Gigaflops

[Lonyo]

I know people have thought about gigaflops for performance ratings, but does anyone have any tables of AMB/Intel CPU's and their ratings?
What Intel/AMD CPU would be about 7.5 gigaflops?

 

[mechBgon]

I'm running SETI@home at the moment, and fired up SETISpy to check the processor performance. Currently it's showing 306Megaflops/sec, and that's an AthlonXP 1600+ at 1.4GHz (at the moment).

 

[EKAtBzboyz]

Today, a 1GHz PowerBook G4 hits speeds of 7.5 gigaflops.
http://www.apple.com/powerbook/processor.html

http://www.cacr.caltech.edu/~johns/papers/dmkd/html/node2.html

http://www.geek.com/news/geeknews/2002may/bch20020530011966.htm

i read somewhere that durons were capable of 3.8gigaflops
dotn quote me though

 

[dannybin1742]

oooooh 7.5gigaflops

gigaflops = apple's BULLSH!T way of GAYging performance

don't buy into it too much

 

[Jeff7]

Well, I found an "AthlonXP model 8" document that says it can produce 4 floating point operations per second. I'm guessing that that would be a Thoroughbred processor. An XP2700 runs at 2.17GHz. So that would theoretically yield a maximum of 8.68 FLOPS per second assuming my thinking is correct. I doubt that you'll ever see this level of performance in any applications of course, as the CPU also has to take care of running the hardware and the OS.
According to the model 6 document, my XP1900 can also produce 4 FLOPS per cycle; it's at 1.6GHz, so that'd be 6.4GFLOPS. SetiSpy reports that my SETI@Home is doing 235.79MFLOPS. My system has always been slow at Seti. I use the GUI screensaver with the blank screen option because of the convenience; guess it just doesn't do as good as the command line version.

 

[imgod2u]

All the "gigaflop" numbers are really just theoretical. Conventionally, it just multiplies the maximum FP operations throughput by the clockrate. So:

The Athlon has 3 parallel FPU's and 3 decoders to feed them, so it can do theoretically 3 FP operations per clock. At 1.4 GHz, that would be 4.2 Gflops (billion FP operations per second) for x87 code and 5.6 Gflops for single-precision SSE code.
On the P4, which has 2 parallel FPU's and an execution trace cache that can issue up to 3 micro-ops per clock to feed it (assuming this is repeatable code). At 1.4 GHz that would be 2.8 Gflops for x87 code and 5.6 Gflops for single-precision SSE code and 2.8 Gflops for double-precision SSE2 code.
The PPC G4 processor has 1 general FP pipeline (fully pipelined) with a throughput of 1 FP operation per clock. At 1 GHz, that would be 1 Gflop. Not impressive, however, combine that with the AltiVec units which function independently from the conventional FPU and can do 4 or 8 (depending on the variety of your AltiVec instructions) single-precision operations per clock, you get a theoretical 5 Gflops in the case of only 1 AltiVec vector per clock and 9 Gflops in the case of 2 AltiVec vectors per clock at 1 Ghz.
This is all theoretical. I would say it's safe to say that this never, ever happens (the maximum being reached that is).