The AMD Radeon R9 Fury Is Currently A Disaster On Linux

[Michael]

I really appreciate all the time and money spent in writing these articles. But the title gives the impression that Fury fares much worse than other AMD hardware. After seeing the numbers, I don't think so. It is just the same issues with Catalyst that is reflected here. Given the state of AMD's drivers, the numbers are more or less as I expected, and I didn't expect to see miracles with Fury either!

It's a disaster in the sense that a GTX 970 that costs easily $200 less near universally outperforms the R9 Fury except for in the case of Valley as the only test so far where that doesn't hold.

[Michael]

How is this a disaster? It's an improvement over the R9 290 and performs at or near the top on a bunch of the tests. Easy on the hyperbole.

That for pretty much all (but Unigine Valley and OpenCL) that the R9 Fury was running around the speed of a GTX 970 that costs $200 less?

[Ansla]

That is true for all AMD cards running Catalyst when compared to a nVidia card with their proprietary driver. The title suggests there is something particularly wrong with the card itself (like it runs slower then a $200 cheeper AMD card). A more realistic title would have been "The AMD R9 Fury works as expected on Linux. Catalyst is a disaster"

[coder111]

I don't see this as a disaster. I mean, AMD Catalyst performance always lagged behind NVidia BLOB performance. This has been the case for as far as I remember. So nothing unexpected.

Fury is still the fastest AMD card out there. If you just want pure framerate and you are OK running BLOB drivers, go buy NVidia. I know I won't.

AMD should soon have half-decent open-source drivers, and if you compare open-source performance between AMD and anyting else, anything else gets blown out of the water.

But I agree with one thing- I'm not rushing to buy it just yet. I'll see how the open-source drivers pan out, and the prices will fall by then as well.

[dungeon]

Ansla

Nope, Catalyst is not disaster too... this is War of Blob's Variabiles and Profiles on Linux

What you see here is that those games Michael choosen mostly has nvidia's specific variabile in script which improve performance... against non profiled games on fglrx

[Michael]

Nope, Catalyst is not disaster too... this is War of Variabiles and Profiles on Linux

What you see here is that those games Michael choosen mostly has nvidia's specific variabile in script which improve performance... against non profiled games on fglrx

I didn't chose any games at all based on NVIDIA-anything... These are the Linux tests I always run.

If you have any other recommendations, feel free to pass me the phoronix-test-suite benchmark xxx command to run and I'll be happy to do so on the Fury system. The only other test recommendation you had was to run OpenArena and as I said I tried that but it wouldn't even run at 4K..

[smitty3268]

Yeah, so it basically performs exactly like everyone figured it would. Just like the scaled up 285 gpu it is.

dungeon, testing OpenArena on a card like this is completely useless. Why not just ask for glxgears benchmarks while you're at it, to get something "fair"? Michael tested real games that people actually want to play and which don't cause the results to go off into multiple hundreds of fps where testing is useless. These were about as good as we're going to get, failing things like getting some WINE gaming tests.

Michael - i would like to see what happens to say, bioshock, if you run it using that eon parameter to turn off the fglrx tweaks. Does it significantly speed up? Could be another article in general, not necessarily related to the Fury card. Also, claiming you thought about returning it because the card was slow is ridiculous. You got it for phoronix testing, not personal use, and the testing went just fine. It's not like you couldn't run the tests with it.

[dungeon]

dungeon, testing OpenArena on a card like this is completely useless.

It is not useless, particulary on higher resolutions... it is one (maybe only one in phoronix suite?) that clearly shows memory bandwidth difference And because it does that "up to bottom" for every chip tested

Why i want it? Well card has HBM isn't it

[Nobu]

70-75% of the performance of TITAN X (in games where the framerate never drops below 80fps) with a hell of a lot less memory and an unoptimized driver seems pretty decent to me. Only reason I wouldn't buy it is because I can't justify spending that much on just a GPU.

[pigeon768]

Now that we have several GCN 1.2 cards on the market, I think it's time you bite the bullet and get the AMDGPU driver up and running. Right now the only viable choice for video cards on linux is an nvidia card with closed source drivers, but I'm really curious to see if the amdgpu work could be a viable alternative.