Sign up for our newsletter and get the latest HPC news and analysis.

NVIDIA's Tesla: 2 TFLOPS on your desk

nVidia logoEarlier this year NVIDIA had several announcements (hardware and the CUDA GPGPU programming framework 1, 2) around its GPGPU strategy as it tried to stave off an assualt by the AMD/ATI combo and it’s Fusion project.

Now they’ve followed up with new Tesla GPU cards and integrated systems. The Tesla GPU is an output-less video card with 128 processors. The company puts the cards performance at about half a teraFLOPS.

Tesla comes as a PCIe x16 card that consumes two PCIe power connectors and 170 watts at max. You can also get Tesla-powered workstations in 1 and 2 TFLOPS configurations sporting 2 and 4 Tesla cards each. The card will run you $1,499 while the workstations go for $7,500 and $12,000 respectively.

Not exactly loose change, but I remember my first TFLOPS system; it cost $10M.

There is good coverage of NVIDIA’s announcement and strategy at DailyTech and PCWorld. (Thanks to reader Paul Adams for the pointer on the DailyTech article.) NVIDIA’s press release is here.

Trackbacks

  1. […] From HPCwire we learn that NVIDIA has released version 1.0 of its CUDA GPU code development kit and C compiler. The release complements the company’s recent Tesla TFLOPS workstation annnouncement. CUDA 1.0 includes new C compiler optimizations and performance enhancements along with additional functionality and C code examples. CUDA BLAS and FFT libraries have been further optimized and include additional functionality and new C code examples relevant to areas such as computational finance and medical imaging are installed with the SDK. In addition, the CPU interface code to the GPU has been enhanced allowing for asynchronous launch calls and asynchronous device to device memory copies. […]

Resource Links: