Please help me choose a GPU

mohmoh
edited August 2016 in DualSPHysics_v4.0
Which of these 2 factors is more important in a GPU for CUDA calculations? -Memory -Number of cores

i.e. Which one of the following GPUs is better and fester for CUDA?
1) 2GB, 1024 cores
2) 4GB, 600 cores

Comments

  • memory=> depends on how many particles you want to simulate
    #cores=> faster computations
  • However there are much better options that those two examples you showed
    Regards
  • Thanks
    The options are dependent on my budget.
    Would you mind introduce the better options?
  • When choosing a GPU for CUDA processing, the thought process is quite simple really:

    1) Create a short list within budget
    2) From the short list pick the ones with the most CUDA cores (this will define the better/newer GPU architectures)
    3) Pick the card with the most memory from the shortlist in point 2

    Typically if you have a card with the same number of CUDA cores that has 2GB of memory and one with 4GB then you end up paying a premium for the extra 2GB plus the 4GB cards tend to have slightly slower memory! The thing you need to remember is that unless you are buying a Tesla device then all of the marketing will be aimed at people wanting to do computer gaming or computer graphics and often manufacturers double the RAM to make a product look enticing but in fact still use an older GPU chip (where much of the actual expense lies for them...) so in reality the processing capability is less.

    Of course, as Alex says if you have more memory then you can process more particles, but I would suggest if you are looking at budget gaming GPUs with either 2 or 4GB of RAM then concentrate on processing power and not RAM size, it is only when you get to the cards with 6GB (or more) that you are getting to the realm of powerful GPU AND a decent amount of RAM.
Sign In or Register to comment.