It seems we’ve been hearing a lot more about AMD’s next-generation GPU (Polaris) than NVIDIA’s Pascal recently. Pascal seems to be very illusive and details on the GPU are very hard to find. Well today we get some new information! A new academic paper has information on the actual theoretical compute performance of the Pascal GPU.
The paper, written by Manuel Ujaldon talks about the benefits of new memory technologies and their impact on the compute landscape. One of the slides actually shows the theoretical double and single precision performance numbers for the flagship Pascal GPU compared to other compute devices like Kepler, Fermi, Xeon Phi, and a Bulldozer-based Opteron. If we look at the chart we can see that the double precision numbers are just slightly above the single precision numbers for Kepler GK110, but if we take a look at the single precision numbers they are simple off the charts!
If we look at the source we can confirm that Manuel Ujaldon did talk to NVIDIA, so his number can’t be completely made up and were likely from early internal testing. Now of course these numbers don’t always translate into how a particular GPU performs in real-world tasks, but what we do know is that based on this Pascal is going to have a lot of compute power!