I was reading about the top 500 supercomputer list. The fastest one can do 35 teraflops which is 35 trillion floating operations per second. I would like to find out how many flops can my cpu do... do any of u know where to get such benchmark?
thx in advance :)
The Passmark Performance Test will give you a Mflops score, you can download a trial version here:
http://www.passmark.com/products/pt.htm
I've never used it myself though, so I can't give you any details.
Just about any decent cpu test will give you megaFLOPS (millions of floating point operations per second). Sisoft Sandra says my mildly overclocked XP2500+ gives me 3.2 gigaFLOPS.
Most X86 processors can get near their theoretical. In fact many of the computers on that list are in fact clusters of fairly ordinary systems. A P4 can do 2 FLOPS per clock cycle, so a 3.6 GHz P4 can do 7.2 GFLOPS (theoretical max). This translates into 4860 single processor systems or 2430 dual processor systems the reach that number. Since you only get about 90% efficiency, the real number is more like 5400 single or 2700 dual processor systems. (Which you can’t do yet because there is no 3.6 GHz Xeon)
If we were using A64s we would need a lot more. It also peaks at 2 flops per clock, so you would need 8100 single processor systems or 4050 dual processor systems.
The PPC processor in the Mac can do 4 FLOPS per clock but maxes out at about 70%-75% efficiency. At 75% you would need 4670 single processor, or 2335 dual processor 2.5 GHz G5’s.
I'm a little confused then... Sandra shows my A64 doing 4088 mflops at 2 ghz... that translates to over 100% efficiency at 2 flops per clock... and the comparison shows a p4 3.2 doing 4330 mflops... that's either less than 2 per clock or very low efficiency... according to your analysis, the 3.2 pentium is much faster than the A64 and should be getting 90% of 6400 mflops, but it doesn't even come close.Quote:
Originally posted by Moridin
Most X86 processors can get near their theoretical. In fact many of the computers on that list are in fact clusters of fairly ordinary systems. A P4 can do 2 FLOPS per clock cycle, so a 3.6 GHz P4 can do 7.2 GFLOPS (theoretical max). This translates into 4860 single processor systems or 2430 dual processor systems the reach that number. Since you only get about 90% efficiency, the real number is more like 5400 single or 2700 dual processor systems. (Which you can’t do yet because there is no 3.6 GHz Xeon)
I could have sworn the A64 does at least 3 flops per clock... if you go by that and 2 for the Intel chip, then they both are running at 67-68% efficiency. That's the only way I can get the numbers to add up:
4088 / 6000 (3 flops per clock, A64 3200 @ 2000 mhz) = 68.1% efficiency for AMD
4330 / 6400 (2 flops per clock, P4C @ 3200 mhz) = 67.6% efficiency for Intel
Maybe I'm missing something, but this adds up to me.
Dan, doesn't Sandra give you the effiency rating? Or is that only on the mem test?
Not that I can find, but then I need to update it, it's a few months old.Quote:
Originally posted by ATilaptops
Dan, doesn't Sandra give you the effiency rating? Or is that only on the mem test?
According to the genus book of world records the world’s most powerful computer runs at 8.5 petaFLOPS.
The Blue Waters project at my campus is supposed to do 1 petaflop, but IBM backed out. They are looking into another supplier.
Nice gravedig though, this is the stuff that is fun to dig up.
I kinda stumbled on it by accident I was doing research for my school project and found a (until today) new term for me. FLOPS I like it