Rendering on Tesla systems

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi All,

I have a question: is it possible to use on the Tesla systems or cards for OpenGL rendering? I do not need to have video oputput, but need to render into machine memory instead.

2

This is definitely possible in Linux.

3

But what is the difference between OpenGL and CUDA on Linux and Windows? If it works on Linux it should also work on Windows.

What is really interesting for me after it works to know about performance figures of rendering on Telsa card vs 8800 Ultra or other.

4

It should work in both Linux and Windows. However, since Tesla GPUs do not have display connectors, you will not be able to create a display context or window, or to “swap buffers”, etc. You’ll have to render to off-screen buffers (FBOs or pbuffers).

Performance should be similar to GeForce rendering on an equivalent GPU.

Mark

5

Very nice!

Hi all, new to this forum and to CUDA programming.

And I have a very similar question.

Is it possible to use Tesla boards to speed up rendering with Gelato (or am I missing a point somewhere here?)? I have a QuadroFX 5600 and I’d like to add 3 more Tesla boards if it would speed up Gelato rendering (I know that I can’t SLI Quadro and Teslas together but independant speed up…).

Thanks in advance.

Best regards,

JF

6

I don’t know if it is possible to speed up Gelato. But I know that if you have a multi GPU you cannot use CUDA on it if you want to use those GPU in SLI mode. So i think you can also buy Quadros for that.

7

Actually, I was corrected on this. It should work fine in Linux, but in windows in order to create a GL device context you need a display adapter attached to the desktop. Since Tesla does not present as a display adapter and is not attached to the desktop, I don’t think OpenGL will work on Tesla under Windows.

If you need a device like a Tesla S870 that can run OpenGL, we do have a Quadro product, the Quadro Plex Model S4 (NVIDIA Quadro Legacy Products).

Mark

8

Hi Mark, could you make it clear how to do that. If i’m not administrator in Linux can I create OpenGL context on the testla only machine and render to the FBO object of that context.

Thank you