AMD APU On Linux: Gallium3D Can Be 80%+ As Fast As Catalyst

[liam]

I would like you to show us where Intel HD chipsets outperform AMD APU's with OpenCL, since that's a large deciding factor for many.

I think Iris Pro is faster than every amd apu in opencl.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/17

Pretty interesting results overall. Also notice the latency graphs – quite often Catalyst has huge stutter spikes, while OSS drivers provide a much smoother experience. Very nice.

I'm glad I saw your post b/c I was going to say the same thing. I think there really might be something seriously wrong with catalyst and it increasingly has me wondering why amd isn't putting all their linux eggs into the the radeon driver.

[V10lator]

Also notice the latency graphs ? quite often Catalyst has huge stutter spikes, while OSS drivers provide a much smoother experience. Very nice.

This is exactly my experience when comparing the drivers. When a game stutters with r600g it stutters all the time (overall low FPS) but when it works it works so much better than catalyst, no (micro) stuttering and the rendering is so freaking smooth.
A good example for this is Garry's mod: Catalyst doesn't even get 60 FPS with maxed out settings while r600g gets them without any stuttering at all. When I realized this the first time I almost felt from my chair as I thought the hardware wasn't capable of doing this.

On the other side the LLVM compiler has to be tuned some more, at least for Cayman GPUs as with it I get even more stuttering than with Catalyst.

[res0r9lm]

I think Iris Pro is faster than every amd apu in opencl.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/17

I'm glad I saw your post b/c I was going to say the same thing. I think there really might be something seriously wrong with catalyst and it increasingly has me wondering why amd isn't putting all their linux eggs into the the radeon driver.

lmao They are comparing i7-4950hq ($700) to A10-5800 ($99)

Kaveri is suppose out in January.

http://www.overclock.net/t/1428372/amdfx-a10-7850k-r5-200m-r7-240-in-hybrid-crossfire-over-500-the-gpu-performance-of-iris-800-the-performance-of-richland

[curaga]

I can't help but wonder if TF2 was actually rendered properly. The numbers alone look a lot like the GPU is not really doing what it's supposed to.

Take a look at its TF2 numbers on Windows. All Linux drivers are underperforming in TF2.

[Kivada]

I like the APU graphics test but i would rather see an "APU"-battle intel vs amd.
What is the best APU for linux?
What is the best open source APU for linux?

The AMD APUs will perform better long term with stuff like OpenCL finally getting off the ground on Linux, once it's implemented in the OSS drivers the AMD APUs will really start to shine.

Intel's chips aren't really APUs and the only ones that are decent are those with the Iris Pro 5200 as due to it's eDRAM does offer ok GPU performance, though most people using them are getting them to use that eDRAM as a level 4 cache for the CPU side instead of as GPU vRam.

In either case if at all possible wait till January 14th to buy as that is when AMD's new APU lineup launches. The new APUs are using GCN based GPUs so they will be able to make use of Mantle when it's ported to Linux.

[TemplarGR]

Question for Mr Bridgman

I asked this a few months ago, but never received an answer. So i will ask again:

In my understanding, 2 big challenges for the performance of the open drivers, are shader scheduling and memory management.

From what i recall, and please correct me if i am wrong, GCN uses hardware shader scheduling.

Also, from Kaveri APU and onward, AMD APUs will have a shared memory pool, so memory management will be extremely simple in constrast to earlier apu/gpus, so it will be easier to achieve the Catalyst levels of performance.

So, if the above are true, the open drivers will have an easier task for competing with the binary driver.

I want to ask, how much truth there is on this conclusion?

I would like to get a Kaveri APU for a Linux system, and the main draw would be the open drivers.

[ua=42]

Ah yes, the i7-4950hq, which costs $750, is slightly faster than the AMD A10 which costs $99. So yeah. If money is no object, you can get an intel system that can beat AMD's APU graphically, but if you are on a budget, you can get something that is almost as fast for $650 less.

[Sonadow]

Ah yes, the i7-4950hq, which costs $750, is slightly faster than the AMD A10 which costs $99. So yeah. If money is no object, you can get an intel system that can beat AMD's APU graphically, but if you are on a budget, you can get something that is almost as fast for $650 less.

There's also the point of power consumption. Where, traditionally, Intel doesn't just handily beat AMD in this department; it knocks it to the ground and rubs its face in the dirt.

Which, to some people, is a high enough priority to justify shelling out that small fortune.

[ua=42]

I understand the wanting a longer battery life. But for me I bought an a4 laptop and spent an extra $100 for an extended battery and when I'm browsing the net or typing I can get 8hours of run time. If i'm playing 3d games I get 4 hours.

I'm kind of curious how many more hours the intel gets for the $$$$ I saved.

[Guest]

Take a look at its TF2 numbers on Windows. All Linux drivers are underperforming in TF2.

Is this true with nvidia?