AMD Resolves Linux Marginality Performance Issue on Newly Shipped Ryzen CPUs

[Raevenlord]

We've brought to your attention how AMD's Ryzen architecture was encountering a bug that prevented it from successfully performing certain tasks under Linux. The issue, to which Phoronix was the first website to call attention to, was later confirmed by AMD, with a further remark from the chip designer that EPYC and ThreadRipper weren't affected by it.

AMD has now solved the Ryzen issue at a silicon level, and new revisions of the CPUs should be clear of the problems. Phoronix's Michael Larabel has confirmed that the new processor passes the previous insurmountable, crash-prone workloads with flying colors, so all in all, AMD delivered a swift response to the issue. The company is also doing good for itself in that it is open to replacing previous-batch CPUs that are affected by the issue, through a contact to its Customer Care department.



View at TechPowerUp Main Site

 

[mouacyk]

Is the "Source" supposed to point to an article or press release by AMD, because it's currently not?

I've waited for 6 months to build a Linux server... how do I buy a segfault-free CPU initially? I'd rather avoid the several hours of testing needed to attain RMA and retest.

 

[Raevenlord]

Is the "Source" supposed to point to an article or press release by AMD, because it's currently not?

I've waited for 6 months to build a Linux server... how do I buy a segfault-free CPU initially? I'd rather avoid the several hours of testing needed to attain RMA and retest.

Updated source link, thanks for bringing it up.

Edit: Not sure, but the only way should be for you to physically check CPU batches to see.

 

[Steevo]

There is no guarantee offered by Intel or AMD on such things, and it's exactly why they specify they have it fixed.

It's also worth noting that both Intel and AMD list errratus on their website for every processor.

 

[PowerPC]

Wow, is AMD actually offering to replace all previously sold Ryzen CPUs? That is insane but I have to commend them for doing it.

 

[chaosmassive]

Wow, is AMD actually offering to replace all previously sold Ryzen CPUs? That is insane but I have to commend them for doing it.

I think only few people use Linux and even fewer who do that kind of computation on Linux
for majority user who using Windows mostly for gaming/streaming/rendering/CAD etc
even they knew that their CPU has such issue, they dont bother to go through all the RMA hassle and move on.

 

[Joss]

how do I buy a segfault-free CPU initially?

Chips affected are those manufactured prior to week 25 (of 2017). It's printed on the CPU itself, not sure about the box it comes on though.
I suppose one can consult with the vendor, they should know what they have in stock.

http://www.phoronix.com/scan.php?page=article&item=new-ryzen-fixed&num=1

 

[OSdevr]

Wow, is AMD actually offering to replace all previously sold Ryzen CPUs? That is insane but I have to commend them for doing it.

Intel had to do that once, cost them nearly half a billion dollars! This is part of the reason why microcode on today's processors can be updated.

@Raevenlord:
Is this the same FMA3 bug that AMD fixed with microcode or is it a different one?

EDIT: Turns out it's a different bug that apparently can't be fixed in microcode. Yet according to that article's comment section the fixed CPUs still have the same stepping

 

[Hood]

Soon there should be Ryzen refurbs for cheap - RMAed 1st gen chips whose only fault is that they still have the Linux bug, which will be fine for most people. But not many, most Ryzen owners will just keep their 1st gen.

 

[biffzinker]

Past thread in the Linux/BSD/Mac OS X section I posted.
New Ryzen revision corrects compiler segmentation fault issue

 

[R-T-B]

Past thread in the Linux/BSD/Mac OS X section I posted.
New Ryzen revision corrects compiler segmentation fault issue

Consider it promoted to news.

As I said in that thread, 1.2v SOC fixed it for me.

 

[Dave65]

Not giving up my 1700 @ 4.1 for a chip that may only get 3.8..

 

[ssdpro]

MD has now solved the Ryzen issue at a silicon level, and new revisions of the CPUs should be clear of the problems.

Two things of interest there.

 

[eidairaman1]

Im presuming they will launch microcode updates for existing parts or replace them?

This sounds like a stepping revision

 

[OSdevr]

Im presuming they will launch microcode updates for existing parts or replace them?

This sounds like a stepping revision

Yes they are replacing processors, so it can't be fixed with microcode.

 

[R0H1T]

Two things of interest there.

The thing with this particular issue was that it wasn;t reproducible, as in no one knew what triggered the segfaults. Not AMD, not any of the thousands who posted on AMD forum could pin it down to one thing. As is the case with such an errata, you test the hardware (fix) & see if the segfaults recur. They;ve probably done days or weeks of testing on the new batch & cleared it. You can;t speculate that it isn't fixed till you get the same segfaults with the latest stepping/revision of Ryzen.