New AMD GPU Performance To Be Boosted By Linux 4.5; How It Compares To The Binary Blob

[BenPope]

I have no idea why my results are so much better than Michaels. I'll spin a 4.5 DRM-Next in a couple of days and give it another go.

No change: http://openbenchmarking.org/result/1...HA-BPPADOKAN98 (for those paying attention, I undid the patch disabling powerplay by default, so I didn't need the kernel parameter).

[ObiWan]

Did a quick run of the benches with my Fury:

https://openbenchmarking.org/result/...HA-1512234GA60

Not as good as Catalyst, but not as bad as Michaels.

Not sure how comparable the Metro benchmarks are as I don't know the settings of Michaels benches.

[humbug]

I believe someone working on the OSS drivers has stated here on Phoronix that they do support profiles, even if they don't actually make them. If some submits a profile, they'll merge it (not sure whether there's any validation taking place before merging). So yeah, while it's theoretically possible, there's no activity in that area today.

I guess from a purist perspective the idea of application specific profiles is not a nice concept. If every layer is doing it's job properly there should be no need for such things (in theory). The reality is somewhat different and both directx and openGL games rely on nvidia and AMD profiling them for maximum performance. On windows most AAA games are accompanied by driver releases. Hopefully after the catalyst linux driver is discontinued then they can have enough resources to start profiling for open source for the demanding games.

[sonnet]

edited

[bug77]

Yep ideally surpassing catalyst performance should only be the start (considering what the hardware is capable of). But it's still a way to go to match catalyst and some people here actually say it cannot be done without application specific game profiles (not sure if/when these are coming to open source?).

I believe someone working on the OSS drivers has stated here on Phoronix that they do support profiles, even if they don't actually make them. If some submits a profile, they'll merge it (not sure whether there's any validation taking place before merging). So yeah, while it's theoretically possible, there's no activity in that area today.

[djdoo]

It is the first time in all of these years I am aboard linux train that I would be really glad if Linus dropped Kernel 4.4 before released and went straight to a 4.5 rc1 version! My Tonga will rev up and I will be able to play games without that Bloody mess called fglrx renamed to catalyst renamed to crimson radeon driver!!
I guess it is easier to change names than hunt bugs in the code and make optimizations so after kernel 4.5 I won't care anymore how they will call that disastrous blob in the future.

And yes the results are very promising, if someone remembered from the old days of r600g when DPM introduced there where many blocker bugs and performance was not at the levels of 50% or more of catalyst and also we have to mention that 3D performance has a lot improved in catalyst too since then, considering that we talk about completely experimental code I am very optimist about AMDGPUs future!
Fiji seems to have problems not only with amdgpu but with catalyst too... Really a pity for such an expensive hardware supporting cutting edge technologies like HBM to perform worse than my Tonga (even with blob sometimes) which costs much less than half the money Fury is...
Good work Michael! Good work AMD OSS team! Thanks guys, happy New Year, health and labour to all of you and to the phoronix forum mates!
Jim, Greece

[humbug]

Not too bad, but... Don't we need this driver to actually surpass catalyst so that it can amount to anything? (Considering how catalyst is lagging pretty far behind Nvidia's drivers)

Yep ideally surpassing catalyst performance should only be the start (considering what the hardware is capable of). But it's still a way to go to match catalyst and some people here actually say it cannot be done without application specific game profiles (not sure if/when these are coming to open source?).

Still we do really appreciate the work that AMD's linux driver team is doing with limited resources. Hopefully with Vulkan the driver teams can stay ahead of the game from the start, without later having to play catch-up to API advancements. Maybe that way there can be more focus on performance?

[blackiwid]

There's work going into the Xserver but it'll need to be sorted at the DRM level and ddx too - I think radeon & amdgpu already support fencing and I'm not sure yet if it'll work when using the modesetting driver - I'll comment once things have been merged upstream

It seems that would be a good thing today in the near feature (6 to 12 months) the propriatary crashing blob will most likely only be needed for some enterprise software or some cracy stuff most consumers dont care and for gaming the free driver will be faster or at least the same speed. Or to enable on 5000 Dollar normal gaming cards that are hardware identical to 500 dollar cards that gpu computing features are allowed to run. Like antifeatures for rich companies to make them pay 10 times the amount of the cheaper gamers.

Or something like that.

[darkblu]

Hey, that was unexpectedly good! Thanks for the christmas surprise, Michael : )

[boxie]

I am looking forward to the day when I can easily switch between OSS and Proprietary drivers - or even using them on a per application basis (i.e. OSS running the desktop and the binary blobs being used for gaming).