It permits single machines to perform complex analysis and visualization at the same time

Aug 7, 2012 14:18 GMT  ·  By

Seeing that Advanced Micro Devices has revealed new workstation series of graphics controllers, and some FirePro APUs to boot, NVIDIA formally introduced the Maximus platform with a similar purpose.

This product launch is a curious thing. Coming more than half a year after the initial announcement, it does not include performance figures of any sort.

That is to say, while the specs of the Quadro K5000 are available, NVIDIA is not ready to ship the product and has no benchmarks to show either.

Still, we cannot, in good conscience, ignore the press release, even if it is just a paper launch.

Quadro K5000 is powered by the Kepler architecture and has 1,536 CUDA cores, 4 GB of GDDR5 VRAM, a memory interface of 256 bits and a bandwidth of 173 GB/s.

The single precision compute performance is, thus, of 2.1 Teraflops, all at a power consumption of 122 W (18W in idle).

As for connectivity, NVIDIA chose two dual-link DVI ports, a couple of DisplayPort 1.2 outputs and four digital outputs.

"With the parallel processing capabilities enabled by NVIDIA Maximus systems, we can now be 10 times more creative," said Alan Barrington, a designer at the Mercedes-Benz Advanced Design Center California.

"With the NVIDIA Maximus-powered environment, we can continue to refine and improve our design, right up to the last minute. We can stay efficient and multitask. We no longer have to settle for less or to compromise on our creativity."

By multitasking, NVIDIA means to say that systems equipped with the Maximus can perform complex analysis and visualizations at the same time. Bindless Textures (1 million textures can be referenced directly in the memory) and FXAA/TXAA film-style anti-aliasing technologies help in that regard.

That said, there is a second product that NVIDIA has revealed: Tesla K20. We only know that it is designed with the GK110 GPU and that it offers SMX streaming multiprocessor technology (3X performance per watt improvement) and Dynamic Parallelism and Hyper-Q GPU technologies (simpler parallel programming and higher performance).

NVIDIA Quadro K5000 and Tesla K20 ($3,199 / 2,572 Euro) will sell as part of workstations starting in December. The former will start shipping standalone as well, in October, for $2,249 / 1,808 Euro.

Photo Gallery (3 Images)

NVIDIA Quadro K5000
NVIDIA Quadro K5000NVIDIA Quadro K5000
Open gallery