13-Way Radeon Graphics On Ubuntu 14.04 LTS

[phoronix]

Phoronix: 13-Way Radeon Graphics On Ubuntu 14.04 LTS

To complement last week's Radeon Windows Catalyst vs. Linux Gallium3D vs. Linux Catalyst OpenGL performance comparison on Ubuntu 14.04 LTS, here's benchmarks of thirteen different AMD Radeon graphics cards on the near-final state of Ubuntu 14.04 LTS "Trusty Tahr" in its current development form.

http://www.phoronix.com/vr.php?view=20112

[Azrael5]

Could you test HD3850 AGP and HD4670 AGP? As the latest powerful AGP videocards?

[peppercats]

Can anyone tell me why the 260x is so poor? Is there a fix for it? I was considering buying one because the radeon OSS drivers are so good now but it seems to perform awfully.

Does the 265 perform just as bad?

[bridgman]

Can anyone tell me why the 260x is so poor? Is there a fix for it? I was considering buying one because the radeon OSS drivers are so good now but it seems to perform awfully.

Does the 265 perform just as bad?

My assumption was that the low performance of 260X was because the dpm work from the 3.14 kernel (enabling dpm on CI for the first time) had not yet been picked up in the Ubuntu 14.04 kernel, but Michael's comment about the same problem showing with the 3.14 kernel surprised me.

The 265 is based on Pitcairn so should run a bit faster than the 7850.

[Michael]

Could you test HD3850 AGP and HD4670 AGP? As the latest powerful AGP videocards?

I don't even have those cards... Nor any AGP systems anymore... Not interested in AGP at all anymore.

[peppercats]

My assumption was that the low performance of 260X was because the dpm work from the 3.14 kernel (enabling dpm on CI for the first time) had not yet been picked up in the Ubuntu 14.04 kernel, but Michael's comment about the same problem showing with the 3.14 kernel surprised me.

The 265 is based on Pitcairn so should run a bit faster than the 7850.

So the 265 is based off a different architecture than the 260x? That's an awkward naming scheme.

[bridgman]

So the 265 is based off a different architecture than the 260x? That's an awkward naming scheme.

I guess, but there probably weren't a lot of naming options for a card that falls between 260X and 270 in performance. Calling it the "270 Lite" probably wouldn't fly.

The number seems more reasonable if you look at it as a performance metric rather than a GPU die identifier, of course.

BTW I checked the 3.14 kernel tree and AFAICS dpm is still enabled for Bonaire there, so if Michael is seeing low performance with a 3.14 kernel I don't think I understand why that is.

[peppercats]

I guess, but there probably weren't a lot of naming options for a card that falls between 260X and 270 in performance. Calling it the "270 Lite" probably wouldn't fly.

The number seems more reasonable if you look at it as a performance metric rather than a GPU die identifier, of course.

BTW I checked the 3.14 kernel tree and AFAICS dpm is still enabled for Bonaire there, so if Michael is seeing low performance with a 3.14 kernel I don't think I understand why that is.

Because the 265 is based off of the 7000 series would you say the OSS drivers for it are more mature than the other 2xx cards? Thank you for the help BTW, I know nothing of AMD cards.

[bridgman]

Yeah, I think that's a safe bet. If agd5f posts and disagrees though, go with what he says

I expect that most of the graphics code would be common between 260 and 265, the differences would be more in the "uncore" bits like dpm.

[DarkFoss]

I agree somethings wrong with the test set-up the numbers are way off on the R7 260x. It's last generation comparable card is the HD 7790. There is no way that the 260x should be out performed by the much older 6450, which isn't even a gaming card in most peoples eyes. Even an R7 260 non x would have a hard time putting up such pitiful numbers.