Mark Gilbert’s Blog

Well at this present point in time I would say that’s unlikely, but in the future, well who knows. Nvidia CEO Jen-Hsun Huang called Intel’s integrated graphics offerings “a joke”, which may be true from a certain perspective. From another angle though Intel have done a pretty good job so far of sewing up the laptop graphics market, only a few niche items really contain Nvidia or AMD chips, the integrated graphics on desktops shows a similar picture too. Surely however when it comes in discrete graphics cards and chips Nvidia is top dog? Well yes, currently, although that may soon change and I’ll tell you why, Intel’s Larrabee graphics chips.

Intel intend to take a new approach to graphics, with the Larrabee chips being based off the x86 instruction set like most CPUs rather than custom graphics oriented instruction sets like most current GPUs. Intel senior vice president Pat Gelsinger said at the Intel Developer Forum last week in Shanghai,

“First, graphics that we have all come to know and love today, I have news for you. It’s coming to an end. Our multi-decade old 3D graphics rendering architecture that’s based on a rasterization approach is no longer scalable and suitable for the demands of the future.”

Early schematics for the chips suggest they are have 16 cores capable of operating in excess of 2GHz, but the number of cores can easily scale up to the thousands. This high number of general purpose cores fites very nicely with Intel’s current Tera-Scale project. The vision of computing and architecture that Intel sees in the future seems to consist of a high number of general purpose cores, which can be used for different tasks and reallocated on demand. For example instead of having a dual core CPU and a dual core GPU, Intel envisage a quad core general purpose chip that performs the tasks of both. Not running a game? Then all four cores can be dedicated to general processing tasks. Doing a lot of 3D rendering? Well more cores can be dedicated to the graphics processing. Work from that basis and scale up, which Intel is already doing, as they have demonstrated a prototype 80 core chip that can perform 1 TeraFLOP, hence the name of the project, while still exhibiting a TDP of 62W. In comparison the first system to achieve 1 TeraFLOP was ASCI Red. It achieved this in 1996 and used nearly 10,000 Pentium Pro processors running at 200MHz and consumed 500kW of power plus an additional 500kW just to cool the room that housed the beast.

I think it is this that Nvidia are really worried about. Not that Intel will develop a better Discrete GPU than them, because that would take a lot of effort, but more that their role purely as a manufacturer of discrete GPUs will become more and more marginalised over time.

Huang was quoted as saying that

“if Intel manages increase graphics performance by ten times by 2010, that’s barely up to par with current Nvidia offerings”

Although with Nvidia spending $1 Billion on R&D and Intel spending $6 Billion on R&D, Intel may be able to do just that and more.

Personally I think the Intel approach will win out, multi-cored, multi-purpose chips seems like the way forward to me.

Cheers