Larrabee doesn't scare me
In a rather scathing attack on Larrabee, NVIDIA CEO, Jen-Hsun Huang, dismissed Intel's efforts to become a serious discrete graphics player using multi-core parallel computing, noting "I don't think a PowerPoint slide of any eloquence will replace the experience we have already created."
When asked about Larrabee at NVIDIA's recent GTC event, Huang joked "which chip is this?" before launching into a diatribe about Intel's chances in the graphics market.
"The fact of the matter is that Larrabee has launched and launched and launched to the point where even you're not interested anymore," Huang told a roomful of journalists.
Different things
"Designing a CPU and designing a GPU is a remarkably different thing," he continued, noting that one was all about locality of content in cache in very high frequencies and speculative execution. "The key word is speculation, so that they can have more parallelism," said Huang, adding "we don't speculate anything. We don't speculate at all."
The NVIDIA chief went on to explain that because his firm's technology was latency tolerant, it could simply use another thread "out of the 15,000 other threads we're executing on at any other time, instead of just one or two threads. We just keep running."
"The two architectures are radically different, the two approaches are radically different, the problems faced are dramatically different," continued an unrelenting Huang.
"In our business, you saw in one year, in one generation, double precision performance increase by a factor of eight," declared Huang, who went on to say that another way of putting this would be to say that if a company was late to market, it would lose the performance battle by a factor of eight in just one year. Intel, he said, had now slipped "by multiple years."
