AMD Radeon R9 290: Still Not Good For Linux Users

[Calinou]

Have you heard of nvidia's fcat - Frame Capture Analysis Tool? Do you know if it works on linux?

No, it doesn't.

[bridgman]

But I think the RadeonSI hasn't implemented basic support for 290 yet, so we're still left in the dark on that one. If you search phoronix regarding the subject, you can find lots of articles mentioning it. I haven't done that though, so I might be wrong about my previous statement.

AFAIK support for Hawaii (R9 290/290X) was enabled a couple of months ago. You need a 3.14-family kernel to get the latest DPM fixes and have DPM enabled by default. What it probably doesn't have yet is much user testing since the support is pretty recent -- most of the code paths are the same as Bonaire (7790, 260X) but there could be Hawaii-specific bugs.

EDIT - just checked mesa logs, looks like most of the Hawaii-specific activity in radeonsi was in the last couple of weeks of November.

[schmidtbag]

Michael:
Based on what I see in the images, it seems as though you only test 1 GPU at a time, meaning, you open up a computer, take out an old GPU, insert a different one, boot up, run tests, and repeat. To make life easier for you, you should try cramming as many GPUs in 1 system as you can at a time. Before anyone freaks saying this is a bad idea, hear me out: Put all high-end GPUs in 16x slots that are documented to run at the full 16x bandwidth. Then, put all mid-range GPUs in slots that operate at 8x bandwidth, and put all low-end GPUs in slots at 4x bandwidth. This will allow at LEAST 4 GPUs to be inserted into a single system at a time, and there are plenty of options to switch between GPUs. I am NOT saying to test all 4 at the same time - that's a terrible idea.

Just so everyone is clear though, not even high-end GPUs use the full bandwidth of PCIe 2.0 @ 16x. At worst, there might be a 1FPS drop by putting mid range GPUs into an 8x slot. If Michael reads this, he could really save himself a lot of time. Besides, he might be able to do SLi/Crossfire tests this way - I'm not sure he knows this but I'm pretty sure he's able to crossfire some of the parts he currently owns.

[Azpegath]

But how does the card fare with radeonsi?

I was interested in that as well. I'm actually thinking about wasting money on this card, instead of buying a 7xxx series card for the same amount of money. But I think the RadeonSI hasn't implemented basic support for 290 yet, so we're still left in the dark on that one. If you search phoronix regarding the subject, you can find lots of articles mentioning it. I haven't done that though, so I might be wrong about my previous statement

Michael: Could we get the same tests done on Windows as well? Perhaps not as many cards, half of them will do. Just to see how good the 290 is "supposed" to be for the tests you ran.

[jakubo]

What about it? It wont work in VM?

I was referring to the article. Since Michael was using the 3.12 Kernel the faster GPUsd wouldnt run at full load under certain conditions. Thats why its needed to switch on performance govenor, or have at least 3.13 kernel. And i wonder if that happened here in this article. There are a couple of articles referring to that issue.

[FutureSuture]

But how does the card fare with radeonsi?

[D_T_G]

fcat

Have you heard of nvidia's fcat - Frame Capture Analysis Tool? Do you know if it works on linux?

[mike4]

Heard that AMD will be much faster in 3.13 or 3.14? Seems they found some bug in the kernel.

Interesting would also be to see 3D performance differences using different WM, Gnome, Unity etc. in upcoming 14.04.

[mmrezaie]

performance govenor?

What about it? It wont work in VM?

[jakubo]

performance govenor?