933 GFlops anyone?

[dangerstat]

Anyone following Nvidia's TESLA personal super computing stuff. Personally I do a lot of numerical work, pretty tempted by a tesla setup. Anyone know how this compares with just using an "out of the box" GT280?

A proper "built" tesla setup is about 5k just wondering if you'd get the same power with a few GT280's thrown into a box?

Links and more info:

http://www.tomshardware.com/news/Nvidia-Tesla-Supercomputer,6616.html

 

[NightmareXX]

Pretty sure that a GTX 280 is about 1 TeraFLOP on it's own.

By leveraging NVIDIA CUDA Technology, the GeForce GTX 280 can transform itself into a fully programmable multi-processor with 240 cores, on-die shared memory, random read and write capability with 1GB of dedicated memory and the ability to deliver close to one trillion floating point operations per second (teraflop).

Source, about half way down

 

[dangerstat]

Pretty sure that a GTX 280 is about 1 TeraFLOP on it's own.



Source, about half way down

Cheers yes I think the Tesla cards are just GTX280 cards with 4 rather than 1 (or so) gigs of VRAM.

Definitely going to have to buy a GTX280 then!

 

[james.miller]

likes like a gtx280 with 4gb of slower ram. the operating frequency of the core and the theoretical performance bare that out.

 

[dangerstat]

likes like a gtx280 with 4gb of slower ram. the operating frequency of the core and the theoretical performance bare that out.

Yes but TBH a cluster with 240 codes would have at least 1.2 TB of RAM so it's lacking to say the least for anything that needs to shift loads of data about, then it's also limited by the max bandwidth of the PCI-X interface.

Be interesting to play with this on something like quick sorting or map reduce IMO.