Why not Larrabee?

Posted: 08.01.08

Anyone not stuck in outer space or maximum security knows Intel is going to introduce a new chip code named Larrabee. At Siggraph they are going to reveal, after almost two years of teases and leaks, the architecture of the device.

It is not a GPU as many have mistakenly described it, but it can do most graphics functions, Intel says it can do all, we’ll have to wait for proof. Right now its slide-ware, but development systems are supposed to become available in November.

ATI and Nvidia will be very busy discrediting the device and pointing out its shortcomings. They should, given that Intel has all but ruined their share prices with disparaging comments about GPUs. Perhaps Intel needs to be reminded of some of its past triumphs; the Itanium and XScale come immediately to mind.

Nonetheless the question is, in my mind at least, is there room in the market for a third major player in discrete graphics?

What is Larrabee’s market potential?

If you look at the market development for discrete desktop GPUs over time your answer is probably no. In the late nineties when the market was just cresting 100 million units a year, the number of suppliers swelled to over 70. Today, it is approaching 400 million units a year; but the market has consolidated to four discrete graphics chips suppliers plus two integrated (only) suppliers. And of that population, two suppliers, ATI and Nvidia, own 98% of the discrete GPU business (which was 350 million units in 2007.) And, the trend line indicates a flattening to decline in the business as the red line shows in Figure 1. However, Intel is no light-weight start up, and to enter the market today a company has to have a major infrastructure, deep IP, and marketing prowess – Intel has all that and more. So yes, there is room for a third player if that player is Intel.

Figure 1: Shipments of GPUs to date

Assuming flat growth in discrete graphics chips, if Intel could reach parity with the incumbents that would give them 33% of 20 million units a quarter. Are seven million units a quarter worth the investment? Intel won’t be able to charge any more than ATI or Nvidia.

It’s a bigger market

However, you can’t assume flat growth in discrete GPUs. We recently increased the forecast on desktop discrete GPUs to take into consideration multi-AIBs (i.e., Crossfire and SLI), and two new developments: Hybrid configurations (which have a GPU and an IGP), and GPU compute. Clearly Intel sees an opportunity for Larrabee in GPU compute, and probably in Hybrid.

This revised forecast gives Intel a market potential of 46 million units in 2010, the first full year of shipments of Larrabee. Assume they can sell the chips for $100 that represents a market value of $4.6 billion and more if they build AIBs.

Now there’s a lot of “if” in that, not the least of which is that they can bite into ATI and Nvidia’s market share. But $4.6 billion is an admirable goal and can represent Intel’s TAM.

Intel has told me they intend to push Larrabee into graphics applications first, so the GPU compute portion of the market may not be realized immediately in their TAM calculation. However, they will come out with several versions of Larrabee (various number of cores, just as ATI and Nvidia do) and so the entire discrete spectrum is open to them.

Figure 2: Potential increase in GPU shipments due to new architectures and multi-AIBs

When is a GPU not a GPU?

One final note. GPUs. Larrabee is not a GPU in the sense an ATI, Nvidia, or S3 chip is a GPU. It is gang of X86 cores that can do graphics processing, so it is a GCPU – graphics capable processing unit, as are ATI, Nvidia, and S3’s chips. It’s unlikely the industry is going to take such subtleties into consideration and adopt a new term like GCPU and rather will incorrectly label and refer to Larrabee as a GPU.

Probably Intel will come up with a catchy mnemonic for Larrabee. It would be wise of them to do that to differentiate it and to drive home their point that existing graphics processing architectures “will be a footnote in the history books However, the industry, analysts, reporters, users, and investors have demonstrated too many times that they are lazy and will find it easier to simply call Larrabee a GPU. A GPU by any other color smellith as sweet….

Then and now

Intel is announcing a 2010 part now. Maybe that will influence some potential buyers to wait, certainly that would Intel’s ambition. However, do you think ATI and Nvidia are just going to sit on their hands till 2010 and wait for Larrabee to show up?

ATI and Nvidia are fleet-footed companies and can turn much faster than Intel. So if they chose to they could have a counter punch to Larrabee by 2010.

And lets not forget that ATI and Nvidia have been building hardware optimizations direct and OpenGL for the last 10 years, designs that are based on the APIs. Larrabee can’t do that and so there will be constraints in Larrabee which Intel is confident it can overcome in software. A lot can happen between now and 2010.