AMD Betting Everything on OpenCL

The Problem of Betting Everything on OpenCL

Bit-tech: If OpenCL doesn't take off, and GPGPU computing doesn't become the big thing that you're predicting, does your CPU architecture alone have enough power to compete with Sandy Bridge?

Sasa Marinkovic: It's not a question of 'if' it's going to take off – if you look at things like IE9 there's already GPU acceleration – you look at Microsoft PowerPoint and it's got GPU acceleration, you look at all the applications coming down the road they all have GPU acceleration, so I don't think it's a question of 'if' but 'when'.

BT: A lot of that GPU acceleration doesn't necessarily use OpenCL, though, does it?

SM: It doesn't, but that's the path that we're on.


Terry Makedon: I think the way to answer it is to say that AMD as a company made a very, very big bet when it purchased ATI: that things like OpenCL will succeed. I mean, we're betting everything on it. Our most important major product launch is Llano. We're an APU company; that is our priority – we've given that direction to the investment community, our CEO has said over and over that our priority now and for the future is APUs, so it's a very big bet.

BT: Cynics might say that the GPGPU computing revolution has failed to materialise. We haven't seen a lot of software using it – even using CUDA – and we've been promised this big thing for years, and so far we still haven't seen it break into mainstream software. You do get it in some big software packages, but most people aren't using GPGPU compute on a regular basis. Is that really going to change?

TM: I don't think we're actually quite ready to see the OpenCL revolution crack the PC ecosystem open yet. I think we're on the tracks, so our first major step is the developer conference. Think of it like a console – what do you need to sell a console? You need some really good games – that's where we're at right now, so we have a very good SDK [software development kit], we've got very good hardware – you take these parts, you take the hardware, you take the development toolkits and the development environment that we need, and the next thing in that chain of events is applications.

So yes, I'll give you the fact that we don't have any major applications at this point that are going to revolutionise the industry and make people think 'oh, I must have this,' granted, but we're working very hard on it. Like I said, it's a big bet for us, and it's a bet that we're certain about.


One of the things we're investigating, and I'm not making a statement to say that we're definitely doing this right now, but we've been looking at things like an app store. So when we start getting these OpenCL or DirectCompute-based applications, or any applications that take advantage of our hardware, we'll have a nice marketplace where people can go and find out about it, and all that kind of stuff. So we're doing things behind the scenes – the first step is getting the hardware, the second step is getting the developers and the third step is getting the application support.

The bottom line is that if you have five pieces of software that do photo-editing, and one of them processes effects much quicker, that application will have the advantage over the other ones, and that will dictate which one becomes the market leader, simply because it takes advantage of the GPGPU improvements.

And obviously the beauty of OpenCL is that it's device-independent, so if a developer's going to come up with an application, let's just say somebody wanted to do a transcoding app, then if they write that in OpenCL it'll work on the whole PC ecosystem, whether it be Intel, AMD or Nvidia.