So after reading all of NVIDIA’s hoopla about their “personal supercomputer”, I decided, well instead of spending ten thousand bucks, I can do this right now! I’ve got the GPU power, it’s just a matter of trying to get this shit to work.
So I thought to myself… well… those C1060 cards that NVIDIA sells for $1600 each is just a GTX card basically WITHOUT video output… so let’s see… I have a GTX 260, why don’t we run some CUDA code on there and use the GPU as a processor…
I downloaded SETI@Home for CUDA and bam… no problem… works fine. Then I thought to myself… if I have 2 PCIe 16x ports… I could load another card on, well sure enough my ASUS P5K does have two of them, so I grabbed my 9800GT which isn’t being used because of RMA and stuck it in.
Well, strange things start happening, computer isn’t happy… I check my power supply… 550 watts… that’s not enough, we need nuclear power, maybe a Dyson Sphere… so I hack my other 430W PSU to power up without being connected to the motherboard and use that as an external power source for the 9800GT, jimmy rigging ensues:
Now I can run crazy crap with CUDA code and have my own parallel processing unit at home, big fun! Sometimes you have to wonder though… have I gone off the deep end?
0 Responses to “CUDA applications and multiple GPU cards.”