In a sea of system integrators that use off-the-shelf components, it is a rarity to see a company showing some real innovation. Colfax International was one of first companies in the world to launch a Tesla-based personal Supercomputer, but that was something we might call… a standard personal supercomputer.
Engineers at the company obviously weren't enthused in such a product, since it puts serious constraints in how many GPUs can you plug-in…"more is always better" in the HPC side of business and we were shown just that.
During the GPU Tech Conference 09 at Fairmont Hotel in San Jose [CA], Colfax International demonstrated world's first GPU server featuring no less than eight Tesla C1060 cGPU cards. But the eight Tesla boards are not the big news here… Colfax created a motherboard featuring eight full-width PCIe Gen2 x16 slots. This means that each cGPU can potentially get full bandwidth.
Here it is - 8 Tesla cGPUs for HPC goodness - 8TFLOPS [at least] for supercomputing calculations
According to the information we got in hand by Mike Fay, VP of Sales, the system is powered by two Intel Xeon 5500 processors [up to the top-of-the-line W5590, 3.33 GHz ones], it can get up to 144GB of system memory [DDR3-1066 or DDR3-1333, depending on module capacity]. Using standard-sized 2GB or 4GB modules, you should be able to have 72GB of DDR3-1333 memory.
Zooming in discovers 72-144GB of DDR3 system memory, 8 Tesla C1060 cards and many more
When it comes to the GPUs, eight 55nm Tesla C1060 boards will get you 8.8 TFLOPS [the 55nm Tesla are a bit higher clocked than old 65nm C1060's, so you get 1.05-1.1 TFLOPS per board]. When it comes to computing power of future Tesla cards, worst-case scenario is 4.99 TFLOPS of IEEE 754-2008 spec dual-precision power, i.e. 9.98 TFLOPS of single-precision power [note: we got this info ourselves, not from Mike].
This bomb of a High Performance Computing goodness is powered by two to three 1200W power supplies, with the 3rd power supply acting as a redundant backup.
No display outputs at the back... this is a pure computational system
The estimated usage for such a supercomputer is oil and finance research, but with companies such as Royal Dutch Shell and Bank of America rolling on the cGPU bandwagon big time, the time is coming for eight-GPU configurations. We have a few ideas of our own, featuring Quadro FX boards. But we'll leave that for another time.
© 2009 - 2010 Bright Side Of News*, All rights reserved.
