Nvidia GK180 appears in Tesla K40

With 7.1 billion transistors and 561 mm² Nvidia GK110 is not only the world’s fastest graphics processor but also one of the most complex chips of all time. It is both expensive and difficult to manufacture, which is why only the Quadro K6000 contains the full version of the circuit. The rest are scaled-down variants with one or more SMX clusters disabled.

A data sheet is now being distributed with the specifications for what is said to be the upcoming Tesla K40 “Atlas” calculation card. This contains 2,880 CUDA cores, as many as complete GK110s, but here we are talking instead about the hitherto unknown circuit GK180.

The new calculation card will be able to deliver over 4.0 and 1.4 TFLOPS at single and double precision, respectively. Another novelty is a full 12 GB of memory, double up from the Tesla K20X, this with a bandwidth of 288 GB / s. In addition, the table indicates automatic overclocking via GPU Boost. Despite this, the TDP value remains at 235 W (245 W in the form factor SXM).

The figures thus do not point to any major differences compared to the predecessor, but are in line with what can be expected from an invented variant of the GK110. However, it remains to be seen if there may be more news than that.

Via: Techpowerup.