CHIP DESIGNER Nvidia has announced the Quadro K5000 workstation video card, the first board featuring its Tesla K20 GPU.
Nvidia's Kepler architecture has for the most part been limited to the firm's high-end consumer parts and its single precision floating point K10 Tesla GPGPU boards. However the firm has finally shown off its K20 part in the Quadro K5000 workstation board.
Nvidia has given the Quadro K5000 impressive specifications with 1,536 cores, citing 2.1 teraFLOPS of single precision performance. The firm has also bunged in 4GB of GDDR5 memory on a 256-bit bus with total memory bandwidth of 173GB/s.
While Nvidia is pitching the Quadro K5000 to a workstation crowd and DirectX 11 support is there, OpenGL 4.3 support along with Shader Model 5, CUDA and OpenCL support are far more important. The firm said the Quadro K5000 card will come with two dual-link DVI outputs and two Displayport 1.2 outputs.
Nvidia said the Quadro K5000 uses up to 122W though it idles at 18W, both of which are impressive figures for such a graphics card and should sit well with workstation builders that have hard power usage limits for components.
Since Nvidia didn't mention double precision floating point performance it is likely that the Quadro K5000's performance will be optimised, or crippled, for single precision workstation workloads rather than high-resolution GPGPU workloads. Nvidia has already said that the Tesla K20 cards will push double precision floating point performance, and given that the Quadro K5000 and Tesla K20 share the same GPU one would expect that Nvidia wants to do some 'market partitioning' with its products.
Nvidia said its Quadro K5000 will appear in pre-built workstations in December, but added that the card will also be available as a discrete unit from October.
Sumit Gupta, GM of Nvidia's HPC business unit told The INQUIRER back at ISC that the Tesla K20 would be out before November, and while that might be so for some select customers, Nvidia is now saying that Tesla K20 cards will appear in December. µ