General Garcia is dead now, but there are other Garcias - Elbert Hubbard
|
|
|
China aims for supercomputing top ten
A CHINESE CONSORTIUM is aiming to build China's most powerful supercomputer, by cobbling together 1920 nodes of Tyan S4982 four-socket server motherboards and bundles of quad-core AMD Opteron processors.
The Institute of Computing Technology (ICT), the Shanghai Supercomputer Center (SSC) and Dawning Information Industry Co., Ltd. agreed a joint project to build the fastest supercomputer system in China.
The SSC system will be built by Dawning and aims to achieve a theoretical peak performance of approximately 230 TFlop/s to make it the largest and fastest supercomputer in China.
The designers are hoping to crash into the top 10 in the Top 500 supercomputer ranking when it officially starts operation in December 2008. ยต
Tags: Amd
Share this:
Share

FFS, GPU's, Cell processors, why is it the fanbois/buyers of uber overpriced toys think their, suposed, all singing all dancing toys can do real pooting, pffffffft !!!!!

i'd of assumed the chinese would have just upgraded what they use now, no wait, parallel pooting, yes thats it, just build say 5 more Suanpans, the cost of which would be less than $100, employ another 5 tibetan monks (cost zero) bingo, at least 6 X the pooting power, eazy peazy :O)

for the benefit of the unwashed masses and the expensive toy fanbois that may have strayed here by mistake, Suanpan = Abacus errrrrrrrm counting frame errrrrrrrm bits of wire/rod with beads on, errrrrm that thing you used to see across the front prams and push chairs O_o
You plonker. They can't do branch prediction, and are a pain in the arse to program. Have you every tried programming them.
Each Cell processor should do more than those opterons and in total require less power to power them. And they are general purpose.
All those who suggest GPUs are cheaper and should be used to replace blades and servers are ignorant of the fact that, all problems sets solved by HPC systems cant be done on GPUs.

GPUs have capabilities to solve only specific problem sets like protein folding, video encoding for example.

You cant do every kind of problem set that a general purpose CPU can do.


So to say why dont we just replace these blades/serves with GPU is totally naive.
We not just get 192 ATI 4870 cards?

That would be a lot cheaper. Add to that the host CPU/PC running them and you wouldn't even need that many.

Why spend so much money doing it with opterons? New Radeons give over 1 Tflop each and you can put 4 in a box. Theoretically this could be done with 230 cards at a cost of around 50k :)
im happy for AMD...

still dont like ATI

saying that my last PC was A64/X800XT and then a Core2Duo/X1600 and then a Core2Quad/8800GT

Phenom just wasnt kickin it for me