I need a help/fawor making business decision

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hello,

Company that I work for is exploiting possibilities of CUDA technology for trading simulations.
We have finished our first trading simulation model based on CUDA. Speedups are quite substantial.
Run times for 620.000.000.000 double precision calculations with result set (25 000 000 doubles) returned to host:

Tesla C1060 - 86 seconds
GTX 580 - 23 seconds

We would like to invest in new Tesla cards C2090. However, it would be great to run some benchmarks before spending money on new Tesla cards and workstations.

Is there any way to run small benchmark application on C2090? I mean is there any service provider for such testing ? Or in case there isn’t, is there any kind soul that would run that benchmark :)

many thanks
Mirko

#2

I don’t know about C2090’s / M2090’s, but you could try running your simulation on a Peer1 or Amazon (AWS) GPU compute instance – from what I understand, they’re running M2070’s (or at least, they have 2x C2050 boards in the servers).

#3

Many thanks for ideas and suggestions.

NVidia helped me with benching as part of my bug report.

thanks nVidia :)