Smaller, Faster and Much Cheaper - Cluster Level Performance Right At Your Desk

I just returned from the London and Paris events. Wolfram, NVIDIA, Supermicro and us presented in-time results  that were impossible before.

Over 100 delegates followed our presentations  and live demo. In spirited discussion we came to the conclusion that in technology development there are rare windows of co-evolution where technology foundations, like computing power, stimulate software developers to take all efforts to optimize their algorithms and schemes as well. Valuate over a million instruments in a second is such a result. It was so far out of reach in sw only and impossible at a desk hardware - but became real in the interplay.
Obvious questions, what will it finally cost? Take the one-million-valuations-in-a-second case.
A quadcore PC would cost less then EUR 2K. For simplicity, one NVIDIA Tesla C2070 GPU card the same. For valuation and calibration tasks we can achieve a speed up of 100 from one CPU to one GPU and 100 from clever UnRisk optimization.
A cluster of 100 core CPUs represented in 25 networked Quadcore machines would cost approx 10 times as much as a hybrid quadcore CPU system with one NVIDIA Telsla 20-series GPU in one little box. With a small fraction of space.
And also UnRisk and Mathematica licenses will be a 10th of the equivalent grid licenses for 100 computational kernels.

In short, UnRisk/Mathematica boxed in a quadcore machine with one Tesla C2070 GPU will total to a purchase of  approx. EUR 12.000 including one year of premium service.
And in cases it will speed up 100*100*4 --> 40.000 compared to one CPU.
With cluster level performance at the desk and a tenth of the price.