Intel told Ars today that its long-delayed Larrabee discrete graphics product has suffered yet another delay, so the company has had to "reset" its overall GPU strategy and reposition plans and its expectations for the first version of the Larrabee product.
Specifically, Larrabee v1 is so delayed that, at the time it eventually launches, it just won't be competitive as a discrete graphics part, so Intel plans to wring some value out of it by putting it out as a test-bed for doing multicore graphics and supercomputing development. Intel will eventually put out a GPU, but may not be the one we've been calling "Larrabee" for the past few years.
If the fact that Larrabee is "launching" not as a GPU, but as a kind of multicore graphics demo unit, sounds like a cancellation to you, that's because it kinda sorta is. It's not a cancellation in the sense that Intel is throwing in the towel on discrete graphics, because that's definitely not the case. The company reiterated that it still plans to launch a many-core discrete graphics product, but won't be saying anything about that future product until sometime next year. Whatever it is, it won't be the hardware/software combination that it previously announced, and that we described in our coverage of Intel's big August 2008 Larrabee reveal. It will be something else, and Intel wouldn't even characterize the relationship of that future something to the current Larrabee product.
The main issue behind the delay, it appears, was the hardware. That's not surprising, because Larrabee is a big, complex part, and it's quite a departure from anything that Intel has done. The hardware delay would have resulted in a software delay, and if Intel were to launch Larrabee with an immature software stack then it would be roadkill as a GPU.
Even though Intel couldn't have the Larrabee software ready on a timeframe that would make it competitive with NVIDIA and ATI (again, Larrabee is really a hardware/software hybrid GPU), the chipmaker can still push out the hardware itself and let others have a go at using it for graphics and HPC. Hence the plan to release it as a development platform for doing multi- and many-core research for HPC and graphics.
The Larrabee delay is obviously great news for NVIDIA, and even better news for AMD. This leaves both competitors to share the market for discrete GPUs, and quite possibly next-gen game consoles, for the next two years. NVIDIA's underlying long-term problems (the death of its integrated graphics processor market and the ongoing decline in the high-end discrete GPU market) are still there. And AMD still has well-documented challenges of its own as it struggles to regroup and return to growth after a brutally punishing few quarters of layoffs and cost-cutting. But execs at both companies have got to be high-fiving each other right now.
As for Intel's long-term future in the discrete GPU market, we'll have to wait until next year before we know more. Thankfully, that's just around the corner.
Source: Arstech
