AMD Confirms Linux Performance Marginality Problem Affecting Some, Doesn't Affect Epyc / TR

[calc]

Could you explain more about what is meant by "performance marginality?" It seems open to several different meanings and I'm not sure what you or AMD means by it.

What that means is "we still don't respect you and think you should buy Intel instead"

Sounds like they are saying that at a very high cpu usage level the system stability becomes marginal (bad). Combined with the fact that many people buy AMD specifically to overclock, since Intel tends to suck at that, they probably originally just assumed the small number of people having the problem were overclockers. It sounds like they are implying that Windows isn't able to reach usage levels that high for some reason so isn't affected, maybe a high performance threading issue under Windows keeps it from happening.

https://www.merriam-webster.com/dictionary/marginal

4.(a) close to the lower limit of qualification, acceptability, or function : barely exceeding the minimum requirements

[schmidtbag]

Yes they have now said I will be receiving Threadripper and Epyc samples (contrary to previous communications saying not) and also looks like some more Ryzen models too.

That's great. These aren't cheap models (though compared to Intel's offerings, they're a bargain). It wouldn't surprise me if you don't get one of those real fancy kits that other reviewers got, but hey, at least they're interested in supplying you and working with you - that's a big step forward.

[shmerl]

I didn't quite get it. Will there be some microcode fix, or RMA to newer stepping is the only way to avoid it?

[Michael]

I didn't quite get it. Will there be some microcode fix, or RMA to newer stepping is the only way to avoid it?

So far they haven't communicated it could be fixed via microcode update.

[Holograph]

It's really rare that I log in and comment on the Phoronix Forums, so... congratulations, you went so far off the deep end that I logged in. If you haven't seen this video, then you really need to see it. (https://www.youtube.com/watch?v=osSMJRyxG0k)

AMD has been tracking this issue since day one, and they've been testing on it since day one, they've just been very quiet and secretive, which is exactly how Intel handled their huuuge bug with hyperthreading recently. It actually took Intel over a year to release a fix for their issue, and they never acknowledged it during that entire time. If you think Intel is going to provide better support, you're wrong. AMD should have been open from the beginning, I agree! But, the fact that they're being open now is instantly better than how Intel has ever handled any situation similar to this, and that's if we just conveniently ignore how bad Intel is at sportsmanship (which the video talks about).

So yes, I will call you a troll if you keep making these ridiculous statements. It's possible that up until now you did not realize how much more Intel has failed at this than AMD has, and I don't know who you would suggest buying from if you disavow both Intel and AMD. Are you going to buy VIA processors? Because that would be a joke of a computer.

I'm neither off the deep end, nor a troll, just because you disagree with me. Congratulations: you made yourself look foolish!

I already called Intel corrupt and I've been in the game long enough to know. I was fortunately warned about FDIV back in the day and held off on the Pentium until I got a 166MHz one. I switched to AMD with the K7 and stayed with them until Core 2 Duo.

The thing is, we have to react to poor actions from corporations as they come. I'll actually claim at times I'll boycott Intel - I end up switching back and forth sometimes.

Punishing the most recent transgression of Intel/AMD (or Nvidia/AMD) is still a better corrective action than buying a product immediately after a company screws you.

And multiple AMD employees speculated that Ryzen was simply not compatible with the standard architecture and mused that they might tell people that they needed to recompile software for Ryzen. Again, I point out this was not a final solution from them. But that they even considered it, that they talked about the possibility, is extremely insulting and shows that I literally can't count on AMD to care that their processors follow the standards of the architecture.

So now I would like you to justify the use of the word "troll" and also tell us how it helps improve communication in this discussion. Some of you seem to think you're actually helping the discussion by throwing this word around.

[shmerl]

So far they haven't communicated it could be fixed via microcode update.

Did you also ask them about mce bug (random freeze / reboots)? Those are actually more annoying than segfaults.

[k1e0x]

I didn't quite get it. Will there be some microcode fix, or RMA to newer stepping is the only way to avoid it?

Either that or OS patches prob if AMD can't fix it. It's unsure if it's an architecture thing or not. You won't get an RMA I'd think because its not exactly widely prevalent. Stay tuned. Like I said before, FreeBSD already appears to have a partial fix for this by moving some of the areas of ram around.

[bridgman]

There were a lot of suggestions going on - including from AMD and I believe specifically including from bridgman - that Ryzen might simply not be 100% compatible, and the fix might have been to compile software for the CPU.

And multiple AMD employees speculated that Ryzen was simply not compatible with the standard architecture and mused that they might tell people that they needed to recompile software for Ryzen. Again, I point out this was not a final solution from them. But that they even considered it, that they talked about the possibility, is extremely insulting and shows that I literally can't count on AMD to care that their processors follow the standards of the architecture.

I don't think anyone from AMD has ever said anything like that. There were questions about compiler settings in the context of understanding what instruction sequences might be related to the reported problems, but I don't think anyone has even hinted that the solution would be to recompile all the software you were running on it. Not everyone runs Gentoo.

[shmerl]

Either that or OS patches prob if AMD can't fix it. It's unsure if it's an architecture thing or not. You won't get an RMA I'd think because its not exactly widely prevalent. Stay tuned. Like I said before, FreeBSD already appears to have a partial fix for this by moving some of the areas of ram around.

Several people already reported that AMD are ready to provide an RMA and even test a setup similar to one they need to replace, but it's not guaranteed to help (it did for some, and it didn't for others).

[chithanh]

Interesting that AMD says Windows is not affected. There is at least one user who reproduced this problem on WSL.

What that means is "we still don't respect you and think you should buy Intel instead"

Coffee Lake here I come

I don't need to remind you about the Skylake HT bug which was reported to Intel in early 2017, and then months of no communication followed? Intel silently fixed the bug in microcode some months later, there was no public statement of any sort directed at the affected users.