Can you use instances of the Amazon EC2 GPU for real-time rendering?

Question

I need a remote PC / server that has a decent 3D map to do real-time 3D rendering ... imagine you are running a 3D game on a remote server and this is a good comparison.

Most VPS and dedicated servers do not have good graphics capabilities for obvious reasons, but Amazon has special instances of the GPU. They sell to calculate GPGPUs, using GPUs to crunch data using tools like CUDA, but I wondered if they could also be used for real-time 3D rendering.

Can someone give a solid answer to this?

Edit: I have to add my own 3d code, and I want to know the capabilities of EC2 for this purpose, not the general EC2 question

Answer 1

Amazon GPU servers feature NVidia Tesla GPUs. Although they are best suited for GPGPUs, they also have more than average real-time rendering capabilities. Although in this respect they are inferior to NVIDIA GTX cards (see GPU specifications on the NVidia Website).

You can now use Amazon for real-time rendering, but your network bandwidth will be your bottleneck. Tesla cards can be used with OpenGL to render graphics to Offscreen buffers very quickly, but then you should find a way to read pixels for each rendered frame and transfer it to the client with an acceptable frame rate. The OpenGL pixel read from the GPU is already very slow (although you can do some hacks using PBO ping-pong), but I really don't think you can stream packets using standard resolutions (800x600 or even less) from a remote server so that the client receives it with the minimum acceptable frequency. I believe that this will be possible in the future :)

PS My answer is based on personal experience with Amazon EC2

Answer 2

Yes, amazon EC2 is well suited for rendering. I have been doing this on a large scale for over 3 years for a mobile application. The bandwidth was great for short animations, which I am switching from EC2 to S3 / CloudFront.