Clearspeed: 1 TFLOPS in 16u

Print Friendly, PDF & Email

ClearSpeed is talking from the IDF this week about it’s dual-core Intel machine accelerated with a pack of e620 cards:

With each ClearSpeed Advance e620 accelerator delivering 80 GigaFLOPS peak double precision performance and over one GigaFLOP per watt of sustained LINPACK performance, the entire cluster had a maximum power consumption of less than 7KW and completed the benchmark in just 14 minutes, half the time required by the non-accelerated system. The energy used to achieve this TeraFLOP performance was approximately 1.5KWh, costing a mere 15 cents assuming a cost of 10 cents per KWh.

The whole thing took 16u of rackspace. For perspective

To underscore what a difference a decade makes, in 1997 Intel impressed the industry with the world’s first-ever terascale system known as ASCI Red that was delivered to Sandia National Laboratories. With 4,510 compute nodes the then state-of-the-art and record-breaking system occupied 2,500 square feet, consumed 850KW and cost $55 million.

Comments

  1. clearspeed is very nice, but is it keeping up with the market? for instance, the current gen of normal cpu chips have 4 cores, each delivering up to 4 flops/cycle at 2-3 GHz. true, they’re more like half a gflop/watt, but that’s not a massive, paradigm-shifting difference.

    in 16U, it seems to me that I could put 16 dual-socket servers with quad-core core2 chips at 2.6 GHz or better. that’s over a teraflop, no? I’d expect it to dissipate around 6KW.