AMD Has Some Last Minute Updates For The AMDGPU Driver In Linux 5.11

[phoronix]

Phoronix: AMD Has Some Last Minute Updates For The AMDGPU Driver In Linux 5.11

The Linux 5.11 merge window is expected to open next week and while AMD has already queued several rounds of updates into DRM-Next ahead of that period, some last minute items were submitted overnight for this next Linux kernel version and what will be the first major kernel release of 2021...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-Linux-5.11-Last-Minute

[StandaSK]

Nice to see the older GCN GPUs getting some love

[Peter Fodrek]

Nice to see the older GCN GPUs getting some love

GCN =Graphics Core OF Next GENERATION is every new GPU from AMD even if it is called RDNA2 ))))

[duby229]

GCN =Graphics Core OF Next GENERATION is every new GPU from AMD even if it is called RDNA2 ))))

That's not true, GCN is a codename given to specific -set- of AMD GPU cores which all have derivatives of the same architecture. The HD7970 was the first high end GCN product and the RX Fury was the last high end GCN product.

RDNA is different -set- of AMD GPU cores with a new architecture.

https://en.wikipedia.org/wiki/Graphics_Core_Next

https://en.wikipedia.org/wiki/RDNA_(microarchitecture)

[bridgman]

GCN =Graphics Core OF Next GENERATION is every new GPU from AMD even if it is called RDNA2 ))))

Yeah, in hindsight using the word "next" in an architecture name can get confusing after you release it.

We never used the GCN terminology internally, just generation names like GFX6 (SI), GFX7 (CI), GFX8 (VI & Polaris), GFX9 (Vega & MI-100), GFX10 (RDNA). Within generations we subdivide further, eg RDNA1 is GFX 10.1, RDNA2 is GFX 10.3.

That's not true, GCN is a codename given to specific -set- of AMD GPU cores which all have derivatives of the same architecture. The HD7970 was the first high end GCN product and the RX Fury was the last high end GCN product.

I think Peter Fodrek was making a joke since the "Next" part of GCN can apply to the next part we produce from now until the end of time.

Rather than Fury, wouldn't Vega20 be the last high end GCN product before GCN morphed into CDNA ?

[muncrief]

I sure wish they'd fix the problems with resume on my RX 580.

I used to have the same problem with my old Gigabyte 990FX motherboard, FX 6300 CPU, and R9 390 GPU system, but was always told it was because my hardware was old.

But now that I built an entirely new system with an ASUS TUF Gaming X570 motherboard, R7 3700X CPU, and RX 580 GPU the problem persists, and my (209457) and other bug reports about it remain ignored after 2 months.

[bridgman]

But now that I built an entirely new system with an ASUS TUF Gaming X570 motherboard, R7 3700X CPU, and RX 580 GPU the problem persists, and my (209457) and other bug reports about it remain ignored after 2 months.

Where did you file the bug report ? Doesn't seem to be a valid bugs.freedesktop.org # and tickets on the gitlab issue tracker that replaced it seem to be 4-digit right now.

I did find a kernel.org ticket but that can't be the one because agd5f has already responded to that one three times in the last 2 months, and last comment was that there seems to be a platform level problem that is leaving the board powered off so neither OS nor driver can access it.

His previous post asked if the problem appeared when you were not using VFIO - only response I saw was that "you could not remove your VFIO setup for a long time" which is fine, question is whether you could remove it long enough to do some testing. Removing the second GPU and booting from a live CD might be sufficient, not sure.

[Guest]

I sure wish they'd fix the problems with resume on my RX 580.

I used to have the same problem with my old Gigabyte 990FX motherboard, FX 6300 CPU, and R9 390 GPU system, but was always told it was because my hardware was old.

But now that I built an entirely new system with an ASUS TUF Gaming X570 motherboard, R7 3700X CPU, and RX 580 GPU the problem persists, and my (209457) and other bug reports about it remain ignored after 2 months.

Haven't faced problems on resume with my 580, but haven't used it for too long either. Looks like it's not a general problem with all 580s, might be some issue specific to your GPU model (or some race condition that I haven't experienced yet).

Yeah, I faced a lot of problems on my R9 390 too. I have found that the best way to get something fixed is to do the work yourself. It sucks, but that's just how it is with AMD.

[bridgman]

Yeah, I faced a lot of problems on my R9 390 too. I have found that the best way to get something fixed is to do the work yourself. It sucks, but that's just how it is with AMD.

The 290/390 is a bit of a special case - I don't think we ever figured out why we were unable to repro many of the power-related problems in house.

Quite often even if we can't repro in house agd5f is able to figure out what is going wrong by sending out a series of "try this and tell me what happens" patches but even that wasn't totally successful for 290/390.

[muncrief]

The 290/390 is a bit of a special case - I don't think we ever figured out why we were unable to repro many of the power-related problems in house.

Quite often even if we can't repro in house agd5f is able to figure out what is going wrong by sending out a series of "try this and tell me what happens" patches but even that wasn't totally successful for 290/390.

Do you have any suggestions I could try for my 580 bridgman? All my hardware is new, but every 5 or 6 times resume hangs as described at https://bugzilla.kernel.org/show_bug.cgi?id=209457. It's really frustrating, but it's so random I'm at a loss at what to try or how to debug it.

To compound the problem I actually have two RX 580s, with one used solely for a Windows VM, and the VM GPU is in the primary PCIE slot. So I've always wondered if my non-standard configuration is confusing the drivers somehow. On the other hand it's odd that it works most of the time, and that the same problem occurred with a GT 710 and R9 390 in my old motherboard.

But as you know confusion is my normal psychological state