The AMD Radeon R9 Fury Is Currently A Disaster On Linux

[duby229]

You are wrong again, because openarena has fglrx profile so that should not be bottlenecked ... maybe you are just not informed on that

No. Absolutely not.

It's fine if you actually play openarena, then in that case those bechmarks will matter to you. But, you can -not- benchmark openarena and claim that those results represent anything else. It's fine if openarena doesn't suffer from whatever bottleneck we're seeing, but other games obviously are.

[dungeon]

Openarena phoronix benchmark shows memory bandwidth difference, you can check it on anything nVidia, Intel and AMD APU, on cards... it is clear bandwidth capped benchmark.

I like to take a look on Hawaii vs Fury difference in openarena, that is all.

It's fine if openarena doesn't suffer from whatever bottleneck we're seeing, but other games obviously are.

It is not only openarena that has profile, Unigine benchmarks has it, HL2, TF2, etc...

[theghost]

Thank you for your benchmarking efforts, Michael.

I think you have done nothing wrong here. Those are the games which are actually played under Linux, yes even with an not optimized Catalyst.
It's clearly AMD's fault who don't give a shit on the profiles for Linux and it's clearly a disaster comparing the pricing.
Who cares if the card could perform much better, maybe, theoretically, under certain conditions.
Under Linux today we don't have these conditions and not everyone is interested waiting another 5 years until a new driver might be in shape.
If you want to play games today under Linux, this card is a disaster for this price, so what.
I can say that, even though I am running an AMD card (suprise, on Catalyst, because the other driver is not in shape yet).

[Michael]

Openarena phoronix benchmark shows memory bandwidth difference, you can check it on anything nVidia, Intel and AMD APU, on cards... it is clear bandwidth capped benchmark.

I like to take a look on Hawaii vs Fury difference in openarena, that is all.

Here's some OpenArena R9 Fury tests (sans the 4K ones as the game hangs at 4K at least for catalyst)

http://openbenchmarking.org/result/1507291-BE-OPENARENA42

If you so desire you can dig through some old Phoronix articles since I commonly use this i7 5960X system for my graphics benchmarks so can compare to those old numbers.

[dungeon]

I dig some even with slower CPU, etc...

http://openbenchmarking.org/prospect...0169ca1ac111b5
http://openbenchmarking.org/prospect...edf1fa68b1445d

From the time that there was not profile, etc... so no, it can't be true that Hawaii is faster then Fury really

[duby229]

I dig some even with slower CPU, etc:

http://openbenchmarking.org/prospect...0169ca1ac111b5
http://openbenchmarking.org/prospect...edf1fa68b1445d

From the time that there was not profile, etc... so no it can't be true that Hawaii is faster then Fury really

Nobody ever said that. The benchmarks that were run today on Fury and benchmarks that were run in the recent past on the 295 Hawaii obviously show a bottleneck somewhere, I don't think anybody knows what that bottleneck is yet, but it's obviously there. The 285 Tonga results look pretty badly bottlenecked too.

As Micheal said in the article, it's not a hardware problem, it's a Catalyst problem.

[humbug]

If you already new the deplorable state of catalyst open gl on linux then why did you buy the card to run with the linux catalyst driver and write yet another article like hundreds of others stating the same thing over and over and over like some trailer park princess?

I run mesa/amd on my linux setup because it's open source and superior to the nvidia mesa. If I was stricly into 3d gaming on linux I'd do something different.

Michael is doing a tremendous service to the Linux community and the PC enthusiast community by publishing regular benchmarks. There is no other site which does this for Linux. In the windows world there are loads of sites doing in-depth testing and analysis which directly drives sales behavior and also pushes hardware companies to get better. The hardware vendors sometimes even directly respond to assertions made in those articles. It's very important for the publications to report data and keep hardware vendors honest and make sure that consumers know the truth. And the existence of open source drivers does not make this results less important. The fact is that as bad as these numbers are the open source performance would be even worse. In terms of gaming on Linux this is the best which AMD has to offer. And many users will choose whichever driver provides the fastest gaming performance. Others will choose open source for ideological reasons, which is ok, but it doesn't invalidate these results. Fact is if the open source drivers were faster nobody would be using these.

[DanL]

Maybe support will get better when FirePro (or whatever the workstation line is called nowadays) cards based on Fiji/HBM appear.

I would send the card back and wait until AMD bakes the driver support more (and the price drops). $600 is a heck of a lot of money to find out something we already knew - Catalyst has inferior OpenGL perofrmance/support to Nvidia and AMD struggles mightily with day one Linux support on new products.

[dungeon]

Nobody ever said that. The benchmarks that were run today on Fury and benchmarks that were run in the recent past on the 295 Hawaii obviously show a bottleneck somewhere, I don't think anybody knows what that bottleneck is yet, but it's obviously there. The 285 Tonga results look pretty badly bottlenecked too.

As Micheal said in the article, it's not a hardware problem, it's a Catalyst problem.

It is not Catalyst problem for openarena, did you readed anything from links that you quoted?

http://openbenchmarking.org/prospect...edf1fa68b1445d

R9 290 had 263fps and GTX 980 has 295fps... that is around 10% difference (that is not any big bottleneck for any driver as you see) , and Fury's Michael results are slower then both now even on a bit lower resolution.

http://openbenchmarking.org/result/1...BE-OPENARENA42

It is also not true that no one knows what kind of bottlenecks for other software are, for those steam apps Micheal tested nVidia goes higher mostly because of driver variabile nvidia driver can use and are already available by default in game scripts and flgrx stays flat because it does not use that and those apps are not profiled in current driver

[Nobu]

It'd be nice to see some vulkan benchmarks with this card, and also with OSS drivers (once they get more or less stable), so we have some idea of what the hardware is really capabable of, how much of this is Catalyst and how much is the hardware itself. Won't really know until then.

Edit:

R9 290 had 263fps and GTX 980 has 295fps... that is around 10% difference (that is not any big bottleneck for any driver as you see) , and Fury's Michael results are slower then both now even on a bit lower resolution.

That tells me the exact opposite of what you state: there is a bottleneck in the Catalyst driver preventing the Fury from performing at it's peak, unless you're saying it's HBM is actually slower than GDDR5. Never mind all of the other things that are better in the Fury...

If there was no bottleneck, then the Fury should perform better than both those cards, especially at a lower resolution.