Modelling Approaches: CUDA

When discussing with another student about what options I want to consider for my project (on Saturday night), in particular, the amount of power I need to draw from the computer to get the quickest compilation time, CUDA technology was recommended.

CUDA is NVIDIA’s parallel computing architecture that enables dramatic increases in computing performance by harnessing the power of the GPU (graphics processing unit).

[Source: What Is CUDA?, accessed 1st March 2010]

I was given a very brief overview of what the technology does, but basically, it allows rendering direct to the GPU. Some interesting information here about the new line of graphics cards from NVidia:

The next generation CUDA architecture (codename: "Fermi") which will be standard on NVIDIA's upcoming (GeForce 400 Series [GF100] (GPU) to be released available Q1 2010) GPU is designed from the ground up to natively support more programming languages such as C++. It is expected to have eight times the peak double-precision floating-point performance compared to Nvidia's previous-generation Tesla GPU.

[Source: Future CUDA Architectures, accessed 1st March 2010]

It looks like a good concept. Looking around at things on forums, there are plugins available for Maya to render using CUDA which would be good. I'd love to sit down and experiment on rendering time using CUDA VS my current computer.

I've been considering what sort of computer I want to use for my project, and just trying to figure out costs. My current system, see specs below, runs maya quite well.

Processor: Intel(R) Core(TM)2 Quad CPU Q9300 @ 2.50GHz (4 CPUs), ~2.3GHz

Memory: 3582MB RAM (4096MB Total)

Hard Drive: 2.8TB

Video Card: NVIDIA GeForce 9800 GTX 512DDR3

Sound Card: Speakers (High Definition Audio Device) 7.1+2 Channel High Def Output

Operating System: Windows Vista™ Home Premium 32bit (6.0, Build 6001) Service Pack 1 (6001.vistasp1_gdr.080425-1930)

Motherboard: Gigabyte GA-EP35-DS3P S775 P35 FSB1333 DDR2 PCIEX16

Computer Case: Antec 900 Gaming Case

Much better then my friends system that I used to render one of my assignments when in Cairns. A total of 8 hours rendering time VS 15 minutes. Drastic improvement. If I could get something similar or better to use for my project, that'd be pretty good. I like the idea of the i7 processors that Intel released this year and how they use hyper-threading to achieve faster speeds. (http://www.intel.com/products/processor/corei7/index.htm). I think if I were to combine a system with an i7 processor and a CUDA enabled graphics card (wishing the GeForce 400 series would come out soon) then it'd be pretty interesting to see just how well the system can render. Especially considering if it's a scene filled with trees (or more specifically, leaves). It'd be the same concept as particle rendering, and that can cause my current system to get upset if the particle count is too high.

Something to think about anyway. I'll definitely have to look further into the CUDA technology.