AMD APUs Don't Appear Affected By Linux 3.12 Change

[Tyler_K]

By the way, Michael, did you force dpm for APUs?

Ummm...Michael, I don't mean to be a dick (i really don't) but I only got one question for ya... did you enable DPM for these?.

You guys could have answered the question for yourselves very easily, but here you go anyway

http://openbenchmarking.org/system/1...203.11/cmdline
http://openbenchmarking.org/system/1...%20Git/cmdline

[Ericg]

You guys could have answered the question for yourselves very easily, but here you go anyway

http://openbenchmarking.org/system/1...203.11/cmdline
http://openbenchmarking.org/system/1...%20Git/cmdline

Thank you, Tyler, the amount of times I've been on OBM.org could probably be counted on one hand so I don't know where everything is at on it

[Tyler_K]

Thank you, Tyler

np.

the amount of times I've been on OBM.org could probably be counted on one hand so I don't know where everything is at on it

I'm about a second hand ahead of you in the count then

As a quick FYI:
In this case, Micheal linked to the OBM page directly (http://openbenchmarking.org/result/1...SO-LINUX312G09) which includes the black box summary. In most of his write ups, he embeds that box (example from the prior related article: http://www.phoronix.com/scan.php?pag..._preview&num=1).

Anwyay, From the black box you can click on the system logs (bottom right) ... or if you're on the OBM page, you can also hit the blue system logs tab, which then provides a link to:

system logs relevant to the tests i.e. http://openbenchmarking.org/system/1...SO-LINUX312G09 ... clicking on either of those links, reveals a whole lot of relevant info which you can further drill down into.

[rudregues]

Nice!

It's really good to see my notebook APU being benchamarked, I already tried myself but failed (I'm a just a little noob haha). But I can try with Xonotic (I could make it run on top of Sabayon, but it was the only benchmarked that worked). I would like to do that benches in Gentoo too, but if in Sabayon it fails it can fail to run in Gentoo too. And trust me, a software that takes much time to compile and fails is extremely annoying.

Michael, don't worry with adblock or similar by me, since I don't use this never. Even with my smartphone. I prefer support all sites I like allowing ads run because I don't have money yet (I'm in college). When I finish my studies and start to get true money I will help (will help GNU Octave too and another programs I like and use).

[Luke]

Another way to support the mission of Phoronix

It's really good to see my notebook APU being benchamarked, I already tried myself but failed (I'm a just a little noob haha). But I can try with Xonotic (I could make it run on top of Sabayon, but it was the only benchmarked that worked). I would like to do that benches in Gentoo too, but if in Sabayon it fails it can fail to run in Gentoo too. And trust me, a software that takes much time to compile and fails is extremely annoying.

Michael, don't worry with adblock or similar by me, since I don't use this never. Even with my smartphone. I prefer support all sites I like allowing ads run because I don't have money yet (I'm in college). When I finish my studies and start to get true money I will help (will help GNU Octave too and another programs I like and use).

If everyone here reported their experimental results with the hardware and games they already have instead of passively consuming content, maybe all the work and expense of getting information on a wide variety of hardware would not fall onto the shouders on one person. Not everyone's results will be as repeatable or reliable as running a fixed suite of fixed benchmarks is, but comparisons can still me made. For that matter, if only the one HD6570 test had been done (on one card), we wouild not now know that not everyone gets bad results from R600g on HD6570.

OK, here is a summary of some of my HD6750/FX8120 (4.4GHZ) results on Critter, and Scorched3d (as reported on some other threads): Scorched3d 85fps at peak (lightest maps) with gallium backend and Linux 3.11 with dpm enabled, compared to a peak of 106fps on fglrx a year and a half ago, with 1080p resolution. Scorched3d does not like the sb backend for some reason. Critter (2d but written in Opengl) gives 630fps at peak right now on Linux 3.12 with either previous default backend or with sb backend and cpu governor set to 4.4GHZ fulltime, compared to 1,000 or so fglrx same settings but CPU governor "ondemand." Both at 1080p with "show nebulas" disabled.

NO DIFFERENCE that I can see between Linux 3.11 and Linux 3.12 on the Radeon 6750 in Critter, in Scorched3d, or in 0ad. In Critter no improvement in the "ondemand" governor setting issue either, I still get over a 50% loss of framerate in Critter with OnDemand. That doesn't mean there is none as Scorched3d is very hard to benchmark, Critter is very light, and 0ad CPU limited. My results may not apply to games like Xonotic that I have never played and do not have, but if I sit on them and don't talk about them nobody else can benefit from my testing.

Open source is supposed to be about community contributions of effort in whatever way each person is capable, not passive consumption. Passive consumption is the Windows/Hollywood way, and Linux would not have even gotten out of Linus's garage that way, Some contribute funds or hardware, some contribute code, some contribute testing. All of these are needed for the open source model to work.

[djdoo]

Bus Improvements?

Well since we tested an APU, of course a low end one maybe some A10 non GCN would give more accurate results, lets examine our guesses!

- Intel CPU governor improvements:
To check this we should run the test with the radeon cards on an AMD CPU system. Also if this is the case there should be improvements to the Catalyst driver's performance also, so running the test with 3.12 and Catalyst should answer this question!(if this can be done now)

-PCI Express Bus or RAM Access
This is my wild guess considering Alex words about GART size and thinking that it is not affecting the APUs so it maybe a BUS enhancement in the 3.12 kernel like it was when PCI Express 2.0 was enabled by default.

[benmoran]

Well since we tested an APU, of course a low end one maybe some A10 non GCN would give more accurate results....

I think they might benefit more than the e350s did. When I was running 3.11-git while it was in rc, setting the CPU to performance manually, did produce a noticable speedup. Once 3.11 hit release, I stopped doing that because the performance has been good enough since then. It does seem to imply that the new CPU governor implementation would have an effect as well.

I have an e450 myself in my HTPC, and the thing pretty much always runs at maximum clock when you do anything heavy on it. I think because they're weaker APUs, they might just be running at max frequency all the time already. The "big" a10 APUs might have different results.

If I can find time I'll run these tests myself.

[Kemosabe]

The title of this article is misleading and raises expectations it does not come true.

[agd5f]

The likely reason AMD APUs (and also Intel GPUs) are not affected much by the cpufreq change is that they are most likely already GPU bound even at lower frequencies. The higher end dGPUs still have a good bit of GPU capacity left so higher CPU clocks help take advantage of that. That's why you see the largest performance increases with the larger dGPUs.

[curaga]

Now, if you DID enable DPM, then okay, great, this is more data to figure out what exactly is going on.

You were already answered, but I think it needs mentioning: my E-350 units only have one power state, and so DPM would not affect them at all. It's reasonable to assume all E-350 are like that.