Intel Has a Luxury of Releasing Larrabee in Its Own Time – Larrabee Engineer

Intel’s Larrabee Set to “Come Back with a Vengeance”

by Anton Shilov
03/12/2010 | 09:53 AM

When cancelling the release of the code-named Larrabee graphics processing unit (GPU) that was due in 2010, Intel Corp. stressed that it would not give up hope to enter the market of discrete graphics chips. According to an industry insider close to the project, there is still a roadmap in place and that the Larrabee will “come back with a vengeance”.

 

“Intel is not a one-shot company. But there is a whole product plan and roadmap still in place,” an anonymous Larrabee engineer told TechEye web-site.

According to the hardware designer, if Larrabee had emerged on time, it would have change the whole industry. Larrabee was supposed to render graphics, feature extremely flexible programmability and retain compatibility with x86 instruction set so to allow developers to easily use many-core Larrabee for multi-threaded non-graphics applications. Nvidia Corp.’s upcoming Fermi family of GPUs also promise to be much more programmable than today’s graphics chips, but it is not x86-compatible.

Even though both ATI, graphics business unit of Advanced Micro Devices, and Nvidia now have time to create exceptionally flexible graphics and data processing architectures that may rival Intel’s GPU in terms of programmability, the Larrabee engineer tends to be optimistic: he claims that Intel may release Larrabee when it wants to, not when the market might need it.

“Intel has the luxury of being able to release stuff in its own time, there are not too many other companies that can do that,” the industry insider is reported to have said.

The optimism towards Larrabee does not seem to be reasoned. Both ATI and Nvidia have roadmaps in place and the industry knows what to expect from these two companies. Meanwhile, the world’s largest chipmaker has already cancelled at least one Larrabee project and the launch timeframe of another Larrabee is not clear. Even though select software developers are supposed to get the cancelled Larrabee chips and tools to experiment with many-core programming, it will still not be obvious for them what to expect from the new Larrabee and whether there is a reason to develop for it.