Hi all, new to the forums so I apologize if this is in the wrong place. Essentially I’ve been contracted by a company to reduce downtime spent processing data, it is important to note that they are moving away from a very small cluster and are looking to either get a server rack or build a new cluster. My confusion lies now in the fact that they are using MATLAB to perform lots of mathematical floating point operations to do numerical analysis, there are no graphics involved. According to NVIDIA’s website, the Kepler series GPUs are what I should be looking at but when I compared them to something like the GeForce GTX 660, the GeForce performs a higher number of FLOPS at a fraction of the cost. Now, hopefully I’m missing something and hopefully it isn’t something painfully obvious I’ve glanced over, is there any advice or comparisons that anybody can offer? It seems like the GeForce is the better deal but there’s no way that that’s accurate.
Kepler is a GPU architectural generation. For example, the GeForce GTX 660 is a Kepler GPU. When you say “Kepler” perhaps you mean “Tesla” which is a NVIDIA brand for GPUs used for computing purposes. Kepler GPUs find homes within the GeForce (consumer graphics), Quadro (commercial/professional graphics) and Tesla (compute/enterprise/datacenter) families of GPUs.
However all currently manufactured desktop GPUs by NVIDIA support computing functions. It’s quite possible, depending on what you want to do in MATLAB, that a GeForce GPU might give you higher throughput at lower cost.
For an NVIDIA take on the reasons to choose Tesla over GeForce, there is this:
Probably other folks will chime in.