NVIDIA's Fermi: Architected for Tesla, 3 Billion Transistors in 2010

in that case scaling is entirely dependent on your code

I wasn't thinking of games. I intend to write my own programs for the eventual cluster. Using the four low end cards was to get the program and system functioning. When the horsepower is required, I'll move to the faster cards.

Having a video port on the system would be a bonus.

What do you expect to gain from four cards? Unless you are driving two 30" monitors anything more than 2 GPUs is going completely to waste.

Keep in mind the majority of games will not scale beyond 2 GPUs because the software wasn't written for it, or they aren't stressful enough. A simple GTX 260 is enough to drive a 1920x1200 with all settings maxed in pretty much any game. And unless the game is specifically coded for SLI by NVIDIA, then it will only use a single GPU.

I'll probably build the system with four low end cards and get some test results before throwing money at four fermi cards.

If that overloads the system, I'll move two to another MB.

Multi-GPU scaling = diminishing returns... 2nd isn't too bad with mainstream games, but 3rd rarely helps significantly and 4th is there mostly for looks.

that's because GPU scaling decreases very quickly after 2xGPUs; CPU bottleneck is not quite the issue

I was thinking more on the lines of transfer speed. Can the CPU keep up?
Somewhere I read that more than two GPU cards ( 9800s and better ) was a waste.

don't see why not, as long as the PCI Express slots are far enough apart

Will a MB be able to support three of these?

I'd rather use that extra $60 to buy a better GPU

If I'm going to pay for the GPU to sit idle, then I might as well pay the ~$5 difference to have it doing something useful per month.

definitely NOT folding;
so that leaves it running idle and using as little power as possible

And when you are not gaming, what is your GPU doing?

Looks impressive, but can it do that simulation in real time?

Crysis with this water....

128 thousand particles simulated and rendered in real-time on the GPU. SPH fluid simulation with surface tension effects.

As demonstrated at the NVIDIA GPU Technology Conference 2009

I use my GPU 99% for gaming

Perhaps!

But lets be honest, what is the GPU used for most? I use my GPU for 80% folding, 15% gaming, 5% misc tasks. If I'm not in a game or watching HD video then the GPU is folding or running some kind of computational program. I understand I'm in the minority, but it doesn't change how I use my hardware.

Also, is amazing to think but Fermi will be very near the same die size as GT200b, around 467mm² by Anand's estimates. Hard to fathom now that G80 was 576mm²

you have been brainwashed!

Certainly big enough to have brought its own!

True MIMD design and so many enhancements... Bit odd but I'm more interested in this for direct computing than gaming.

And yet, a beast has been slain before, it will have to come armored this time

What a beast.