Analysis - The CEO of Nvidia, Jen-Hsun Huang, is a pugnacious guy who isn’t without charisma.
He doesn’t pull his punches and you wouldn’t want to get into a street fight with the guy. Earlier this year he’s said some pretty controversial things about Intel, for example.
But at a technical conference for financial analysts yesterday, he didn’t concentrate on dissing the opposition. Instead he proposed the most ambitious of strategies – to fundamentally change the nature of computing – and in the course of that to floor the Big Boy Intel. And he may have a point.
Huang said: “I think it’s fair to say that 2009 compounded by the recession is going to drive an inflexion point that we’ll all remember. The PC is just not sexy any more and in order to make it sexy and bring back excitement into the industry we have to revolutionize the architecture in some very important ways.”
That, fundamentally, uses what’s already a parallel processing semiconductor, the graphics chip, and harness it as a coprocessor that sits alongside the microprocessor. He said that the basic raw materials have advanced to a point where Nvidia can do some pretty revolutionary things with it.
“Our view of the computer architecture of the future is an idea that there is a sequential processor and a parallel processor co-processing. The original architecture of the microprocessor is no longer adequate. There are sensors everywhere collecting world data that is coming in in gigabytes, moving to terabytes. We need to make something of that data."
He said that Nvidia believes that the answer is a parallel architecture that doesn’t just have two or four cores but hundreds moving to thousands of cores. “If you’re willing to make the bet that you don’t want to replace the microprocessor you will create a co-processor and you can make the architecture of this new parallel processor completely unique.”
Nvidia has turned the graphics processing unit, which is a parallel processor, and has made it general purpose. He said that he believes that providing graphics only with these semiconductors “is a disservice to humanity”.
He said that in high end systems, the GPU spend is far higher than the CPU spend. “There’s a lot of evidence that a re-allocation is afoot. Every single computer ought to have a GPU in it. Our fundamental strategy is to bring GPU to the marketplace, revolutionize the market and re-allocate the total spend. The total market is about $26 billion.
“You compute for a long time and you vizualise at 60 hertz. We believe we can bring computation acceleration to that market. The GPU could be used for all kinds of simulations and the incredible speed up that people have seen. Imagine if it took a day to do a simulation and then it took an hour, it would change your life.”
Huang said that he hasn’t come across a single Powerpoint slide that he hasn’t loved. “Powerpoint slides are just beautiful and then you see the real baby and it’s the ugliest baby you’ve ever seen. GPU computing has reached the “tipping point”. We felt the GPU was our best shot at getting parallel computing to the world. Some people have tried to do it but they’ve all failed for one reason, because there’s no volume. If there’s no volume, there’s no apps. If there’s no R&D budget, how do you keep ahead?”
The real problem, he continued is software. Startups and others have experimented with parallel processors but the volumes were low, demand was low, and software engineers just sat on their hands and waited and didn’t write software. Huang said that everyone has just waited for the last 15 years, while Nvidia appeared at a time when the CPU hit the wall. The CPU is no longer scaleable, but, he claimed, the GPU is already a parallel processor, with 3D graphics the most extreme example of parallel processing.
Nvidia is willing to take the hit on its architecture because, Huang believes, it’s the best way to speed up real world applications.
Unfortunately for Huang, of course Intel is not just sitting on the sidelines waiting for its business to disappear. While most people positioned its Larrabee architecture originally as an alternative to ATI-AMD and Nvidia graphics processors, insiders at Intel believe that there’s much more to this multicore architecture than just graphics.
In fact, their ideas about Larrabee are not that dissimilar to Nvidia’s and while Huang may well be a pugnacious street fighter with a knock out punch, it’s not fair nor realistic to expect a bantam boxer to knock out a heavyweight in a 13 round contest.
But if anyone has the will to do it, that anyone is Jen-Hsun Huang and it’s too early to say that he may not prevail.