Combineing Computers

[danielbrthwt]

I was looking at my wall and i thought i could wall mount a few computers, i was also thinking of somehow combineing these computer(ethernet , firewire) and turning them into one computer.

if this is possable how would i do it and would i need special software or hardware ?

Thanks

[Sparda]

You can cluster computers (with either Windows Server Datacenter or Linux), but it may not be useful to you depending on it intended use. Clusters only work well if the application you intend to run is multi-threaded enough to utilize all available processors.

[digip]

It would be cheaper to run multiple Nvidia CUDA based cards or any OpenCL based cards and then write your program to utilize the GPU's for the processing, such as things like OCLHaschcat does for password cracking. CPU Clusters from multiple computers are fine in big data centers where they need redundancy, but that is more and more becoming the old way of doing things, since it isn't cost effective for the overhead.

Mainly clusters are used today for redundancy of services in the corporate network, but if you want multiprocessing power for a specific single function of your own, GPU processing would be the way to go and something you can do at home for just the cost of the extra cards(given your main-board allows the expansion for multiple cards).

[danielbrthwt]

Chears that going to be really helpful

[Infiltrator]

Chears that going to be really helpful

To give you an idea, the new Nvidia GTX 580 graphics card supports over 500 Cuda Cores. If you can build a computer with 3 or 4 of those cards, you will have massive amounts of power, instead of clustering computers together, occupying space in you room's wall and consuming lots more electricity. You can have a very efficient and fast computer.

Plus applications will need to be written for Cuda or otherwise it won't work.

[danielbrthwt]

To give you an idea, the new Nvidia GTX 580 graphics card supports over 500 Cuda Cores. If you can build a computer with 3 or 4 of those cards, you will have massive amounts of power, instead of clustering computers together, occupying space in you room's wall and consuming lots more electricity. You can have a very efficient and fast computer.

Plus applications will need to be written for Cuda or otherwise it won't work.

I had a look at thoes graphics cards and they look amazing but the reason why i thought that makeing a cluster would be better was that it provided you with more flexabilty because if you do 3 way SLI with the Nvida graphics cards and you wanted to upgrade (which you probley wouldent have to do i a while with those cards) you would have to replace all 3 of them which would be expencive but if you had 10 wall mounted computers in a cluster you could replace all the parts indvidually (Unless clustering requires you to have the same hardware on each computer)

[digip]

The amount of heat and overhead electricity would be worthless. Most computers sit idle and never use more than 50% of their CPU anyway, because of throughput limitations in the board, HDD's, etc, as where running on the GPU, lowers costs all around and saves space. Cooling will be an issue if you go too many cards, but even 1 cuda card with a program written to use GPU computing will out do any of todays desktop CPU's.

See these vids for what I mean:

Edited December 27, 2010 by digip

[digip]

Also

[Infiltrator]

I had a look at thoes graphics cards and they look amazing but the reason why i thought that makeing a cluster would be better was that it provided you with more flexabilty because if you do 3 way SLI with the Nvida graphics cards and you wanted to upgrade (which you probley wouldent have to do i a while with those cards) you would have to replace all 3 of them which would be expencive but if you had 10 wall mounted computers in a cluster you could replace all the parts indvidually (Unless clustering requires you to have the same hardware on each computer)

If you are just building a normal cluster system with traditional CPUs, you don't have to use the same hardware. But if you are using GPUs, it would be a good practice to have the same brand for compatibility purposes.

[danielbrthwt]

i think from what you guys have been saying i think it would be worth it and power efficent to just make one really good super computer

[Infiltrator]

i think from what you guys have been saying i think it would be worth it and power efficent to just make one really good super computer

Yep, just like the Fastra in the vidoes above.