CUDA with AMD ATI Radeon 5870

Dear All,

I was putting together a GPU system with the AMD ATI Radeon 5870 , which boasts 2 teraflops per GPU node. All this on a asus P6T7 WS Supercomputer with 3 GPU’s (plus one more at half speed). The CUDA software says it would work with any GPU’s, but I am not quite sure. So, I am posting here to find a good parallel processing software solution, as a suggestion or fixes which would allow the same archatecture to run other GPU’s than just NVIDIA’s (sorry I know this is an NVIDIA site). That, or suggestions on a cheeper aquisition of Teslas.

Sincerely,

Stephan Watkins

CUDA only runs on NVIDIA hardware, I’m not sure where you read it would work on any GPU. We do also support OpenCL, which is supported by multiple vendors.

I would recommend contacting AMD to find out if they have a solution.

AMD just released a driver for OpenCL which supports both GPU and CPU. You may want to try it.

My best advice for OpenCL development is to use a Mac platform:
Mac OS X 10.6 laptop (even entry-level w/ GeForce 9400M) for development, because OpenCL is fully supported natively on both GPU and CPU, enabling to optimize code for both.

Then run the code and tune it on a Linux server W/O X11, to avoid speed limits and watchdog problems :-)

I would stay with one hardware vendor, for me, and I know just one that have a full range of OpenCL-enabled GPU, enabling development on low-end laptops with IGP, and offering up to server-grade racks for production sites. The other vendor still have big performance hits on it’s old range of GPU, and new GPU (that are promising) are only available for desktops and hard to find in retail actually.

The first have disclosed it’s next generation, that will stay compatible but offers an architecture that is totally oriented toward gpu-computing, offering a clear upgrade path for 2010, and enabling easier port of CPU algorithm as-is.

It’s not a matter of name, because these 2 companies do impressing products, I owned products of both, and was really satisfied, but in term of actual implementation, GPU range, availability of laptop OpenCL solutions, OS implementation of OpenCL, support, nVidia is far ahead, at least one generation ahead!

Thanks for the advice. I already went with the desk top, but have found two examples using the new radeons’s, although only for a simple math test program. I would prefer NvIdia only because it already has most things worked out, but they are 3-4X the price due to this fact (excluding the external server type GPU’s which I would invest in as a company, not as an individual as it would make a PC in the 10-12k range or more).

I also posted on the OpenCL, and AMD boards, and after reviewing my post realized I had learned a bit and may have looked a bit knowledgeless. I don’t think there will be much of a problem with the OpenCL and the AMD’s SDK 2 beta 4, but am now in the waiting process before I get to sit and watch ten million failures. When that happens, I hope my post don’t look two nieve or screwey.

I am though, based on liturature, looking for something that will let me operate a script editor for OpenCL, as it seems many of the commands, etc…may be different enough to be missed, or to cause constant “typo’s” in something like Emacs. I did find somone whom used the AMD’s kernel analyzer on Linux with WINE, but I wanted to avoid putting too much on the system, as I have found the more you add like additional tweeks just to get Windows apps to run on Linux end up causing other problems, especially if you want everything to be minimal, and avoid confussion by generating excessive files, etc…but I guess this is more the person involved than other aspects. That, and I can’t stand not being able to open every single file on the computer with a simple script editor, especially if I officially own it, thus don’t want windows apps.

Thank you for the suggestions.

Thanks for the advice. I already went with the desk top, but have found two examples using the new radeons’s, although only for a simple math test program. I would prefer NvIdia only because it already has most things worked out, but they are 3-4X the price due to this fact (excluding the external server type GPU’s which I would invest in as a company, not as an individual as it would make a PC in the 10-12k range or more).

I also posted on the OpenCL, and AMD boards, and after reviewing my post realized I had learned a bit and may have looked a bit knowledgeless. I don’t think there will be much of a problem with the OpenCL and the AMD’s SDK 2 beta 4, but am now in the waiting process before I get to sit and watch ten million failures. When that happens, I hope my post don’t look two nieve or screwey.

I am though, based on liturature, looking for something that will let me operate a script editor for OpenCL, as it seems many of the commands, etc…may be different enough to be missed, or to cause constant “typo’s” in something like Emacs. I did find somone whom used the AMD’s kernel analyzer on Linux with WINE, but I wanted to avoid putting too much on the system, as I have found the more you add like additional tweeks just to get Windows apps to run on Linux end up causing other problems, especially if you want everything to be minimal, and avoid confussion by generating excessive files, etc…but I guess this is more the person involved than other aspects. That, and I can’t stand not being able to open every single file on the computer with a simple script editor, especially if I officially own it, thus don’t want windows apps.

Thank you for the suggestions.