The Linux 3.13 Kernel Is A Must-Have For AMD RadeonSI Users

[bridgman]

Question: Will DPM work out of the box for HD6450 with kernel 3.13?

IIRC it's enabled by default in 3.13 but still problematic on *some* systems, so it was disabled on BTC (HD 64xx through 68xx) in 3.14 for now.

[tom.higgy]

Thanks for the updated RadeonSI benches, Michael. You've got paypal.

Between previously running 3.12 with radeon.dpm=1 with mesa from git to now running 3.13rc7 and mesa-git with the performance patch to go with it, my mining performance has gone from 70-150 Mh/s to currently hovering around 490Mh/s using a single 7850 and bfgminer's OpenCL backend.

I haven't recently benched any games, but the OpenCL performance is much improved (at least for this one use case).

That's interesting. The bitcoin mining hardware comparison lists 7850 at 280-360 MH/s, presumably with Catalyst. A waste of power but still interesting.

I guess it would be nice to see some early comparisons on OpenCL on Clover, Catalyst and any other available implementations.

[Kivada]

Clearly irony is not a word you are familiar with! If you're really that interested in a certain game's performance you could always test it yourself.

So everyone everywhere should waste the money, time and resources to buy and return sever pieces of hardware every time they want to test something out on Linux?

Sounds like an excellent plan...

[Veerappan]

That's interesting. The bitcoin mining hardware comparison lists 7850 at 280-360 MH/s, presumably with Catalyst. A waste of power but still interesting.

I guess it would be nice to see some early comparisons on OpenCL on Clover, Catalyst and any other available implementations.

Yeah, that's what I had been expecting as well. There was a previous mesa/kernel setup that I had that was capable of ~230Mh/s with this card, but then a software stack upgrade dropped that to ~1/2 of the speed.

I re-installed my OS, did some software/kernel upgrades recently, and now I've got bfgminer currently reporting 492.2Mh/s with the 7850 running at 65C.

[Dukenukemx]

Factoid: Radeon in general has fairly poor memory allocation in comparison to Catalyst, and the better your card is the more obvious this becomes because then Memory becomes the bottleneck. By poor memory allocation I mean the driver is kind of 'dumb' about where things should be in memory, whats safe to move out of GPU memory / just drop completely, and the likes. Its on the 'volunteer todo list' last I heard, because optimized memory algorithms were never on the original AMD roadmap and plans.

One additional note... In most cases, especially high end cards, Catalyst will ALWAYS BE FASTER. Kernel and Mesa would never accept code that said

If (CardID == 7970)
{
Codepath 1
}
else If (CardId == 7770)
{
Codepath 2
}
else // See: Low and medium cards
{
CodePath 3
}

Meanwhile Catalyst just might, and probably does, because they have a financial incentive to make sure that every single card gets the most performance that it can get, even if it means micro-managing code paths. The Kernel and Mesa devs will accept the code that works the best on the most cards as possible, and is the most maintainable, even if that means maybe only hitting 90% performance of the possible because you've got a high end card

That can be a good thing and a bad thing. Depends on what graphics card you own. For example, CodePath 1 are for expensive graphic cards, which get a lot more attention from AMD. Cheaper graphic cards are not generally given a lot of attention, and older graphic cards are even worse. Code is changed, but mostly just bug fixes.

Remember, AMD and Nvidia's business is to get you to buy new hardware as often as possible. Optimizing older and cheaper hardware would be against their interests. While the open source drivers are optimizing the code for one path, it might be better in the long run for everyone then to have separate optimized code paths.

Just look at AMD's patch notes. They generally favor newer and expensive graphic cards over cheaper and older.

[Calinou]

Why is radeonsi that slow? It was faster on older benchmarks! Your benchmark thing seems to be really inconsistent.

[Rich Oliver]

I'm running three monitors off a 7790 2 1920 x 1200 + 1 1920 x 1080. Up until now I haven't been able to get my screens configured properly with radeon. Most distros it doesn't even see them as separate monitors, so its update and then install catalyst, but maybe I'll give a radeon arch install a try with 3.13.

[smitty3268]

Even if it runs acceptably with GLAMOR, the current state of it means you are wasting a lot of power and generating a lot of heat to do what an Intel APU could do without breaking a sweat.

Modern desktops running in OpenGL mode don't really even touch GLAMOR anyway. Glamor accelerates XRender, 2d acceleration code.

That can show up in certain apps, but not usually the desktop shell itself anymore.

[zanny]

Modern desktops running in OpenGL mode don't really even touch GLAMOR anyway. Glamor accelerates XRender, 2d acceleration code.

That can show up in certain apps, but not usually the desktop shell itself anymore.

Would that apply to Wayland? IE, since Wayland requires opengl-es, desktop Wayland wouldn't use XRender at all, so what would 2d acceleration be doing then?

[sasy360]

Im on HD7950 and mesa/oss driver since some months. 2D performance was all the time fluently enough for me. I did not recognize stutters or lags but sometimes some visual glitches on overlays - but those stay only for seconds
The raw performance of the HD7950 is so big, it is no wonder that even if it is so unoptimized that it runs 2D fluently. My guess is that goes also for any HD 78xx or 77xx. 2D is just not a real problem today. Or is it for anyone reading this? Contrary experiences despite low benchmark numbers compared to catalyst?

2D performace is very important even these days. And the current Glamor state with radesosi driver is total crap.
I didn't pay for some glued, genereic 2d driver BS. Definitely no more AMD graphics for me.