I recently became aware of two technologies which I highly doubt I will have the opportunity to test due to monetary constraints but are very cool nonetheless.
NVIDIA announced a sub $10,000 personal supercomputer. But more interestingly, it turns out GPUs can already be used as parallel processors using a special programming language, CUDA, supported by NVIDIA. Here’s the coolest visualization I could find demonstrating its capabilities- smoke rendering sped up with a GPU:
Here is a whitepaper on using CUDA to speed Matlab but, even better, there is a free/beta Matlab add-on named Jacket which looks much simpler. The benchmarks given on the Jacket website claim improvements of 50-400x even with relatively small matrices (bigger matrix => more dramatic improvement).
I was also informed about a recent whitepaper on parallel computing with Matlab on Amazon EC2. It looks very complicated since you need to set up a licensing server in addition to the normal EC2 configuration which, as I found out earlier, is quite a challenge. EC2 seems like a great solution for quant trading though because you don’t have to worry about hardware failures and it’s arbitrarily fast.
If you’ve had the opportunity to play around with any of this I’d be curious to see benchmarks.
