Hi there, the name is Lee, and I’m currently pursuing a degree in physics. I’m in my final/senior year and am pursing a project requiring a Monte-Carlo based integration routine. I currently have an AMD Phenom II setup (the X3 720), and am currently trying to decide upon a GPU. I was opting for the 5850 for a good bang-for-buck choice - considerations were made of power consumption, throughput, the usual necessities - but after placing my order with Komplett.ie & waiting a few weeks I’ve been told it will be near Christmas when the card is back in stock. I was wondering if I should stick with the AMD/ATI (possibly getting a 4870 instead), or would it make much difference if I opted for the NVidia route? Would this complicate matters with compilation? I’m running Kubuntu 9.10 and Fedora 12 (as of yesterday) and as I understand NVidia has better Linux support than ATI. Any help with this matter would be appreciated.
If you are in a hurry and need performance now I would suggest you carefully evaluate the current state of the various OpenCL implementations. Several of them are still in beta and may be significantly less optimized than other approaches. (E.g., I’ve heard that Nvidia OpenCL is currently slower for some things than cuda.) Overall the ATI top-of-the-line cards have higher potential performance (particularly for doubles) than the Nvidia cards, but I’m not sure how much you actually get with the current drivers. The ATI cards are also vector-processors, so if your algorithm is not a good match for that you may lose a lot of the performance.