BOINC FP and INT measurements?

[Bender10]

I saw someplace, a paragraph explaining the FP and INT speeds for the cpu's. I thought it mentioned that these numbers were not very accurate. Because, I was just looking at the results for the PE client test I ran, and the measured speeds were a bit different.

Stock machine: 1014.87 FP
1892.81 INT

PE Machine: 1010.26 FP
2039.09 INT

To me they look close enough to be the same. But the INT on the PE changed a bit...

Is there a Rocket Scientist in the building?

[drezha]

Basically it depends on what else is running as well.

Hence the Bart PE has less thing running at the same time so gets the test done quicker.

[Bender10]

Ok, I should have thought of that.

I'm sure the stock machine has other junk running. It was a donation, and I did not do a clean re-install. I just cleared out the obvious junk and plugged it in...

Another plus for PE I guess.

[Nflight]

Is there a Rocket Scientist in the building?

Well Maybe, but I am more of a researcher then the actual Scientist. Although I can go pretty deep into detail, lets put this into perspective or confuse the living daylights out of, either way I am going to comment here!

#1) http://hint.byu.edu/documentation/Gus/HINT/ComputerPerformance.html#SGI
This will give you insight to the parameters of the comparison of Different Divisions of performance.

#2) http://en.wikipedia.org/wiki/Floating_point This relates to the FP as if it were a floating point, and this is how I would perceive this two letter configuration.

#3) http://en.wikipedia.org/wiki/BOINC_Credit_System#Cobblestones This is clearly the description that describes the most beneficial understanding.

#4) Your comments and insight would help in getting me to understand what direction you want to go from here.

[Bender10]

I just wanted input as to if I should accept the 'benchmark' values listed for FP and INT speeds at face value, Or to see these values as a 'point in time' reference.

If I take this info at face value, and take into accout the running processes on the computer. Using the same computer to run each configuration (PE and, stock XP install). Running a diskless (operating system?) computer built for one purpose, with less overhead (processes) will process faster than the same stock computer with who knows what is running in the background.

I guess I never gave background processes much thought (like most people). Because I was never trying to use the computer to 100% of it's ability.

*DC is still new enough to me, and I am sure to be covering ground you have all been over...

[drezha]

BOINC is very sensitive to whatever else is running whilst it's benchmarking.

And you cant really change that. It also benchmarks every so often (once a week I think) to keep the benchmarks in check. Obviously reducing the processes running then benckmarking it again might help but not overly so.

Anyhow, all DC process are effected by the PC use. It just happens that BOINC gets hit in the benchmarking phase whilst Folding etc get there slowdown in the processing. BOINC get's it's points from the benchmarks and if it takes longer processing, you'll still get the same amount of points because lower benchmarks*longer time = good benchies*shorter time. (at least in theory thats the score ;) )

[Bender10]

Ok, That makes sense. And it does help me out. I've been hearing too much stuff about custom clients that change your benchmarks vs running the stock client. Looking around I see aquite a few boxes with what looks like inflated benchmarks when compared with similar boxes.

I'm really looking to optimize my machines for power consumption, (I'm starting to get a traffic jam in the basement) components required per machine, noise level, and if they will process a bit faster, so be it...

If I can strip my machines down to the basics and still crunch, I'll have more parts to help my friends out when they need a confuser fixed or rebuilt.

I'm digging right now to see if I have a motherboard (that werks) so I can put together a permanent BOINCpe box...

[drezha]

Optimised apps were (not sure about now) ones that used SSE, SSE2 etc to make the benchmarks higher to claim more credit.

Back when I BOINC'd, they were needed for the linux client because for some reason, the linux client apppears to be penalised and got lower benchmarks than the exact same PC on windows.

Nowadays there's not really needed as the BOINC client itself uses SSE IIRC.

TBH I doubt you'd see a huge, if noticable increase in points stripping them all down to the bare essentials.

[Bender10]

Ok thanks for the help.

I'm not really looking for a speed increase, just a reduction in power consumption (no HD? or even CD), and parts required.

Just tinkering.

[Nflight]

Replace the HDD with a NAND Drive, that will save you some bucks! :icon_wink:

[Bender10]

I was going to try a CF card......