Can't find a FLOPS benchmark test for PCs...!

I've been looking for a good FLOPS benchmarking test for the past week and I came up empty. This all started when I read somewhere on the net (from 3 independent sources) that a Pentium 4 2GHZ will do 8 gigaFLOPS and Pentium 4 3ghz will do 12 GigaFLOPS... Since I have a (old) Pentium 4 @ 2ghz myself I would like to test it and see if it really does 8 GFLOPS... The only desent program I came across that did FLOPS benchmarking was SiSoft Sandra 2005. I did the CPU benchmarking test and I got these results:

Dhrystone ALU: 4970 MIPS
Whetstone FPU: 1443 MFLOPS
Whetstone iSSE2: 2652 MFLOPS

As you guys see, the closest score to 8gigaflops was the top one... but is still off by a large margine... Anyone know why this is happening? I have also looked at the new Athelon 64 FX-57 scores listed by SiSoft Sandra and this was stated:

Dhrystone ALU: 12034 MIPS
Whetstone FPU: 4444 MFLOPS
Whetstone iSSE2: 5816 MFLOPS

See the thing is that I saw the FX-57 stated by AMD at 25 GigaFLOPS... and none of the SISoft Sandra are even close. So can ANYONE with more SisSoft Sandra 2005 experiance tell me what the hell is going on becease im lost. Thanks for reading my post Any replies are appreciated.

The test that showed 5 GFLOPS was on a x64 machine. I have a core2duo which I overclocked, and on my x64 OS I get about 26 GFLOPS.

Dhrystone ALU: 4970 MIPS

Those are NOT FLoating point Operations Per Second. They're Generic Instructions thus stress, and benchmark, a whole different part of the CPU that has nothing to do with FLOPS. CPU's are mainly designed to process generic instructions, which is why you see processors such as Cell blow away Desktop PC's in specific kinds of instructions such as Floating Point. Cracks me up... Cell is only an in-order processor, which means half the processor will be at idle even if it is at, so called, '100%'.

Whetstone FPU: 1443 MFLOPS

Those Floating Point Instructions stress the Floating Point Unit.

Whetstone iSSE2: 2652 MFLOPS

Those are SSE2 instructions which are yet again, different to the above.

If you want the true total, you have to add all the FLOPS that stress different parts of the core together to get a closer value. However, in reality it can be difficult to acheive that, because more often than not, you'll run into bottleneck in the bus, memory, schedulers, or cache. Those numbers are theoretical only, or under very special circumstances.

www.hardwaresecrets.com...

www.hardwaresecrets.com...

By the way, FX-57 and Pentium 4 are old hat. My processor.... Dhrystone ALU: 55403 MIPS... 11 times as fast. Once I get my new cooler, it will be 25% faster. That was with a newer Sandra which didn't have SSE2 test. Link to 2005 version?

[edit on 22/1/2009 by C0bzz]