Intel Beignet Is Working Out Surprisingly Well For OpenCL On Linux

[phoronix]

Phoronix: Intel Beignet Is Working Out Surprisingly Well For OpenCL On Linux

Beignet is the project out of Intel's Open-Source Technology Center for exposing GPGPU/compute capabilities out of Ivy Bridge hardware and newer when using a fully open-source Linux stack. While Beignet differs greatly from Gallium3D's Clover state tracker, this Intel-specific open-source OpenCL implementation is working out quite well for Ubuntu Linux...

http://www.phoronix.com/vr.php?view=MTc3Njk

[pali]

You can try daily builds of Beignet on Ubuntu 12.04 from my PPA:

https://launchpad.net/~pali/+archive/ubuntu/graphics-drivers/+packages

Now it is compiling with llvm3.5 without problems.

[liam]

Thanks so much for this, Michael.
I really hope the desktop will start writing to opencl soon instead of simply relying on the cpu (including simd). Gnome is especially bad about doing that, imho.

[rikkinho]

it works with trusty?

You can try daily builds of Beignet on Ubuntu 12.04 from my PPA:

https://launchpad.net/~pali/+archive/ubuntu/graphics-drivers/+packages

Now it is compiling with llvm3.5 without problems.

it works with trusty?

[pali]

it works with trusty?

No, packages are only for precise.

See: http://www.phoronix.com/forums/showt...04-Precise-LTS

[schmidtbag]

IMO, OpenCL puts some serious use to IGPs. Considering the price of intel processors, in a system with discrete graphics the GPU is a decent chunk of change going to waste. Even if the GPU were worth as little as 3% of the entire processor, in some cases that's enough money to get you another 2GB of RAM, maybe more. I suppose that if you're not playing a game there's no reason you couldn't just use the discrete GPU, but, if you can combine IGP with discrete in a single OpenCL task, that would be a nice bonus.

I wish folding@home would support beignet, but, what I REALLY wish is if intel graphics could support physx.

[liam]

IMO, OpenCL puts some serious use to IGPs. Considering the price of intel processors, in a system with discrete graphics the GPU is a decent chunk of change going to waste. Even if the GPU were worth as little as 3% of the entire processor, in some cases that's enough money to get you another 2GB of RAM, maybe more. I suppose that if you're not playing a game there's no reason you couldn't just use the discrete GPU, but, if you can combine IGP with discrete in a single OpenCL task, that would be a nice bonus.

I wish folding@home would support beignet, but, what I REALLY wish is if intel graphics could support physx.

I thought physx was closed to all but nvidia?
Wikipedia seems to back me up on this:

Any CUDA-ready GeForce graphics card (series 8 and newer, with a minimum of 256MB of video memory^[13]) can take advantage of PhysX without the need to install a dedicated PhysX card.

Nvidia really pisses me off. They've got some amazing engineers, but they do crap like that, continuing to only partially support opencl, and http://streamcomputing.eu/blog/2014-...snt-want-know/

[ChrisXY]

Is all the code completely intel specific? Because the gallium3d clover state tracker is at opencl 1.1. And they both use llvm backends...

Edit: Also, when multiple llvm versions are installed, PATH should be set to the bin/ directory from the correct llvm version when compiling beignet, because it's running clang while compiling.

[Sdar]

Nvidia really pisses me off. They've got some amazing engineers, but they do crap like that, continuing to only partially support opencl

AMD really pisses me off. They've got some amazing GPUs, but they do crap software that can't compile large kernels, continuing to only partially support opencl.

It seems that the best Opencl compiler right now is the Intel cpu one.

[zeealpal]

I was quite surprised, but when I ran SmallLuxGPU (the engine used in LuxMark) on my Intel i5 4440 with a HD4600 iGPU, the iGPU rendered faster than the CPU, and together it was (naturally) twice as fast as just using the CPU, which I was quite happy with.

If OpenCL support is perfected to work on more complex software, that basically turned my cpu from a quad to an 8 core