Posts tagged with General Purpose GPU

What is GPGPU computing and why all the fuss now?Ok, so what is GPGPU? It stands for general purpose, graphics processing unit. What a mouthful! In simple terms, it means using the GPU for things other than graphics. So, let’s go back in history a bit and see why we are even discussing this.

In the old days, the only processor in the PC that could do floating point math was the CPU (forget about co-processors for now). Floating point is critical for more precise computations. Also, the CPU was designed to undertake general purpose math; it could be used or programmed for just about anything. At this time, 3D hardware used fixed point, not floating point math for all the 2D and 3D workload. These GPU’s were also very specifically “hard-wired” for graphics processing

Then, in 2001, with the advent of Microsoft’s DirectX 9, or more commonly know as just DX9, things started to change. DX9 started the revolution towards floating point math on the GPU. All the “shaders” or “mini-cores” on a GPU were becoming more programmable. From this point on, all modern GPU’s were capable, at least in theory, of being used for more than fixed function graphics processing.

Fast forward to 2008 and it’s all about the magnitude of these new GPU’s. With 800 parallel shading or processing units in one ATI Radeon™ HD 4800 series graphics chip, you have huge compute power. That’s up to 1.2 trillion floating point calculations per second! So, we now have one huge, general purpose compute capable GPU. In some ways it is more powerful than the most up to date quad core CPU. It is also massively parallel or “multi-core” with up to 800 of these processing units, each capable of executing at the same time.

It took some time, but AMD (thanks to its acquisition of ATI) was a forerunner in this field with research projects going back over 5 years; to see what else could be efficiently and effectively computed on this growing number of processing units. With the launch of the R520 GPU in September 2005, ATI successfully demonstrated the first applications of GPGPU compute to the world. But it is not all that easy; most software is written to execute serially, not in parallel. AMD has been working hard to ensure that the majority of today’s processes or applications can work on first dual core, now quad core and say up to 8, 16 or 32 cores (CPU cores or GPU cores) in the next few years. In fact, all the key companies in this area are working very hard with internal and external groups to find better ways to program both CPU and GPU cores in this new world of massively parallel compute opportunity.

And let’s not forget cost and accessibility from the equation. In 1996, an Intel Pentium Pro-based computer with 10,000 processors, the size of a large room, delivered 1 TeraFLOPS and consumed 500 Kilowatts of power. Today, the ATI Radeon™ HD 4870 X2 delivers 2.4 TeraFLOPS, costs less than $600 and only consumes 285 watts of power. This just makes it all so much more accessible.

So, in simple terms, over the last 10 – 15 years, as the GPU has evolved from low performance, single core, fixed point math capable to massively parallel, fully floating point math enabled with up to 64-bit precision. It is now an intensely powerful (but hugely underutilized) piece of computer kit and is sitting inside millions of PCs. It’s just begging to be used, not just when playing the latest intense 3D games, but for all the other massively parallel tasks we do each day.

But, there is one final part to the questions I posed at the beginning – why all the fuss now?

Well, as I have said, the opportunity has arrived now with performance, availability and price coming together. But there is another side too. It also comes down to competition. In my opinion, Nvidia and AMD are locked in a brutal battle for supremacy in the traditional GPU market place. The performance crown has gone back and forth between the two companies over the years. Today, ATI is seen as the leader again, with the whole range of ATI Radeon HD 4000 series GPU’s. With this kind of battle raging in the market, all potential weapons, including new uses for GPU’s are fair game to create buzz and interest. Nvidia’s strong marketing around GPGPU is probably part of this. So, that’s my thoughts on what GPGPU is and why it is legitimately important and finally why it is so talked about today.

Come back soon for more on this subject!

Paul Ayscough was Director of Advanced Marketing at AMD and an evangelist for graphics and gaming.  His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.