ATI y Folding@Home, entrevista exclusiva

Intro:

Hello Andrew, what is you job at ATI?

My job title is Software Product Manager. My role is mainly divided into two parts. The first part is internal to ATI, where I help develop the CATALYST strategy, write requirements for features, work with developers etc. The second part of my role is external facing - creating presentation, white papers and other materials to present to customers and press.

What was the motivation from ATI to support distributed computing? Did you just wanted to test GPU power? Participate in a noble cause? learn how to code for GPGPU?

Since the launch of the Radeon X1800 we knew that our hardware was capable of doing more than just rendering graphics (we actually started working with Stanford’s Folding@Home project before we even launched the Radeon X1800). We knew that any application that required significant parallel processing would greatly benefit from ATI’s powerful GPUs. So we started talking with Stanford, and both parties thought it would be a great application to accelerate with the GPU – and of course it’s a very noble cause.

What is next in terms of GPGPU from ATI, physics?

We’ve got a number of great GPGPU initiative coming soon (including physics), but I can’t go into too much more detail right now – you’ll see a lot more noise about this in the coming months.

Why did you choose Folding at Home as the first supported GPGPU?

We had been aware of Stanford’s Folding@Home project for quite some time, and thought it would be a great application for GPGPU.

Do you plan on adding more distributed computing program support, one of the many initiatives using BOINC?

Yes, we would definitely like to work with more scientific projects (including BOINC based projects) to get GPU acceleration built in.

Do you know why NVIDIA has shown little or no interest in supporting distributed computing? Is there a hardware reason behind it?

No, I don’t know anything about it. (And if I knew I wouldn’t comment on it either).

Hardware:

How is the expected lifetime of a video card affected by F@H?

The lifetime of the video card won’t be affected by taking part in the Folding@Home project.

Will you change Your warranty policy due to this increased and new use?

There will be no changes to the warranty.

Have you thought of problems regarding high temperatures? GPUs normally work at high temps but because of the nature of the process on the GPU they rarely sustain 100% charge for long periods of time

The GPU does not reach a temperature that is considered dangerous for the GPU when running Folding@Home.

When can we expect to have more GPUs supported to fold? budget and mainstream families, for example

Hopefully in the next couple on months.

If we use Crossfire for Folding@Home, will the driver see Crossfire as one big GPU? or will it “see” both cards separately?

We realize that adding Crossfire support for Folding@Home is a high priority item, and we hope to get Crossfire support included in the next few months. Actually Stanford just released a Multi-GPU version of their Client (Nov 3rd) to support multiple ATI GPUs (not running in Crossfire mode though). This is all I can say on that subject right now.

Do you have plans for special F@H (or GPGPU or more general stream processing) setups? Like those of Evans & Sutherland’s, but somehow using PC technology?

No, ATI’s Folding@Home GPU Client will work on any PC with a Radeon X1950/X1900 Series product.

Drivers:

We saw that Catalyst 6.10 include the F@H client. Will this be the default way to deliver the client?

Catalyst 6.10 was just posted Oct. 31st, which supports the Folding@Home GPU client for the Radeon X1900/X1950 family of products. However, Radeon X1950 PRO is not supported in Catalyst 6.10 and thus it cannot use the included F@H client either, and in those cases the client should be distributed separately.

When F@H on GPU will be supported on different OSs?

In the next few months we plan on adding Vista support.

How much room to tweak do you have at software level to improve performance (estimative value if you have)?

We’re working with Stanford on further performance increases using ATI’s own custom interface for GPGPU purposes. (In the range of a 150% - 200% performance increase ) We’re hoping to release this in the next month or two.

Future:

How will DirectX 10 affect F@H?

DirectX10 offers increased flexibility and should provide us with new opportunities for optimizations in the Folding@Home code. But our current focus is helping Stanford make use of our own custom interface for the GPU, which as mentioned above should provide a substantial performance improvement.

How will new geometry shader of DirectX 10 affect performance of GPGPU?

Geometry shaders are designed to operate on polygons and other graphics primitives. In stream computing applications, such as F@H, operations are performed on generic data rather then vertices, polygons, and pixels. However, this will not really matter much for DX10 applications running on GPUs with a Unified Shader Architecture, since they will be able to use the same set of processing units to operate on any type of input data.

When building R600 did you think about the possibilities of GPGPU? did you improve anything at hardware level to get more benefits?

R600 is still a secret and unfortunately I can’t talk about upcoming products – but you can expect new GPGPU features plus an amazing increase in performance

Andrew, in the name of CHW’s Staff and all our readers, I want to thank you for your time