CUDA specific setup..

[modi123]

I am beebopp'n around this Monday at work looking into parallel processing problems with CUDA from NVIDA. It is moderately more appealing to me than, say, a multi processor cluster or setup just based on space/power requirements/cost vs uptick in performance.

From what I gather all I need is a moderately okay CPU and one to many NVIDA cards tossed in a machine and the CUDA libraries deal with making the same/different cards all play well? That seems deceptively.. simple.

If I whip up a Monte Carlo simulator or something along those lines I should be able to see a performance gain between running it on my regular CPU versus shoving it off onto my NVIDA card?

Just a reality check for me looking into investigating applying a few random NVIDA cards in the closet to this build before investing something substantially more.

 

[josephnunn]

For a deep learning setup to train neural networks.

A Full Hardware Guide to Deep Learning - Tim Dettmers

What I'm in the process of putting together, it is certainly the mostest for the leastest at this point in time.

Building a 32-Thread Xeon Monster PC for Less Than the Price of a Haswell-E Core i7

Joseph

 

[Deleted member 82943]

I built a system around dual e5 2670s 128GB ddr3 and a Titian X. I couldn't be happier.

 

[LightningCrash]

When you get the hang of it, you might try out the Amazon GPU cloud servers. The smaller GPU instance is $0.65/hr and the larger one is $2.60/hr.