"condor" wrote in message <firstname.lastname@example.org>... > Hi I am building a new workstation and my question is about where should I spend my money? > > CPU or GPU (or cloud)? > > My code can be parallelized easily ... However I haven't ever tried a GPU and I have no idea of what could be the difference if any... > > I know that the C2075 is capable of 515Gflops ... > What about the CPU? Maybe around 120Gflops??? > > What would be the difference in speed? 515/120= 4x? > > Furthermore, choosing the GPU, I don't understand why I should buy the AccelerEyes toolbox? > Looking some documentation here http://www.mathworks.it/discovery/matlab-gpu.html it seems possibile to use the GPU (if the compute capability is higher or equal to 1.3) with the parallel toolbox that I have. My only real need to to send for loop to the GPU ... So do I really need the accellereyes toolbox? > Furthermore a complete package of Jacket would be around 4000$ (crazy!) > > I am thinking also to another solution...to distribute parallel computation on Amazon (or others) cloud ... this would be cheap at least in the short term, what are the cons in your opinion? >
Another thing ...
A Tesla C2075 has 445 cores at 1.15 Ghz = 511.75 Ghz --> costs between 1500$ and 2000$ A normale video card as the new 680 GTX has 1536 cores at 1.06 Ghz = 1628.16 Ghz --> costs around 600$
Is the 680 GTX 3x more powerful than the c2075? What's the difference and how many Gflops achieve a video card?