At ISC10, a new Top500 list was published and showed one major upset: China’s “Nebulae” supercomputer unseated long-time Top500 member Roadrunner for the #2 spot.  The #1 spot remains the Cray “Jaguar” at Oak Ridge National Labs, with an impressive 1.7 PetaFlops of performance, but China has made an impressive showing.

The Nebulae computer clocked in at 1.27 Petaflops by combining Quad-Core Intel Xeon processors with NVidia Tesla C2050’s.  While I can’t find concrete numbers anywhere, piecing together various news articles shows:

  • 120640 Cores (Straight from the Top500 List)
  • of Quad-core Intel X5650 Processors
  • 4640 Tesla GPU’s (from EETimes)
  • Which comes to 1:6.5 ratio of GPU’s to CPU’s

While the .5 probably means they have some nodes with Tesla’s and some without, it’s a surprising density.

Update 6/1/10: A reader pointed out that, unbeknownst to me, the Top500 list includes the count of GPU cores in the total core count.  From his website, he theorizes:

Based on all this information, it is almost certain that Nebulae is built on 4640 nodes, where each node has two X5650 processors and one C2050 GPU, for a total of 9280 processors and 4640 GPUs:

4640 nodes * (12 processor cores + 14 SIMD units) = 120640 cores

Why so many?  This comment from the EEtimes sums it up nicely:

The graphics processors are also relatively power efficient. The Nebulae system which uses 4,640 Tesla chips, consumes about 2.55 megawatts compared to about 7 MW for Jaguar.

So it made #2 on the Top500 list with only 1/3rd of the power.  But that’s not all, China has another system in the Top10, the Tianhe-1 which couples Intel processors with ATI Radeon 4870’s at an impressive 1:1 ratio (2 CPU’s and 2 GPU’s in each node).  Granted, they have only half the nodes of Nebulae at 71680 cores ( = 17920 CPU’s, or 8960 nodes) but that’s enough to put them in the Top 10.  And China isn’t stopping there (from the Daily Pioneer):

The super computer named “Xingyun”, has been developed in Tianjin, and works at double the speed of “Tianhe-1”, the previous fastest machine in China.

The Tianhe-1 was developed by the National University of Defence Technology in October 2009, Li Jun, president of the Dawning Information Industry Co. Ltd., was quoted as saying by Xinhua.

“Its peak performance reaches nearly three quadrillion calculations per second, three times the peak speed of Tianhe-1,” Li said.

This new machine is expected to be delivered by the end of 2010, so China may have another upset in the next Top500 if ORNL doesn’t step up the game with Jaguar.

If you want to see the list, but don’t want to dig through the text, then check out this neat interactive Treemap that BBC published online.

Tags