Open Source Graphics Processor (GPU)

[svartalf]

[shodruk]

I wrote OpenGL ES 2.0 commands streaming framework for Parallella.

viewtopic.php?f=34&t=708

It streams OpenGL commands from Parallella to Raspberry Pi over network, and render 3D graphics on RPi.

With this framework, Parallella can run full OpenGL ES 2.0 programs and render 3D graphics through RPi with over 60 frames per second.

[svartalf]

I saw that just as I came into the sites. Coolness. How detailed are the scenes you're rendering over this link?

[shodruk]

[svartalf]

[jlambrecht]

[TomasD]

Maybe we could think of any other possible video acceleration? E.g. using a CPU with dedicated video subsystem, so we can deliver the video through PCIe? I believe there could be a kind of cheap solution to that. E.g. those AllWinner boards are pretty cheap. I presume CPUs also.

[sunnystormy]

I apologize if I'm coming into this conversation rather late...

I read the whole thread and I noticed some comments that said that using the Parallella as a GPU is a vain effort because it won't reach the performance of current off-the-shelf GPUs. I'd like to point out that merely *having* the prospect of an open-source GPU is well worth it.

I don't mind if the output is somewhere in-between the PS2-XBOX360 range of GFLOP performance; simply having the option of being able to do it yourself is quite liberating!

I'm still waiting for the next available opportunity to purchase the Parallella, but the idea of using two of these boards (one for computation, and one for GPU processing/shading) linked together by an Ethernet cable already has me very excited! I just wanted to give kudos to whomever suggested it.

[svartalf]

Actually, it depends on what you're aiming for.

Trying to best a current Modern NVidia or AMD Radeon with just a 16-core parallel processor trying to do something like llvmpipe from Mesa3D... Not going to do so hot.

Trying to best all but the top-end ImgTec SGX or similar... Quite a bit closer. You won't edge out the top-end mobile GPUs, but you will get the midrange with some clever coding.

Trying to go your own way, using alternate algorithms and adaptations to OpenGL/OpenGL ES... Up in the air. You can't say for positive either way.

Now a 64-core unit, on the other hand...little more interesting in the latter two categories. If only Adapteva can get the costs on the things down.

[sunnystormy]

I think for me it would just be about getting it to work the way I want to. I don't care about "besting" anyone.