NVIDIA Reveals GK110 Specifications and Availability

Yesterday, we reported that NVIDIA revealed the GK110 GPU, or “Big Kepler”, as it has been dubbed, which is going to go in their K20 Tesla graphics card for professional applications (it will most assuredly have a professional price, too). From a technical perspective, we were able to give some idea of its performance, but not much else. However, NVIDIA have now revealed more specs of the device and when it will be available to gamers. It appears that the only parameter left out is the clock speed.

The GPU contains 7.1 billion transistors, 2880 CUDA cores, 15 SMX clusters, talking to up to a whopping 24GB GDDR5 over a 384-bit data bus, not the 512-bit that had been widely rumoured, which is sure to disappoint hardcore performance enthusiasts. This also means that the card will have an odd amount of memory, rather than a power of two like on the GTX 680. The GK110 compares with the GK104 GPU used in the current GTX 680, which has 3.5 billion transistors, 1536 CUDA cores, 8 SMX clusters and a 256-bit data bus, so the GK110 is basically a lot wider, but not twice as wide as its little brother, so we don’t expect it to deliver twice the performance at the same clocks, even though it has twice the transistors. Note that a bigger chip will have to run at lower clocks too, further reducing performance.

And availability? Eager gamers brace yourselves: it looks like it will be 6 to 9 months away, appearing in Q1 2013. Also, it may possibly be branded GeForce GTX 780. The price for this monster card should be interesting, too. Specs for the GK110 are below.

The die shot above is of the GK110 GPU and the card pictured is the K20 without its top cover. Note the lack of a rear connector panel as it’s not intended to be used for graphics, but GPU compute instead. Power is supplied via 6+8 pin connectors mounted at the back of the card.