Quote:
Originally Posted by repudiado  You really didn't read the article. We are talking about a computer (only one) priced in 4000 € who can surpass the process capacity of any of these common clusters with hundreds of PCs basically employing the outstanding capabilities of the modern GPUs, not CPUs. To make that possible you need to redesign the system core functions in low-level programming which is what this belgians did. They are using Linux because of the stability matters and experience they have. Windows is pretty unstable and probably they don't have too much practise with MacOS. Apple has announced they will use the same technology for the next OS version, to take advantage of the enormous processing capacity of the GPUs who are not been used when people are not playing with some stupid 3D game. |
I did read the article, I was just pointing out to another way to create some very nice computing power (comparable to a supercomputer many many times more expansive than the combined price of the hardware used to make the Beowulf itself) Sorry if my mention of the Beowulf Clusters offended you, I wasn't belittling the computer mentioned in the article you pointed to.
As for cuda (the technology used to interface in C with the gpu of the nvidia gpus - btw, due to their more massive parallelism the ati gpushould even perform better (240 alu for nvidia vs 800 for ati) yes, it is great and I do hope more and more application will use it (it doesn't really need any OS support to work). Problem is, we are talking here about massively prrallel programing, a totally different beast compared to our traditional programing - for example, when it comes to massivly parrallel computing you have no loop instructions but you "simply" fork you code to each core (see how much problem people have to code certain things to take advantage of 2 or 4 cores). Some applications can take advantage of this very easily others (like for example a ray tracer) have a lot more problems. I was hoping to see a support for cuda in mental ray very soon (after all, nvidia owes mental image now) but after talking to someone who really knows very well what he is talking about, chances are pretty slim to see that happen soon (read 2 to 3 years) and anyway no much hope to see it have a huge influence on the render time
Opposite to that, a program like photoshop could really hugely beneficiate (and will probably pretty fast) of this technology, and obivously, when remaining in the graphism world program that either compute pure physic and program that compute fluids (be it a FumeFX in max, the fluid mechanic in maya or a tool like RealFlow
Linear Mode
