Nvidia announces Tesla K20 with GK110 – 7.1 billion transistors

When Nvidia launched the Geforce GTX 680, it was no secret that the GK104 graphics processor was not, in fact, the “real” Kepler. Instead, it was a scaled-down model adapted to the broad consumer market. Pretty soon it became clear that the full-fledged circuit GK110 would not appear until towards the end of the year.

[quote][…] the largest, most powerful GPU we’ve ever built, which is much more powerful than Kepler.[/quote]

At this year’s GTC, Nvidia’s CEO Jen-Hsun Huang now presents two new calculation cards in the Tesla series based on circuits with Kepler architecture. Initially, the Tesla K10 is launched, which in practice is a Geforce GTX 690 adapted for GPGPU calculations, which means, among other things, lower clock frequencies and that the video outputs disappear in favor of better ventilation. So far no surprises.

However, the second model makes anyone prick up their ears. The upcoming Tesla K20 is no ordinary Kepler but contains a new graphics processor with as many as 7.1 billion transistors – twice as much as the GK104.

According to Jen-Hsun Huang, the Tesla K20 is based on Nvidia’s largest and most powerful graphics processor ever. However, no exact specifications are revealed, other than that the model will have three times higher floating point performance (double precision) than its predecessor Fermi.

Previous information claims that GK110 can contain up to 3,072 calculation units and a 384- or 512-bit memory bus. This without exceeding the maximum power output of the PCI Express standard of 300 W.

However, Nvidia will not start selling the Tesla K20 until the fourth quarter of this year. It thus remains to be seen whether exactly what the new circuit contains for news and whether the same model will also enter the consumer market as a new Geforce.