NVIDIA has had its GPU Technology Conference, where it made several big presentations. The company showed no less than two graphics cards in the Tesla segment, and one builds on the rumored GK100 (GK110?) with 7.1 billion transistors.
GPU Technology Conference is a platform for announcing new Quadro and Tesla solutions from NVIDIA, and to reach out to developers. The company presented a graphics card called K10 for HPC (High-Performance Computing) that looks very much like GTX 690. K10 builds on two GK104 circuits, and since we know that the focus is not on high generic performance, but double precision floating point calculations it is actually worse than last generation Tesla. GK104 is still good in single precision calculations, and these are the people NVIDIA target with K10.
 NVIDIA Tesla K10 |
 NVIDIA Tesla K20 |
The most interesting is instead K20 that use the coming GK100. Some calls it GK110, but with NVIDIA’s previous naming scemes GK100 sounds more logic. With K20 NVIDIA will deliver three times better double precision performance than previous generations of Tesla based on Fermi. It will also be the first graphics processor built for could computing.
The biggest graphics circuit ever?
The circuits swallows 7.1 billion transistors, and this could easily be the largest graphics circuit ever since NVIDIA’s GK104 with 3.5 billion transistors measures 294 mm². Judging from the picture above it will get 15 SMX clusters, which would kill rumors of both 2304 and 3072 CUDA cores. The only thing we know for certain is that GK100 will appear in Q4 in the form of K20, and that it should appear in retail about the same time.
No active posts found.