Printable View
A system containing AMD's next generation Stream Computing technology and featuring a dual-core Opteron CPU, was the basis of a teraflop beating system demonstrated at a recent press event in California.
Get the lowdown by reading this HEXUS.headline.
That sounds interesting, it says something about stream processing.
Sounds similiar to what GPUs are like at the moment, they're already capable of massive amounts of Flops if I remember correctly.
I can remember seeing somewhere GFX cards would be a better alternative to CPUs in supercomputers due to their floating point calculation potential
Bear in mind this makes little sense to me and my be utter rubish so take it with a pinch of salt :)
GPU's can be used for non-graphic processing, but apparently they are a nightmare to code for in that way
Very good AMD, but Intel did this on a single chip... :p
Just goes to show how weak-arsed CPUs are when it comes to 'massively parallel data and no real branching'
Around December 2005, ATI managed to fire up a pair of X1900 cards in CrossFire
Using a particularly generous (but still scientifically 'accurate') counting system - they managed to clock over 1 TeraFLOP in a single system
Now the CPU guys are claiming a similar thing X years later
However, the GPUs were chugging along at around 600MHz - whereas the CPUs going for performance like this seem to be in the 3-4Ghz range
I guess we are left imagining a world where either :-
1) CPU-style processors become much more parallel
2) GPU-style processor clocks ramp up 300-400%
...then all we'd be missing is an operating system and word processor to make them look pig-slow
:)