PDA

View Full Version : PCI-Express x16 and *software* renderers


dotTom
10-25-2004, 07:39 AM
Does anyone know when we're likely to see software renderers (for example MentalRay [MR]) use the symetrical bandwidth of PCI-Express x16 GPUs to accelerate renders? It strikes me as mad that I have this high perf X800 PCI-Express card, with bitchin' vector units and a high bandwidth bus back to the CPU and MR just doesn't use it.

OK, so I'm being unfair and PCI-Express GPUs probably haven't gotten enough market share for render ISVs to start using them as "math co-processors" just yet, but does anyone know if we might see something in 2005?

Again, it's probably too soon so say, but with something like MR, if I off load some of my rendering calcs to the GPU does that count as another (costly) render node licence?

On a related note: kudos to Microsoft for stating that they won't licence "per core" on future CPUs (unlike IBM and Oracle) - here's hoping MR takes the Microsoft route. Multi-core is a solution to the general perf problem, its not in the same market sector as SMP.

Lorecanth
10-25-2004, 09:39 AM
from what I saw at the maya masterclasses you can already do it. The scene just has to be an mi file and there's a tag that adds in the graphic cards capabilities. In any case you can and will sometimes end up with artifacts sometimes, but its more a matter of your scene and how much can actually be offloaded to GPU. If you want to see a real performance now download the demo of gelato, nvidia's baby defintaly is scalable based on how fast your graphics card is.

Also in reference to actually using the bandwidth in PCI-E don't waste your breath. There was really plenty of bandwidth on agp 8x for any graphics card available today. What PCI-E allows is for dual graphics cards and overall more bandwidth for any other attached devices (for example a kick ass HD Raid array).

CGTalk Moderation
01-19-2006, 05:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.