Palo Alto (CA) -Stanford is the next University that receives support from major IT companies to develop new techniques, tools, and training materials to exploit the parallelism capabilities of multi-core processors. And no, that headline is no mistake: The initial group of sponsors includes a colorful mix of rivals and partners: AMD, Nvidia, Sun Microsystems as well as Intel, Hewlett-Packard and IBM, showing first signs that these companies could actually be working together to solve the multi-core programming dilemma. Too good to be true?
Stanford’s new Pervasive Parallelism Lab (PPL) is good news for the software industry as research results coming out of this effort are likely to include programming tools in one way or the other, even if there are different interests involved. Nvidia, AMD and Sun are the founding members of this initiative and are apparently driving this effort. Not surprisingly, Nvidia touted its participation and it is obvious that Nvidia wants its Tesla stream processor cards to be part of this program - which, in its very basic concept, may be similar to AMD’s interests surrounding its Firestream stream processor cards. But AMD may have a different idea how its x86 CPUs can be exploited and Sun adds another completely different variable to that equation.
At that point we haven’t even mentioned the "affiliate member companies" yet, which include IBM, Intel and Hewlett-Packard. Add everything up and it is easy to see that the PPL, which includes a staff of 13 people initially, will have to cover a lot of ground: Nvidia’s GPU, AMD’s GPU, AMD’s and Intel’s CPU and most likely Intel’s massively parallel Larrabee accelerator card as well as vendor interests from IBM and HP. The research effort has received $6 million in funding and Stanford representatives said that they hope to come up with programming tools by 2012.
In general, PPL representatives said that they will work on a "top-to-bottom parallel computing system, stretching from fundamental hardware to new user-friendly programming languages" to enable developers to exploit parallelism automatically: A perfect research result would allow developers to simply drop their code into a system, which optimize their code for parallel processing. Initial research steps will include the development of a testbed called FARM (Flexible Architecture Research Machine). The technology is expected to be completed by the end of this summer and is described to be capable of combining "versatility with performance by blending reprogrammable chips with conventional processors."
If that in fact is possible and this technology can achieve decent performance results, FARM could be just what developers are waiting for. But at least for now we can’t help but to wonder how Intel ended up with Big Blue and HP in a train whose direction is apparently being determined by Nvidia, AMD and Sun. Intel has an entirely different idea how visual computing will look like in a few years down the road than, for example, Nvidia. Nvidia has recently voiced its concerns over Intel’s advances in a very public fashion and attacked Intel with cheap shots such as calling Intel’s Larrabee "Laughabee".
Perhaps it all will work out just fine and Stanford will come up with some real solutions for developers to tap into the power of multi-core engines, no matter if they are GPUs or CPUs. But we have no doubt that the research group, which by the way includes people like Vijay Pande, who is in charge of the Folding@Home project, will see challenges to address the interests of all member companies involved.
Intel, of course, has another and most likely much more capable horse in this race. A little over a month ago the company announced that it is sponsoring a multi-core programming initiative at UC Berkeley and the University of Illinois at Urbana-Champaign (UIUC). Together with Microsoft, Intel is pouring $20 million into a research initiative that also hopes to solve the problems developers are facing with multi-core processors today. The goal of this initiative is to create tools that can stimulate the development of multi-threaded software.
Back in March we believed that, given the enormous problem of a slow adoption of parallel programming, the $20 million investment was not enough and the PPL at Stanford certainly does not change our opinion. Companies such as Nvidia and Intel could be too conservative with their investment into such research initiatives: Think about the fact that companies such as Intel will depend on the availability of multi-core software within a few years and a $20 million investment that is placed to secure a future revenue base seems to be out of proportion with the multi-billion-dollar cost of developing a new microprocessor.
The PPL will be officially unveiled at Stanford on Friday at 2 pm PST.