It looks like ARM (NASDAQ:ARMH) is jumping aboard the OpenCL train, meaning all the major players are now behind the standard. It may not sound like all that big a deal, but it is, OpenCL is now officially supported top to bottom.
Jem Davies, ARM Fellow and VP of Technology, Media Processing Division at ARM is going to talk about the future of heterogeneous compute, and how they are going to support it during the fourth keynote at AMD’s Fusion Developer Summit (FDS). This late addition to the lineup fills two big holes at once. The first one is that with official ARM support, as long as you use their GPUs, you are have a very good chance of having your OpenCL code work across most devices. There are two major CPU architectures out there, x86 and ARM, both of which should soon run OpenCL code. Even Intel officially grins and bears it, and will honestly pretend that they don’t support it because Apple has a gun to their head.
On the GPU side, you have AMD, Nvidia, and a few of the big ARM focused IP merchants all behind the spec, so it is almost ubiquitous there too. Unless you have a really odd architecture, or an Intel ‘semi’-GPU, you are good to go with OpenCL now.
The other key thing, and it didn’t hit me until just recently, is the title of the keynote, it keeps referencing heterogeneous compute. I kept asking AMD about their plans for multi-GPU and CPU/GPU splitting of code, and they kept giving the same response. “No comment, but it is high on our list.” Then they smiled. A lot.
Then it hit me. All the keynotes for FDS are lined up around one key thing, heterogeneous compute. That is what I have been asking about from the beginning, CPU and GPU, working together on the same code base. Duh. One keynote is about the CPU, one about the GPU, one about the languages you need to write for it, and one about the integration. It is so tidy it is almost like they planned it that way.
Unfortunately, no one I talked to at AMD has yet spilled the beans on when the integration roadmaps will be laid out, but you can see one hell of a clue on slide 8 here. With Ontario/Zacate and Llano, we are somewhere between arrow 1, Physical Integration, and arrow 2, Optimized Platforms. We have not seen the ‘Fusion Interconnect’, nor have we seen bi-directional power management, but we have seen the other four on the first two arrows. Actually, Intel sort of has bi-di power management, but it is borderline to say that one of the two parts is a GPU.
Why is that important? You can pretty much guess the topics that will be covered in the AMD side of the keynotes now. The only question up in the air is how much of arrow 3, Architectural Integration will be in the keynote about the next GPU architecture. Fusion Developer Summit is going to be an interesting time.S|A