Larrabee we (didn’t) know you well (enough)
Opinion On Friday afternoon, 1:30 p.m., Dec. 4, 2009, California, time Intel dropped a little surprise bomb on the PC industry. After three years of hoopla and massive investment the company was to cancel its Larrabee product introduction.
Everyone who posted to and/or got quoted on the web knew everything about Larrabee – and they knew nothing. Intel showed slides, ran demos and tests on stage, and the crowds either loved it or hated it, but they’d leave the presentation absolutely knowing what Intel was doing and was capable of. There’s an important lesson here for all – if you don’t trust your politicians, why do you trust web posts?
As soon as leaks about Larrabee began to surface in 2006 the PC industry went nuts. Intel lobbed spit balls at Nvidia and GPUs in general, Nvidia offered Intel a can of whoop ass, ATI kept turning around to say, what? huh..? And all three of them watched Wall Street dump shares hoping one of them would grow up.
Intel started to righten itself in the second half of 2009 and apply the management and engineering professionalism the company is known and respected for.
The stupid and distracting brickbats stopped, the meaningless and contrived demos about ray tracing stopped, and real stuff started to appear, like how to program the beast.
The company even acknowledged that maybe the GPU wasn’t dead and could find a place in the universe.
Still, the misspoken claims of when the many core GPU emulator chip would be ready lived on the web. The web is like cancer – it’s immortal, it never dies and it will cause you pain if you’re not careful. The unmanaged quick remarks made in 2008 in Shanghai by a senior executive from Intel about Larrabee being ready in 2009 glowed in the dark. The sad thing is the people on the ground, the engineers and planners actually, the folks I spoke with, who were actually working on the project never, not once, ever said anything but maybe 2H 2010, but they were just engineers, who listens to them or Dilbert?
Still the big shots seemed bent on some self destructive quest and embarrassed themselves with ill fated and hopelessly flawed demos of what was allegedly A0 silicon. Why? Who knows?
It took almost six months but in Portland a few weeks ago Intel showed its real colors and blasted the TFLOPS meter off the scale in a real benchmark. No theoretical calculations of what a processor might, could, do if the wind was just right.
The web cannot lie
Once again the web know-it-alls interpreted the event in whichever terms suited their agenda, and since they KNEW exactly what Intel was doing and planning, they told us – and that too lives on the web forever, and was repeated by sites that didn’t know, but wanted to be associated with the knowledge.
Meanwhile in Intel-land it was forth down and some tough decisions had to be made. Larrabee was working pretty good, but it was just part of a much bigger project and program, and all moving parts weren’t moving at the same speed.
Pause now and let me take you back in time. Intel has tried doing a graphics processor and a graphics chip in the past. The i860 which morphed into the i960, with a clever but underutilized barrel shifter, the disastrous i740, and mediocre IGPs left the company tainted as a can’t do graphics company. This time, top management and senior engineering folks said never again. It goes out right, it kicks ass, or it doesn’t go out – not negotiable.
For three very long days and nights managers, engineers, and even a lawyer or two held meetings, showed slides, played with spreadsheets and came to the difficult decision that "Larrabee silicon and software development are behind where we hoped to be at this point in the project," said Intel spokesperson Nick Knupffer. "As a result, our first Larrabee product will not be launched as a standalone discrete graphics product."
So instead of launching the chip in the consumer market, Intel will make it available as a software development platform for both internal and external developers. Those developers can use it to develop software that can run in high-performance computers.
Intel has made a hard decision and we think a correct one. Larrabee silicon was pretty much proven, and the demonstration at SC09 of measured performance hitting 1 TFLOPS (albeit with a little clock tweaking) was impressive but it was a computer measurement, not a graphics benchmark.
Intel’s next move will be to make Larrabee available as an HPC SKU software development platform for both internal and external developers. Those developers can use it to create software that can run in high-performance computers.
I think this makes a lot of sense, and leaves the door open for Intel to take a second run at the graphics processor market. The nexus of compute and visualization, something we discussed at Siggraph, is clearly upon us, and it’s too big and too important for Intel not to participate in all aspects of it.
Intel will let us know sometime in 2H10 what the next phase of Larrabee and their graphics ambitions will be. In the meantime ATI and Nvidia get a reprieve. But don’t count Intel, or their multi-core x86 based chip out, they’ll be back…