Some AMD Grenada Cards Have Been Borked On The Open-Source Driver For 2 Years

[phoronix]

Phoronix: Some AMD Grenada Cards Have Been Borked On The Open-Source Driver For 2 Years

A frustrated Phoronix reader pointed out a bug report that's been open nearly two years regarding AMD Grenada (basically, Hawaii cards in the R9 300 series) support on the open-source Linux driver being in a tough position for a subset of users...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-Grenada-Linux-2

[randomsalad]

Sitting on an AMD Radeon R9 390, the ASUS Strix OC variant.

I can from personal experience say that the card works mostly OK with the amdgpu kernel driver, only problem is broken DPM. This can be resolved by force-disabling DPM by doing

# echo "high" > /sys/class/drm/card0/device/power_dpm_force_performance_level

, but obviously that means no power savings when the card is idle. Otherwise, the card would continuously cause very obvious visual tearing when trying to do anything except stare at a blank desktop or when running a game demanding enough to keep the card in the high-performance state.

Other than that, it's been working very well for me on both amdgpu+mesa and amdgpu-pro.

[Dreakon]

I had issues with my Gigabyte G1 R9 390X up until kernel 4.9 or so. Currently running on 4.11 with the AMDGPU driver and it's been working pretty well.

[yoshi314]

This is completely in contrast to my experience with opensource amd driver. I did not have cards from that particular gen, but older units and i just had to file bug reports and add extra info when i had issues. Even most serious stability problems were resolved in a matter of few days max.

I honestly find this surprising.

[Guest]

This is completely in contrast to my experience with opensource amd driver. I did not have cards from that particular gen, but older units and i just had to file bug reports and add extra info when i had issues. Even most serious stability problems were resolved in a matter of few days max.

I honestly find this surprising.

It seems Phoronix launched some anti AMD strategy last days. It is known for propaganda articles.

[starshipeleven]

It seems Phoronix launched some anti AMD strategy last days. It is known for propaganda articles.

I disagree on the "last days" part.

[bug77]

This is completely in contrast to my experience with opensource amd driver. I did not have cards from that particular gen, but older units and i just had to file bug reports and add extra info when i had issues. Even most serious stability problems were resolved in a matter of few days max.

I honestly find this surprising.

Michael had a 290 acting up which is basically the same card. The issue was never widespread, which made it difficult to pinpoint and fix (I'm not sure if it was ever pinpointed or just fixed as part of larger changes).
So yes, I expect "but this works on my machine" experience to be much more common. Yet, there's a chance some quirk of this silicon just likes to rear its head now and then.

[dungeon]

This is completely in contrast to my experience with opensource amd driver. I did not have cards from that particular gen, but older units and i just had to file bug reports and add extra info when i had issues. Even most serious stability problems were resolved in a matter of few days max.

I honestly find this surprising.

You should say "usually" instead of "max", as it is proper to say 9+ years max

9 Year Old Mesa Bug Fixed, Addresses Stability With RadeonSI Gallium3D

[DarkFoss]

I wonder how many of these types of issues are caused by a faulty bios ? I'm sure some of them gets by on windows but get exposed under linux., especially once the vendors have added their own custom cooling. Just looking at the VGA Bios Collection at Techpowerup there are 88 bios's for the R290, 100 for the 290X. In both of those listed only 3 in each group are from ATI.

[JPFSanders]

Michael had a 290 acting up which is basically the same card. The issue was never widespread, which made it difficult to pinpoint and fix (I'm not sure if it was ever pinpointed or just fixed as part of larger changes).
So yes, I expect "but this works on my machine" experience to be much more common. Yet, there's a chance some quirk of this silicon just likes to rear its head now and then.

I have another 290 and always have had issues.

Any problem on Linux tends to be exacerbated by the huge disconnection between components, the performance/crashes problems with the 290 are /were both caused by the kernel and mesa.

Mesa has had some lockups these tend to leave the screen frozen, these are rare now but used to happen with certain frequency a year or so ago, the DPM/DRI part of the kernel is the worse one as it is the one that takes the longest to fix, crashes there black the screen and leave a piece of audio playing continuously.

The DRI/DPM issues can be solved by A:

a.1) Plugging a second monitor (makes the memory clock go high)
a.2) Setting the DPM to the low profile

This gives moderate performance and works very well with a 270

Or B:

b.1) Setting the DPM to the low profile

This gives low performance; 25fps aprox regardless of resolution, 800x600 = 25fps, 1920x1080 = 25fps, antialiasing/no antialiasing = 25fps

This used to work wonders for the 290, but as pointed by Michael the behaviour of the 290 + Radeon changes from kernel version to version.

The most stable option is B, I have spent hours playing on the low profile and the card is stable.

I blame the Kernel side more than anything (Mesa has done a stellar job patching things), the quality of the drivers tends to be spotty, if your hardware is popular and the developers have it tends to be well supported, if not and you're unlucky to have something like the 290 which doesn't appear to be homogeneous (not all cards seem to be the same) you're in for a ride.

Having said all of that, I'm not complaining, I'm grateful to have the entire graphics stack ecosystem opensourced, I'm really grateful to AMD for publishing the card details, and that is why I bought a 480 to replace the 290 instead of an nvidia.

My only gripe is the lack of tweak-ability of the cards in Linux, but this is something I know will improve in the future.