Optimized for Computation
Double-precision floating point for scientific and engineering codes
2GB on-board memory 320 stream cores drive up to 500 GFLOPS single-precision performance
Asynchronous DMA, allowing data transfers without interrupting streams processor or CPU
Accelerating data-parallel algorithms in a variety of applications
Seismic processing
Financial analysis
Engineering analysis Rendering
Computational chemistry
Genetic research
Physics
Video editing
Security
==============
I wonder where are its bottlenecks and when would it be actually useful for rendering, since there has to be some sort of rewriting of the core code.
I also wonder how it’s going to be available since if everything would work OK there will be quite a high demand from everybody from everywhere…
I really thought it wouldn’t be possible that soon, but how about dreaming of Realtime VRAY
this is the part I liked,
“it offers up to 500 gigaflops of computing power, according to AMD, or about 100 times the performance of one of its dual core Opterons.”
Well, our internal research so far has shown that GPUs in their current state are not quite applicable to full raytraced rendering like what V-Ray does. Yes, they are very fast, but the communication between the GPU and the CPU and the GPU and the main memory is painfully slow, completely negating the speed you gain from the GPU. The only way to avoid this is to move all calculations completely on the GPU. However, GPUs are still way too limited to support arbitrary algorithms and complex shading effects. Nevertheless, some simplified form of raytracing with a limited set of material/lighting effects may be possible.
I think that was running on the CPU, not the GPU… even so, there wasn’t GI or glossy reflections or anything, as far as I remember.
Vlado - what about nVidia Tesla? also no go with V-ray?There are a number of emerging technologies and it’s hard to keep track of everything, so I’m not saying ‘no’ to anything just yet.
At about 50 minutes or so, a company called PeakStream claims they have developed an api able to somehow bridge the gap between standard cpu application coding and stream processing on GPGPUs.
All with little to no intervention from the coder.
They show an oil drilling research simulation of soundwaves with gains in the order of 15X, and claim montercarlo simulations on the financial market to be 15 to 20 times faster than on an average cpu.
This on an ati 1900 graphics card.
I searched, and couldn’t find their site.
It does, however, sound very interesting to my ignorant self…
