add-on pci card to augment 3d performance

yeah it would work fine just but would be more like a cluster because of the shared PCI bandwidth I dunno which is a cheaper option either.

PS probably can't do it with this exact board I'd suggest blades would be the cheaper option.
 
well, there's good old PixelFusion now known as Clear Speed... too bad that they had change market segment as well. (Number Nine was planned to be their card manufacturer, but went bankcrupt and was sold to S3 who wasn't interested about PixelFuzion 150 chip...)
 
Nappe1 said:
well, there's good old PixelFusion now known as Clear Speed... too bad that they had change market segment as well. (Number Nine was planned to be their card manufacturer, but went bankcrupt and was sold to S3 who wasn't interested about PixelFuzion 150 chip...)
/. is saying that the company is clearspeed tech. maybe they are one and the same.
epic
 
epicstruggle said:
Nappe1 said:
well, there's good old PixelFusion now known as Clear Speed... too bad that they had change market segment as well. (Number Nine was planned to be their card manufacturer, but went bankcrupt and was sold to S3 who wasn't interested about PixelFuzion 150 chip...)
/. is saying that the company is clearspeed tech. maybe they are one and the same.
epic

they definately are.
 
There's a story I'm tracking down about a fellow who connected something like sixteen NV35GL GPUs in parallel.

The NV35GL (workstation) has CAT5 connectors intended for parallel procesing, but the interesting thing about this story was that the person wrote a program in OpenGL for weather prediction or something massively parallel that had _nothing_ to do with graphics.
 
PCI probably doesn't have enough bandwidth to help with games, but it could help for apps that are very shader intensive. There is a Sourceforge project called Chromium with a goal of distributed rendering. I messed with it a while back, but didn't get any fantastic results with my setup. Although I can't even remember what the results were now. Below is the summary from the web site. http://sourceforge.net/projects/chromium/

"Chromium is a flexible framework for scalable real-time rendering on clusters of workstations, based on the Stanford WireGL project code base."
 
It's very interesting reading about this, but I'm wondering how these clearpseed guys intend to sustain their 250Gflops+ of performance. After all, you need to keep your units fed and 133MB PCI, or even 4GB PCI-X isn't going to be the one doing it.

Without gobs of extremely high-performance memory, this thing's going to have hopeless performance on anything with a large data set. It's certainly not just the matter of plugging a pccard into a laptop and hey presto we got ourselves a supercomputer! Hell no.

*G*
 
This piece of hardware is a vector processor, or an array processor, depending on which term you choose to use.

Vector/array processors have been around for decades, albeit typically in very high-end kit (read: One Per Country sort of stuff). They aren't new technology really, and with enough resources I'd guess they aren't that difficult to design and build.

There is, however, a very good reason we're not all using them today: they're are beneficial only for a very restricted range of problems. For example VPs are great where you have very large arrays and you need to do the same thing to each array element. Scientific simulation, weather predicition, that sort of thing, they're great. When was the last time you tried to predict the weather?

Even for some scientific algorithms (those not typically involving grids, or FFTs, or whatever), getting one of these things to provide a decent return can be a real challenge.
 
Back
Top