CPU Clustering - 4 CPUs that runs 1 Computer

[oliverboueid]

Run a computer equal to the power of 4 other CPUs

Can i run a single computer that has the power of 4 other computers?

let's take the example of 4 Computers the that have intel core i7-2600K overclocked to 4.4GHz each.


Can we do a Clustering system and run all the computer's CPUs as a "Single Computer" so that i will have a clock speed of 4 times 4.4GHZ equals 17.6GHz So that i can render Faster during a 3D animation process or Simulate faster...etc? is there's any operating system that do clustering(LINUX for example) ?

Could i, or could it be done?

Please could anyone help me find a way to do this?

http://en.wikipedia.org/wiki/Computer_cluster

Could it also be Done with VGA graphic Cards like the titan Computers in the blog of the http://www.nvidia.com/page/home.html

Thanks

 

[cl-scott]

While I haven't ever looked into it a great deal, this sounds like a Beowulf Cluster, or at least that would be a good place to start your search. Generally speaking Linux, or Free/Net/OpenBSD are excellent choices for this sort of thing.

 

[ss202sl]

There are already servers with multiple cpu sockets to allow just what you're talking about. The real issue is software - most software won't be able to utilize multiple cpu's effectively. There are exceptions of course, and what you are talking about is being done.

There is also software like the SETI at home program that utilizes many computers all over the world.

 

[ThatsMyNameDude]

you obviously do not know what you are talking about. Just get a supermicro motherboard with some e7s and match it with some ram and a tesla and do parallel computing.

 

[Pinhedd]

Run a computer equal to the power of 4 other CPUs

Can i run a single computer that has the power of 4 other computers?

let's take the example of 4 Computers the that have intel core i7-2600K overclocked to 4.4GHz each.


Can we do a Clustering system and run all the computer's CPUs as a "Single Computer" so that i will have a clock speed of 4 times 4.4GHZ equals 17.6GHz So that i can render Faster during a 3D animation process or Simulate faster...etc? is there's any operating system that do clustering(LINUX for example) ?

Could i, or could it be done?

Please could anyone help me find a way to do this?

http://en.wikipedia.org/wiki/Computer_cluster

Could it also be Done with VGA graphic Cards like the titan Computers in the blog of the http://www.nvidia.com/page/home.html

Thanks

Yes absolutely. What you're looking for is called a "Single System Image". There are a number of implementations, many commercial, but some such as OpenSSI are free.

Clustered computers work well for distributed workloads but will have a hard time handling sequential or non-distributed parallel workloads. Inter-node latency can get very high and as such, a cluster of four single socket computers is not a substitute for a single four socket computer.

If you already have several similar computers lying around you may want to try out OpenSSI and see if it does what you want.

 

[oliverboueid]

Yes absolutely. What you're looking for is called a "Single System Image". There are a number of implementations, many commercial, but some such as OpenSSI are free.

Clustered computers work well for distributed workloads but will have a hard time handling sequential or non-distributed parallel workloads. Inter-node latency can get very high and as such, a cluster of four single socket computers is not a substitute for a single four socket computer.

If you already have several similar computers lying around you may want to try out OpenSSI and see if it does what you want.

AH Okay, do you mean http://openssi.org/cgi-bin/view?page=openssi.html ???

should i donwload it and run it as an operating system ? or could it be run on Windows 7 X64 that i have ?

please send me more details about this

thank you

 

[InvalidError]

Inter-node latency can get very high and as such, a cluster of four single socket computers is not a substitute for a single four socket computer.

That is heavily dependent on the workload's nature.

Things like 3D rendering (what the OP was asking about) should have nearly perfect scaling regardless of latency since each system could work on its own frame(s) and the only sequential part is stitching all the individually rendered frames or segments back together since the process requires little to no communication between systems.

Inter-node latency is not much of a problem when the processing can be broken down into almost entirely independent smaller jobs.

 

[Pinhedd]

That is heavily dependent on the workload's nature.

Things like 3D rendering (what the OP was asking about) should have nearly perfect scaling regardless of latency since each system could work on its own frame(s) and the only sequential part is stitching all the individually rendered frames or segments back together since the process requires little to no communication between systems.

Inter-node latency is not much of a problem when the processing can be broken down into almost entirely independent smaller jobs.

Yup. That's why I said that it works well for "distributed workloads" but not much else. Each node in an OpenSSI cluster sees and has access to all the resources across all of the nodes. It is thus important to make sure that the workload fits within a single node.