I’ve run the first realistic tests of the kind of computation that I am planning to perform on my new machine with the GPU “supercomputer” card. Here is a “before” picture:
And now, the exact same program running on the GPU:
I’d say that’s quite an improvement. To say the least.
The calculation in this case computed the self-gravitational forces in a cluster of 10,000 stars… it seems that the GPU can perform this computation at least 20 times a second. That’s quite remarkable.