GeForce or Quadro for Scientific computing? Scientific computing.

Accelerated Computing CUDA CUDA Programming and Performance
1

I want to purchase a good graphics card for running medium.heavy duty CUDA scientific computing applications. Specifically particle based algorithms like
smooth particle hydrodynamics. Should I go in for the GEforce line or the Quadro line of NVIDIA cards? (Tesla cards, are just too expensive)

I know Geforce and Quadro have their specific strengths : GeForce for hard core gaming and Quadro for CAD applications. But on the scientific computing side I do not know which is better.

Thank you so much.

2

Read this thread: The Official NVIDIA Forums | NVIDIA

3

Quadro cards tend to cost more than the equivalent speed Tesla card:

  • Tesla C2070: 448 CUDA cores @1.15 GHz, 6 GB memory = $2.5k
  • Quadro 6000: 448 CUDA cores @1.15 GHz, 6 GB memory = $4k

For scientific computing, you are basically choosing between Tesla and GeForce. The closest GeForce to the above is:

  • GeForce GTX 470: 448 CUDA cores @ 1.25 GHz, 1.25 GB memory = $250
    (with the caveat that the Tesla cards are 4x faster at double precision, support the non-graphical TCC driver in Windows, bidirectional DMA, ECC, etc)

and of course, the top end GeForce card available is:

  • GeForce GTX 580: 512 CUDA cores @ 1.54 GHz, 3 GB memory (special model) = $590
4

I use GTX580; it is great for development, if single-precision is enough for your work.

5

recently i have been working on project which was completely rendered using max quicksilver renderer (fully GPU based). I used 4 computers built mainly for CPU rendering i7 cores: first based on quadro 4000 fermi, another two geforce 560ti and forth geforce gtx460. for all calculations gpu core was loaded on 100% for approximately 12 hours a day for one week. All geforce cards has hihger clocks and much more cuda cores than qudro 4000 as u can see in nvidia tech specs…

results:
all geforce cards running about 60 degrees celsius froze whole computer in 1 hour, later in every 30 min, hard reset needed to restart or ended in bluescreen.
Only quadro card was able to run calculations 12 hours a day for one week on temperatures about 90 degrees !!! i also tried to underclock geforce cards to -20% (gpu, ram) same results… so i stopped using them after 2 days of experiments to prevent demage to gf cards. So i was forced to buy another quadro to finish this project in time.

conclusion:
quadro cards are worth its price, they are a bit underclocked compared to geforce equivalent but they wont let you down in hard times and are capable run 24/7 on full load if necessary.
So if you are planing longtime computing on full load and wont throw away money for fried cheap GF cards or maybe motherboard this is your choice or maybe as mentioned tesla.