If software and emulation is so great, why hasn’t Intel fixed its integrated graphics? asks Jen Hsung Huang
Nvidia’s often outspoken CEO, Jen Hsung Huang, has never been a particularly big believer in Intel’s Larrabee GPU, and even once dubbed it 'Laughabee' in a particularly vicious bout of Dad-humour. However, now that Intel has revealed how Larrabee will work, the head of Nvidia is apparently now even less convinced that Intel can make an impact in the graphics business.
In an interview with professional muck-stirrer Fudo at Fudzilla, Jen Hsung Huang said that ‘there are lots of questions that we are looking forward to seeing answers to from Intel.’ He then reeled off a list of questions about Intel’s forthcoming graphics chip, pointing out that ‘Emulation depends heavily on software. If it was so easy, why doesn’t Intel focus this expertise to fix the countless incompatibilities of their current integrated graphics? After all, it’s just software, right?’
‘Emulation is simply inefficient,’ said Jen Hsung Huang, adding that ‘graphics is the most computer intensive application we know. This is the green era, the era of efficiency. Why wouldn’t Intel focus on extreme power efficiency like their CPUs?’ As well as this, he also asked, ‘why is x86 desirable if Larrabee is not binary compatible with volume PC applications? Isn’t binary compatibility the most important attribute of x86?’
Interestingly, Nvidia’s CEO doesn’t see Intel’s developments in ray tracing as a threat to Nvidia either, confidently claiming that ‘Nvidia is the world leader of ray tracing solutions’ following its purchase of Mental Images, and its subsequent demo of real-time ray tracing in CUDA at Siggraph.
Although a lot of Larrabee’s architecture is still under wraps at the moment, Intel’s decision to focus on software rendering with x86 processors was very unorthodox. However, in a recent interview with Custom PC, Intel’s senior performance analyst Francois Piednoel, explained that ‘Larrabee is not like a regular CPU. It’s based on x86 instructions, but it’s parallel. Without saying too much, you will get the benefit of x86, but you’ll have much more than this. We’re not stupid; we are designing something that will be a very good generic graphics card as well.’