Supermicro GPU Workstation Hits 4 Teraflops

Super Micro Computer has begun shipping its latest high-performance workstation, the 7046GT-TRF SuperWorkstation, which features dual Intel Xeon 5500-series processors and up to four NVIDIA Tesla C1060 GPUs. With the full complement of GPUs in place and used as general instruction-processing units rather than graphical engines, the 7046GT-TRF is capable of performance reaching four teraflops.

August 14, 2009

2 Min Read
NetworkComputing logo in a gray background | NetworkComputing

Super Micro Computer has begun shipping its latest high-performance workstation, the 7046GT-TRF SuperWorkstation, which features dual Intel Xeon 5500-series processors and up to four NVIDIA Tesla C1060 GPUs. With the full complement of GPUs in place and used as general instruction-processing units rather than graphical engines, the 7046GT-TRF is capable of performance reaching four teraflops. Housed in a 4U rackmount-convertible tower chassis, the system supports up to 11 full-height expansion cards, eight 3.5" hot-swappable SAS/SATA drives and up to 96 GB of DDR RAM.

According to James Huang, product marketing manager for AMAX, a systems integrator based in Fremon, Calif., "Many applications, like fluid dynamics analysis, statistical analysis,computer-based animation, advanced financial analysis, and DNA decodingcan be processed in parallel algorithms." Since early customers for the 7046GT-TRF have been in highly technical fields, "This workstation is going to be huge for them," he says.

Huang compares the performance of the GPU-based SuperWorkstation to that of a 30-node cluster built using traditional CPUs. Customers are not only looking to replace time on costly supercomputers with this sort of advanced workstation, but hoping for alternatives to more common clusters and workstations. Huang says, "Most of the interest we've seen is driven by physical space, electrical power, and computer management considerations. Customers are comparing the cost of these systems to that of 100 teraflop data centers filled with traditional clusters."

One of the early customers for the 7046GT-TRF was the National Institutes of Health, and according to Huang, "the solutions we're putting in place include applications for medical, defense contractor, and university customers." However, he adds that, "Fortune 1000 companies are adopting this system as part of a seismic shift into parallel computing."

Because of the system's complexity, and the fact that virtually every customer is deploying the SuperWorkstation in a heterogeneous computing environment, Huang says that most purchasers are working with system integrators such as AMAX to put the computers into production.While Supermicro does not provide suggested retail prices, a spokesman for the company said that a fully-loaded system (4 C1060s, 48G DDR3, 2 Xeon 5530s, SATA HDDs) would sell for around $10,000. When compared to the cost of a 30-node cluster, even when blade-computing technology is taken into account, it's apparent that bargain-hunting high-performance computing users will be looking closely at GPU-based systems from Supermicro, AcceleWare, NVIDIA, Mercury Computer Systems and a growing number of competitors.

SUBSCRIBE TO OUR NEWSLETTER
Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like


More Insights