Dhrystone MIPS 2.1 Benchmark

[dmk11]

I don't know if this would qualify as "highly technical", but here goes.

Who developed this benchmark? I haven't been able to find (by googling it) about the details of this benchmark. Seems like every processor manufacturer runs this benchmark to compare themselves to each other.
I found some blurb that says that version 1.1 of this benchmark does NOT access the system memory - all the programs are small enough to fit in the cache, but I haven't been able to find anything that mention whether version 2.1 does this or not. Anyone knows?
Can anyone enlighten me on this subject?

thanks!

 

[SuperTool]

I believe Drhystone benchmark tests integer performance. I think Whetstone tests Floating point.
The reason you don't want it accessing memory is because that would essentially nullify any differences between processors, because memory access takes so much longer than any integer operations, you would basically be measuring memory access times instead of integer peformance.
Instead, the proc loads the data into the cache before running the benchmark. That way, its isolating what it is testing to the ALU instead of the RAM subsystem.

 

[Locutus4657]

Don't worry, MIPS are a poor excuse for a benchmark anyway..



<< I don't know if this would qualify as "highly technical", but here goes.

Who developed this benchmark? I haven't been able to find (by googling it) about the details of this benchmark. Seems like every processor manufacturer runs this benchmark to compare themselves to each other.
I found some blurb that says that version 1.1 of this benchmark does NOT access the system memory - all the programs are small enough to fit in the cache, but I haven't been able to find anything that mention whether version 2.1 does this or not. Anyone knows?
Can anyone enlighten me on this subject?

thanks! >>

 

[dmk11]

I guess I see your point, SuperTool. On the other hand, since external memory access is such a big hit on performance, they should take that into consideration between integrated processors, where the memory controller is on chip.
I guess it depends on what the cache hit percentage is. anyone knows? I remember it somewhere in the 80% or higher. Does it matter if you're running Object Oriented code (tends to be bigger, less likely for it to fit in the cache)?

Yes, MIPS is useless, but that seems to be the only benchmark that all the processor manufacturer provide, you gotta start somewhere...

 

[Locutus4657]

Assuming you're talking about general purpose CPU's, you could do a google search for the latest version of SPEC. It's fairly subjective like most benchmarks but it's very well respected and if you want to talk CPU horse power with the real guru's (the guy's that have been in the industry for decades) you'll have to come up with something better than MIPs. SPEC should be available for all CPU's CISC and RISC for all OS's.



<< I guess I see your point, SuperTool. On the other hand, since external memory access is such a big hit on performance, they should take that into consideration between integrated processors, where the memory controller is on chip.
I guess it depends on what the cache hit percentage is. anyone knows? I remember it somewhere in the 80% or higher. Does it matter if you're running Object Oriented code (tends to be bigger, less likely for it to fit in the cache)?

Yes, MIPS is useless, but that seems to be the only benchmark that all the processor manufacturer provide, you gotta start somewhere... >>