compare the CPU architectures, then I think getting the same # of physical cores makes more sense, because the extra SMT "cores" are just a side-benefit of the architecture, the same way that the looser memory model is a benefit of the ARM architecture.

[Michael]

ServeTheHome did a narrower comparison but they involved Intel engineers to bring x86 acceleration features into the picture and the Graviton won very few, by small margins, but Intel spanked Graviton by a factor of 10 in several tests and in some the margin was so broad STH didn't even bother publishing it because the results were too embarassing for AWS... "there's no comparison."

So it's interesting to see how Michael's results without acceleration enabled worked out.

I recommend that you read both. Patrick at STH also reminds the reader to consider the total balanced application performance instead of specific benchmarks because even if SSL is 4.3x faster on Intel, SSL is a tiny portion of for example a Wordpress site's workload and disabling that acceleration only lowered performance by 20%. Despite big differences in a few key areas which may be important to a few researchers or businesses selling specific data processing services, overall both platforms remain close when it comes to costs and peak performance.

Checking their site now and not seeing any Graviton3 benchmarks? Maybe it was Graviton2? In any event some of the benchmarks used do have AVX-512 / various Intel optimizations.

[uxmkt]

So if one is using DDR5 and the others aren't, then I'll hold my breath before jumping on the Graviton train.

[coder]

Interesting to see how far their Arm architecture has come in terms of performance.

Yeah, but he's comparing 16-core Graviton instances with 8-core / 16-thread x86 instances.

I'm not complaining, as I think it makes sense to stay within comparable price brackets, but it's something to keep in mind. Also, that Amazon is able to offer time on its own CPUs for less $ for business reasons, as well as technical ones.

It's too bad there's nothing comparable in the desktop market. (Honeycomb LX seems comparatively ancient and lacks single thread performance).

Last year, Mediatek announced intentions to provide a high-performance SoC with Nvidia graphics in a mini-desktop form factor they claim will be suitable for gaming. We can only hope the Linux driver situation will be better than their track record suggests.

Also, I've seen Qualcomm upstreaming driver work, but I'm less hopeful they'll have a desktop offering at a compelling price point.

Maybe one of the upcoming RockChip RK3588 based boards will turn out to be interesting, depending on how quickly Linux support arrives.

That, too.