AMD EPYC 7502 + EPYC 7742 Linux Performance Benchmarks

[phoronix]

Phoronix: AMD EPYC 7502 + EPYC 7742 Linux Performance Benchmarks

Now that you have read our AMD EPYC "Rome" 7002 series overview, here is a look at the initial performance benchmarks from our testing over the past few weeks. This testing focused on the new AMD EPYC 7502 and EPYC 7742 processors in both single (1P) and dual (2P) socket configurations using AMD's Daytona server reference platform. Tests were done on Ubuntu Linux and compared to previous AMD EPYC processors as well as Intel Xeon Scalable.

http://www.phoronix.com/vr.php?view=28142

[tildearrow]

Yes, I do have a test request: BLAKE2 between all aforebenchmarked processors.

[Michael]

Yes, I do have a test request: BLAKE2 between all aforebenchmarked processors.

Actually had an issue with blake2 IIRC over either the GCC version or the compiler flags.

[tildearrow]

- dav1d results in FPS -

1080p

Xeon Platinum 8280	301.76
2 x Xeon Gold 6138	303.54
EPYC 7551	324.01
EPYC 7601	344.46
2 x Xeon Platinum 8280	393.34
EPYC 7502	484.93
2 x EPYC 7601	553.46
2 x EPYC 7502	629.9
EPYC 7742	655.09
2 x EPYC 7742	750.63

4K

EPYC 7551	131.54
2 x Xeon Gold 6138	143.66
EPYC 7601	144.87
Xeon Platinum 8280	167.04
2 x EPYC 7601	190.33
2 x Xeon Platinum 8280	197.53
EPYC 7502	225.75
2 x EPYC 7502	273.78
EPYC 7742	299.5
2 x EPYC 7742	322.56

4K Frames Per Dollar

2 x Xeon Platinum 8280	0.009867772
Xeon Platinum 8280	0.016688745
2 x EPYC 7601	0.021628728
2 x EPYC 7742	0.023205786
2 x Xeon Gold 6138	0.026643299
EPYC 7601	0.032925759
EPYC 7551	0.036358316
EPYC 7742	0.043093447
2 x EPYC 7502	0.052650807
EPYC 7502	0.086827646

[tildearrow]

Typos:

Using a Linux x86_64 kerneel default configuration

But it's interesting to see here the roughly 15% better Python performance generation-over-generation for EPYC.\

[dungeon]

0.5 TB of RAM... Gnome and browser tabs could leak freely there

[mroche]

Any chance some (or all) of these tests could be rerun with GCC 9.x that supports the new Zen 2 target using -march=native so we can see if there are any further performance improvements to be had to either platform?

Cheers,
Mike

[Michael]

Any chance some (or all) of these tests could be rerun with GCC 9.x that supports the new Zen 2 target using -march=native so we can see if there are any further performance improvements to be had to either platform?

Cheers,
Mike

I thought I wrote in the article, but I do have some GCC (including latest Git) vs. Clang vs. AOCC benchmarks coming out in next few days.

[chithanh]

Thanks for these tests.

The MySQL/MariaDB performance is another scenario where the EPYC Rome performance didn't beat out the Xeon Scalable Cascadelake performance, but AMD is investigating.

That MariaDB result is interesting. Did you run pgbench too with same/different results?

[Michael]

Thanks for these tests.


That MariaDB result is interesting. Did you run pgbench too with same/different results?

Will check on PostgreSQL, forgot why I hadn't included it in my original testing, been a terribly long past few weeks.