Is high performance computing (HPC) hot? If this week’s Supercomputing 08 show is any measure, the answer is yes. In a down economy, one week before a major American holiday, the show still drew 10,000 people to Austin, Texas. That’s more than Microsoft’s recent PDC and WinHEC developer shows drew combined.
The annual show highlights all the latest discoveries and new products in high performance computing. It was where the most recent Top 500 list of supercomputers was unveiled earlier this week as well.
Two of the biggest newsmakers from the show are two firms not normally associated with the HPC market, but making increasing waves: Dell and nVidia.
Dell (NASDAQ: DELL) is based in Austin, so CEO Michael Dell didn’t have to drive very far to appear as a keynote speaker. Dell spoke of what he called the fourth wave of supercomputing built on commodity hardware and affordable to the masses thanks to falling prices.
He announced a supercomputer his firm has built with the Lawrence Livermore National Laboratory and several other companies called “Hyperion,” a 98 teraflop
Just as you can build your own desktop or laptop, you can build your own HPC system from a Dell Web site. The firm ranked third among vendors on the Top 500 list, with 19 computers on the listing.
Dr. Reza Rooholamini, director of Dell’s Clustering and Solutions Group, reckons there might be more that could make the list but the firms that owned them did not wish to submit their names to the list.
“Dell is a serious player. They have really ramped up their organization for the HPC market,” said Steve Conway, an analyst for IDC who specializes in the HPC market. “They are really playing the standard solution game very heavily.”
Rooholamini said Dell has its own way of competing against HP (NYSE: HPQ) and IBM (NYSE: IBM), which dominate the supercomputing market: simplicity.
“The goal for us has been to make this way of building and delivering these supercomputers industry accepted and industry practiced,” he told InternetNews.com. “We never set off to have any technological differentiation. The way we deliver is different, we give customers all they need or what they need to actually build one of these themselves. I think these are all differentiators.”
Dell builds and certifies the supercomputer and applications that will run on it independently and then sends it to the customer site, where it can be quickly assembled, said Rooholamini, rather than delivering a lot of boxes and a bunch of engineers to build it on site. This allows for faster deployment, which customers care about, he said.
Next page: nVidia graduates from games to HPC
Page 2 of 2
nVidia graduates from games to HPC
As part of his speech, Michael Dell talked up the new Tesla Personal Supercomputer, introduced at the show. It’s a desktop PC with four GPU processors in it that can generate four teraflops of performance. Just two years ago that would get it on the Top 500 list.
A high-end, dual processor, quad-core Xeon workstation can manage, at best, 192 gigaflops, according to Samit Gupta, senior product manager for Tesla at nVidia (NASDAQ: NVDA). The price is $9,995, fairly cheap for that level of processing power. In September, Cray released a “mass market” HPC server machine. The CX1, offered 786 gigaflops of performance for about $25,000.
Each board has 4GB of memory on it, a way of compensating for Amdahl’s Law, which says there are limits to parallelism because connections can slow things down. In this case, that would be the PCI Express bus. Fast though it may be, it’s no substitute for a direct CPU to memory connection on the motherboard. Much of the computing work is done in that memory to reduce traffic over the PCI bus.
“This product category democratizes supercomputing,” Gupta told InternetNews.com, a statement Dell also made during his speech to introduce the Tesla computer. “When the PC came out, it brought computing to all of us,” said Gupta. “In the same way, this supercomputer makes the cluster available to everyone.”
It does so by giving researchers and researchers and academia a multi teraflop supercomputer for under $10,000. Gupta said nVidia figures there are about five million researchers, academics and scientists in the U.S. who could use one and 15 to 20 million total worldwide.
Conway said Tesla augments x86 processors, but it does not replace them. “That’s the battle that’s going on. What you’re seeing is the very beginnings of the hybridization of the die. nVidia might add x86-like features to what it does and Intel and AMD are planning that with the GPU part,” he said.
Gupta doesn’t think AMD’s Fusion or Intel’s Havendal efforts will gain traction at the high end. “The problem with fusion products is that at the end of the day, the size of the silicon is limited,” he said. “So you can devote all of it to make a CPU, or all of it to make a GPU, or split it over the die. They will end up with a mid-range CPU which will address some of the market, but not the high end of the market. The high end of the market wants the biggest, baddest CPU and GPU it can get.”
Tesla Personal Supercomputers are available now from Dell, Lenovo and a variety of system builders in specific vertical markets.
