The specifications of the MDGRAPE-3 have evolved since October 2003, when I wrote "Protein Explorer, the First Petaflop System." This system, which will be operational in 2006, will contain 6,144 processors instead of 5,120. And each chip, now running at 350 MHz, will have a peak speed of about 230 gigaflops, with 20 pipelines being able to compute 33 floating-point operations per cycle. This will bring the total power of the full system to an astounding 1.4 teraflops. But please note that this is a specialized system, designed exclusively for molecular dynamics calculations and simulations at the RIKEN Genomic Sciences Center, based in Yokohama, Japan. The whole machine will not be very big, with only 32 standard 19-inch racks. Each rack will contain 16 boards hosting 12 chips each. CNET News.com gives more details in "Japan designers shoot for supercomputer on a chip." Read more...
Here are some details on the chip itself.
Samples of the chip, which was designed for life sciences research, can now perform 230 gigaflops, or 230 billion operations per second, while running at 350MHz, better than standard general-purpose chips. In a worst-case scenario, the chip performs 160 gigaflops at 250MHz, said Makoto Tanji, a researcher with RIKEN's high-performance computing group. Tanji spoke at the Hot Chips conference taking place at Stanford University.
The computational power comes, he said, because the chip is specialized for workloads that involve numerous, similar calculations on a comparatively small set of data. This sort of workload is common in the life sciences and bio-nanotechnology field, where researchers need to examine, for example, how a single protein interacts with thousands of different molecules. Consequently, the chip and the computers based on it can be directly compared with general purpose supercomputers only in a limited field, but the processor excels there.
As I wrote above, the whole system is evolving. Below are two diagrams coming from a paper submitted during the SC2003 Conference, "Protein Explorer: A Petaflops Special-Purpose Computer System for Molecular Dynamics Simulations" (PDF format, 10 pages, 138 KB).
||Here is the new block diagram of the full MD-GRAPE system, with its 6,144 chips and its host, a 512-processor PC cluster. (Credit: RIKEN).|
||And here is the block diagram of each board containing 12 MDGRAPE-3 chips. (Credit: RIKEN).|
Michael Kanellos, the writer for CNET News.com, adds some interesting numbers about power consumption and pricing.
In terms of power consumption, the 350MHz MDGrape 3 consumes 14 watts of power, or 0.1 watts per gigaflop. A 3GHz Pentium 4 runs at 82 watts, or 14 watts per gigaflop, he said. The Blue Gene/L chip and Earth Simulator come in at 6 and 128 watts
The 350MHz Grape 3 can provide a gigaflop of computing power for $15, compared with $400 per gigaflop for a Pentium 4, $640 per gigaflop for the chips inside IBM's Blue Gene/L and a whopping $4,000 per gigaflop from NEC's Earth Simulator, currently the world's most powerful supercomputer.
If the above bumbers are correct, this means that the future teraflop machine will only consume 76 kilowatts for its processing units, excluding other elements, such as storage systems for example.
And with the prices quoted above, all this computing power would cost only about US$20 million, again not counting all the peripherals. This sounds very reasonable. If you think my numbers are wrong, please free to correct me.
For more information about this project, you can read the MDGRAPE-3 page (not completely up-to-date) from the GSC Bioinformatics Group at RIKEN.
Sources: Michael Kanellos, CNET News.com, August 24, 2004; and various websites