Should I buy Tesla or GTX295

Accelerated Computing CUDA CUDA Programming and Performance
1

I was planning to buy several tesla cards as NVIDIA’s mad science offer seemed to be promising Fermi cards for the price difference but this is not the case.

So I’m left with the option of Tesla C1060 cards for £620 each
OR GTX295 cards for £330 each

I’m using them for CUDA programming of large neural networks and real time visual feature extraction so I don’t care about gaming performance or graphics output.

I can see that the Tesla has double the memory and the GTX has double the cores (split into 2 devices).

I’m looking for the fastest option which would seem to be the GTX, I hope there is enough memory there?
I could even mix and match, 2 tesla and 2 gtx in one machine?

What’s your advice?

2

Lots of questions :)

C1060 will cost roughly twice as GTX295 and is only one GPU vs the GTX295 which is dual - that gives a factor of X4 in favour of the GTX.

Fermi will roughly run twice as fast as C1060 so you can compare it to the GTX.

Some algorithms can split the code to run in chunks and then you go around the RAM limitations. My algorithm for example needs to calculate

much much more GBs than 4 GB (that C1060 has) or 6GB( that Fermi will have) - so I need to break it into chunks. Therefore breaking it

to work on a 1GB card (such as the GTX) is a non issue as well.

As for mixing - My development computer has GTX280 as the primary and C1060. works great and I usually run my CUDA kernels on the C1060

so that faulty kernels will not crash my windows.

Hope that helps…

eyal

3

To some extent it depends on your application (i.e. how much RAM you need) and who’s paying for your hardware. If it’s someone else get the Tesla’s and take advantage of the Fermi offer.

4

So thats one vote for the GTX and one for the tesla…

5

I’ll just repeat what everyone else said. Some applications are able to crunch large data sets on GPUs with a small amount of memory because they can break the data up and process it in chunks; some can’t, and so they need the larger memory available from a Quadro or Tesla card because all of the data has to fit into the GPU’s memory at once.

So if you can break up the data, go with the GTX295. If not, go with the Tesla (or a Quadro, if you want the video output).

6

This is correct only if your 4GB of Tesla RAM is enough :) if its not enough, you’ll have to chunk things even on a Tesla…

Unfortunatly for me I have to do it all the time (and 6GB RAM will not be enough either, as a matter of fact a 72GB of CPU RAM isn’t enough either ) :)

eyal

7

Go 4 Tesla ONLY IF you need that extra memory. TESLA is enterprise grade - more thoroughly tested and has ECC for memories, I believe.

Also I hope GTX 295 has the enhanced memory coalescing capability that TESLA series have. And, I hope they support “doubles” as well.
You may need to check the compute capability of both these cards. TESLA is 1.3 and supports “doubles” as well. What is the compute capability of GTX 295? If it is 1.3, it should be good enough.

If you dont need any of these features (now and in future), GTX 295 is the best option. They run faster than TESLA (more clock speed for shaders and memory) and they are priced less.

8

ECC is a new feature of Fermi. The C1060 does not have ECC memory.

The GTX 260, 275, 280, 285, and 295 are all compute capability 1.3 devices.

The GTX 295 runs at 1.242 GHz, and the C1060 runs at 1.3 GHz, so there is no clock advantage. The GTX 285 runs at 1.476 GHz, though.

9

Seibert,

Thanks for clarifying!

And so, I think I would go for GTX 285 instead of TESLA provided I dont need that much memory…

10

Thanks everyone, I’ve gone for tesla’s for now as I do need quite a bit of memory for some applications. I’ll be buying more later I think but I’m going to wait for fermi and see what the options are then…