Alexis Cousein wrote:
subsystem wrote:
In the patent, it was prefered that each *APU* have 32 GFLOPs
performance.
Not each PE.
Yes -- *preferred*. I'd also prefer to have 1 Tflops on my desktop.
Also, it's a *patent* application, so even the "preferred" numbers are
probably not talking about the first implementation.
Yeah I also think that running 32 SIMD processors @ 4GHz within a 100 W
envelope is kind of optimistic for the 2006 timeframe.
It will probably launch at 1 to 2 GHz which will still result in TEH
CRAZY peak floating point performance.
Cheers
Martin
|