ARM Discusses Future in Keynote: Heterogeneous Computing, CPUs and GPUs

Posted by Jem Davies,

22 June 2011

As promised, I gave a keynote speech at AMD’s Fusion Developer Summit last week, and it gained quite a lot of attention. There is a video of my speech here and the slides are attached at the bottom of this blog. Please take a look.

In my speech I set out the background - that Moore’s Law will continue, but the effects of it won’t give us as much of a frequency uplift and power reduction as we have been used to. We will have to use the additional transistors more cleverly, through the use of domain-specific processors and heterogeneous computing. We will make use of GPUs to perform the right sort of computation, and utilize coherency (as described in my colleague Bruce Mathewson’s blog), to reduce the cost of offload and increase the efficiency of heterogeneous computing. ARM is leading the industry here, and any company not doing this will simply get left behind in the race for best performance within an energy budget.

I also talked about standards and how they are so essential to abstract away some of the inherent complexity in today’s and future systems to allow the largest number of developers to make use of the capabilities we are providing for them. I explained how much we in the industry need that thriving ecosystem of developers, and how much we need to care for them and look after them. I also mentioned economics, not just technology but the economics of developers making money, and the economics of code reuse: how those developers need to be able to reuse code modules across numerous different platforms and the essential role that standards play in that.

As Jon Peddie pointed out, in the questions at the end, I failed to mention OpenCL in my words about standards (it was on the slides)! I tried to joke about this, but the serious point is that ARM is committed to OpenCL. As a supplier of GPU, CPU and fabric IP, we are in a unique position in the industry to offer the best implementation across ARMv7 CPUs and Mali-T604. With our hardware architecture, and (just as importantly) our software architecture, we are supporting all the GPU computing APIs that are needed by the industry to take best advantage of heterogeneous computing.

Like my talk? Hate it? Got any comments or questions? I’d be delighted to discuss it here…

Compute Power with Energy Efficiency, Jem Davies.pdf (1.65MB)
Number of downloads: 586

Jem is an ARM Fellow and likes to think of himself as "The Godfather" to technical talent in ARM. After spending some time in his youth writing software for satellites and traffic-lights among other fascinating things, Jem spotted the technical inflection point of the mobile industry: graphics, video and other visual computing. As VP of technology in the Media Processing Division of ARM, Jem is busy with a lot of projects involving the future of cool ARM technology, which will revolutionise how people experience and interact with digital devices.

Share This Entry:

All company and product names appearing in the ARM Blogs are trademarks and/or registered trademarks of ARM Limited per ARM’s official trademark list. All other product or service names mentioned herein are the trademarks of their respective owners.

Back to Multimedia →

6 Comments On This Entry

Please log in above to add a comment or register for an account

(2 Pages)
1
2
→

rightnottobeararms

27 June 2011 - 02:07 PM

First : the quality of the video posted on the AMD developers site is terrible . I mean don't they have better cameras, servers and bandwidth. I was expecting more from an IT hardware company. The videos are unwatchable. Second: Is ARM interested in supercomputing? Your expertize in GFlops/Watt might prove useful in building an exa supercomputer. I am sure you read the news about the Japanese new supercomputer. For sure a floor size computer can compute all that a pocket sized computer can and more

-1

Jem Davies

27 June 2011 - 02:19 PM

Whilst it's not our main focus at the moment, ARM is of course interested in super-computing.

If you look at my talk, I noted that the big problem to be solved in servers and supercomputers is that of power - more money is spent by customers on power and cooling than on the hardware itself. We think it would be entirely sensible to build GPU compute server farms with our GPU designs that use low power and dissipate less heat...

rightnottobeararms

27 June 2011 - 02:53 PM

Yes chip heating is a waste of electricity but given that the generated heat can be harness it can be fed back as electricity to the grid. So if the chips are liquid cooled and the heated liquid runs a turbine generator that generates electricity the generated heat is put to use. I did watch your presentation by curiosity. But I understand that you are focused on palm power.

Jem Davies

27 June 2011 - 04:22 PM

Unfortunately, the laws of thermodynamics make it inefficient to harvest low-grade waste heat energy - it is much more efficient not to generate it in the first place...

We are not just focussed on "palm power"; we aim at many segments. Our GPUs are present in a number of segments, including Digital TV and Set-Top Boxes, not just battery-powered devices.

Amit Dube

30 June 2011 - 06:43 AM

Heterogeneous computing seems to be an answer in short to medium term. As Moore's law predicted the density, computing power and frequency in last century, is it possible to predict about the Heterogeneous computing? I mean as task complexities addressed by dividing it in smaller subtasks (homogeneous with in group and heterogeneous between the groups; like CPU / GPU).In medium to long run, what will be the guidance for these domain specific solutions? Single SOC or Multiple Chips, as an expert can you comment on this?