AMD EPYC 7502 + EPYC 7742 Linux Performance Benchmarks

[mroche]

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.

If it’s in there, I must have missed it. Looking forward to those (and the PostgreSQL) benchmarks!

Cheers,
Mike

[MaxToTheMax]

So except for certain lightly-threaded and database workloads, this looks like a blowout. AMD looks to have the space to itself, at least until Ice Lake-SP in 10 months. If I were AMD, I would wait like 7 months, and then start trickling out details about Zen 3 in order to get people to wait for that release instead of buying ICL-SP.

If ICL-SP is actually going to be competitive with Rome, we should expect to start seeing benchmark "leaks" soon. Silence is a red flag here.

How about PCIe 4? Infinity Fabric still consumes half of the lanes when in 2P configuration?

According to AnandTech's review, this is up to the motherboard manufacturer. By default it still uses half but you could make a motherboard that allocates fewer lanes than that to IF.

So the EPYC 7742 can't idle?

Maybe the boost algorithm doesn't idle the CPU until after a certain amount of time. If it really can't idle, then that would be a serious problem. It would definitely be worth measuring idle power consumption against Xeon, as that is something a server customer would care a lot about.

[Zola]

Seams that EPYC 7702P will be a absolute workstation killer deal. Hopefully AMD won't screw up with motherboard's & generally availability this time around.

[Adarion]

Michael
OMG, thanks for the tests. What a beast! And still interesting to see the differences from benchmark to benchmark.
But you forgot the most important part: Make a screenshot of the bootup-logo penguins on a 2 processor-machine with SMT enabled. I want to see those 256 penguins! Do they still fit on the screen? Will it have to scroll for them? Or will it end in a kernel panic?

[carewolf]

Never in Computer History Intel lost this big.

Well, once 15-20 years ago, P4 vs Athlon(64).

[Michael]

Michael
OMG, thanks for the tests. What a beast! And still interesting to see the differences from benchmark to benchmark.
But you forgot the most important part: Make a screenshot of the bootup-logo penguins on a 2 processor-machine with SMT enabled. I want to see those 256 penguins! Do they still fit on the screen? Will it have to scroll for them? Or will it end in a kernel panic?

It doesn't fit on the screen, even on 32c systems I don't believe it ends up looking anything special as it gets chopped off.

[devius]

Well, once 15-20 years ago, P4 vs Athlon(64).

Exactly. The difference was even bigger back then, with the Athlon 64 performing about 60% faster than a similarly clocked Pentium 4.

[elatllat]

So comparing to https://www.phoronix.com/scan.php?pa...-spectre&num=1

• AMD Zen/Zen+
  • l1tf: Not affected
  • mds: Not affected
  • meltdown: Not affected
  • spec_store_bypass: Mitigation of SSB disabled via prctl and seccomp
  • spectre_v1: Mitigation of __user pointer sanitization
  • spectre_v2: Mitigation of Full AMD retpoline IBPB: conditional STIBP: disabled RSB filling.
• EPYC 7742
  • l1tf: Not affected
  • mds: Not affected
  • meltdown: Not affected
  • spec_store_bypass: Mitigation of SSB disabled via prctl and seccomp
  • spectre_v1: Mitigation of __user pointer sanitization
  • spectre_v2: Mitigation of Full AMD retpoline IBPB: conditional IBRS_FW STIBP: conditional RSB filling-

I was expecting more.

[brent]

EPYC 2nd generation looks like a killer. What remains a problem though is the lack of EPYC-based products, that is ready to buy servers and more importantly, workstations. I hope we'll see some improvements here.

[schmidtbag]

If I were interested in getting a server, the 7502 looks very enticing. Packs so much performance for the money.
It's weird how AMD even made their own 7551 look slow, and that thing is no slouch.

Also interesting how a single 7742 is roughly as good as 2x 8280s while being less than half the price.