INTEL W5590X2(HT) 12GBram (96FPS)

Started by jbeau, November 09, 2010, 01:16:13 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

jbeau


16 processors
12GB ram
Dual GTX 480's on Cubix
Can't wait for a hybrid GPU solution from KS.

guest84672

The new integrated CPU/GPU architectures from AMD and INTEL may give a significant boost to high-end rendering algorithms like KeyShot has. Until then, we will continue to use the CPU exclusively while accessing the GPU for additional rendering effects.

jbeau

Too bad KS doesn't have any plans to access Cuda or openCL. It seems like the rendering industry is shifting to work with these new graphic card architectures. KS Effects are nice, but just imagine KS unbiased lighting in realtime ;)

guest84672

We didn't say that. OpenCL is certainly under investigation, and we are looking into it how we can fit it in. GPUs have a lot of limitations and are not the holy grail as Nvidia wants you to believe. It always a bad idea to develop something that requires specialized hardware. It has failed over and over again in the past.

jbeau

Thanks for the clarification ;D Exactly what I wanted to hear.

I imagine that openCL could be used for unbiased physical skylight, object lights(emissives) instead of using photons. The biggest limitation currently is the VRAM which uses a lowest common denominator of RAM if you have multiple cards. The total VRAM directly scales with the amount of geometry and textures that can be used in the scene. If you manage your content to work with instances or proxy's and optimize your textures then this really isn't so much of an issue. I've been able to get 8 million + polys with realtime pathtracing and it renders in 15-30 secs at 720P with dual GTX 480's. Besides GPU clusters are much more affordable then CPU's but the main issue at the moment is heat and wattage, which has forced me to use an external solution.

guest84672

The may be more affordable, but don't scale as well as CPU's.

jbeau

Cuda/openCL doesn't care if you have onboard SLI, as long as there is an active slot open it just works and the scalability in speed more then doubles in SLI depending on the card. You really have to see MLT(Metropolis Light Transport) in action to appreciate the unbiased physical lighting being calculated. Not to mention, motherboards now come with 10 slots of 16xPCIe, as well as external GPU solutions like a CUBIX (a poorman's quadroplex) at a fraction of the cost.  

timb

New IBM chips available next year.  I'm sure they will be expensive when they first come out, but the technology looks very promising.


http://www.popsci.com/technology/article/2010-12/ibm-unveils-nanophotonic-chips-could-lead-exascale-computing-revolution