device which delivers 4 TeraFLOPS of computational goodness. And if 

C1060. (The S1070 is basically four of these in a box. The C1060 is just 
a normal PCI card.) This still delivers 933 GigaFLOPS of compute power.


...of course, all of those are *peak* values. So the S1070 delivers "up 
to" 4 TeraFLOPS. (Depending on how well your software utilises it.)

All of those are also for *single-precision* floating-point computation. 
The C1060 delivers 933 GigaFLOPS peak power for single-precision, but 
for *double-precision* it delivers a far less impessive 78 GigaFLOPS. 
(And that's still *peak* performance.) The S1070 correspondingly 
delivers only 345 GigaFLOPS in double-precision arithmetic.



Hell, if I had any clue how much performance a current CPU delivers, I'd 
know if these numbers are impressive or not! ;-)



I don't think I'll rush out and buy a Tesla any time soon though. 

GigaFLOPS. That's presumably single-precision again, but it's comparable 
to the (vastly more expensive) Tesla C1060, so it's not unreasonable to 
expect double-precision to be similar too.

Post a reply to this message