5

I'm planning to buy a serious GPU for running a parallel algorithm on (budget 2k-4k). Now I see everywhere supercomputers featuring nVidia Tesla GPU cards "made especially for GPGPU".

While this seems very nice on first sight, a better reading makes me have serious second thoughts on that: compared to e.g. a Radeon HD 7970, its performance (in terms of flops) is significantly lower, its cost price is significantly higher, and I can't seem to find any benchmark comparison between the Tesla and normal gaming GPUs.

I have found that the Tesla features ECC-memory. Is this the only difference? Or am I missing a deeper architectural difference between both? Perhaps relevant info: I will be using OpenCL, not Cuda.

skaffman
  • 398,947
  • 96
  • 818
  • 769
user1111929
  • 6,050
  • 9
  • 43
  • 73

2 Answers2

4

There are two technical differences I know of between the brands, when you comparing similar cards.

1) Nvidia cards tend to have better double precision FLOPS than AMD - by a factor of 2 sometimes. AMD usually does better for single precision FLOPS.

2) ECC memory is available for both brands for the GDDR5 memory. The difference is that Nvidia uses ECC on the internal memory (registers and such) as well, where AMD does not.

In my opinion, choose the card based on your application. If you use more single than double precision, go AMD, otherwise Nvidia. If you need the ECC for high fault tolerance, maybe Nvidia is your best choice. Sometimes many cheaper cards does better than 1 or 2 top of the line cards - think of PCI-e bandwidth. Read up on benchmarks, and try to determine which card is best suited for your needs.

mfa
  • 5,017
  • 2
  • 23
  • 28
  • Oh... so basically, you are saying that it more depends on nVidia vs ATI and not on Tesla vs non-Tesla? I just mentioned AMD because that's the only brand that has a 28nm-card currently, but I may as well ask the same for Tesla vs GTX580. Tesla is much more pricy, but there is no reasonable difference in the numbers to explain this... – user1111929 Feb 02 '12 at 01:30
  • 1
    The claim is that Tesla cards are tested much more than GeForce cards before they leave the factory. I am still looking around for more validation of this claim, but I believe it to be true. There should be a decent number of CUDA benchmarks for both card types, but you may have to search around as they might not be in the same place. Here's what Nvidia says: http://www.nvidia.com/object/why-choose-tesla.html – mfa Feb 02 '12 at 01:50
  • 2
    @user1111929 Some notes: Tesla cards are built to endure running 24/7, so heat dissipation has been optimised. When runnning a gaming GPU at 100% for months at a time, the fan will start to get wobbly and will become noisy and might eventually fail. Still i can't see how that justifies the price difference. As for DP performance, a GTX580 has a DP performance of 1/4 of the SP performance (so, 1581.1/4 = ~395 GFlops) while the HD7970 is rated at 947 GFLops at DP. – Emanuel Ey Feb 10 '12 at 16:52
  • 2
    Also of note is that the Tesla cards don't actually have ECC memory. They instead do error correction in software, so using ecc comes at a price: you'll lose some of your available memory, and (i assume) will also take a performance hit. – Emanuel Ey Feb 10 '12 at 16:53
2

I don't know if your problem is similar to mining bitcoins, but there is a LOT of info on parallel GPU setups here...

https://en.bitcoin.it/wiki/Mining_hardware_comparison

Gus
  • 6,719
  • 6
  • 37
  • 58
  • Thanks for the link. I see however none of the recent Tesla desktop cards listed there? The second-to-latest generation is listed there with a factor >10 behind on the top AMD cards (non-overclocked), but I'm not sure how relevant that still is for the last generation. – user1111929 Feb 02 '12 at 01:25