BOINC Credits

[The Borg]

Hi all,

Which indication of the BOINC benchmark has the greatest influence in calculating the number of credits?

If you compare two sets of results, which machine would be better to use as a cruncher - the one with the higher floating results (Whetstone) or the one with the higher integer results (Dhrystone)?

 

[petrusbroder]

Well, it is both ...

Here is the explanation of the credit system in the BOINC - wiki.

As you see they have constructed a atrificial entity called cobblestones, which we now call credits.
100 cobblestones is the crunching work done by a computer which does

* 1,000 double-precision MIPS based on the Whetstone benchmark and
* 1,000 VAX MIPS based on the Dhrystone benchmark.

.

So: since there is an definition, the BOINC manager runs a benchmark program, gets the Dhrysones and the Whetstones and calculates the claimed credits, which then are submitted...

 

[petrusbroder]

Now to the question you seem to ask: what is better ...

I would say it is the computer which does the most floating point operations.
Why?
Because most of the calculations are FP-calculations, and if the computer gets the WU crunched faster (because it has a strong floating point arithmetic unit) the comp crunches more WUs/day and thus gets more credits/day.

That is why the AMD processors were stonger (before the C2D came) ... they had better floating point units and thus crunched the WUs faster than the pentiums and celerons etc...

 

[BlackMountainCow]

Well, nothing to add to what Peter said!

I don't know of any current BOINC project that favors integer over floating point.

 

[Rudy Toody]

I have noticed that the BOINC Benchmarks for MIPS can be erratic. I sometimes have to rerun 4 or 5 times to get the highest possible number.

My 3 FX-60s should produce a number around 5000 MIPS but usually the benchmarks come up with 2500 to 3000 MIPS.

In fact, today I noticed one machine lagging the others and its MIPS were 1500 too low. I ran the benchmark and got just under 5000, so now I'll watch it catch up (or hold its own.)

 

[The Borg]

Thanks guys, makes a lot of sense.

 

[Rudy Toody]

An afterthought: I don't know if the MIPS differences are because the CPU is AMD or the OS is WinXPx64.

 

[Rattledagger]

For most projects roughly 99% is floating-point, so in theory the computer with highest floating-point-benchmark (Whetstone) should be the best.

But in practice this doesn't hold, since cache-size, memory-speed and so on can also greatly influence the speed in some of the projects, and it's also possible a particular cpu-type gets an advantage/disadvantage due to how the application is programmed/compiled...

Example, my computer A benchmark... 1965 MFlops but uses 60 s/TS in Seasonal Attribution.
B benchmarks 1508 MFlops, 45 s/TS in Seasonal Attribution.

After the benchmarks, A should be 30% faster, but in practice B is 33.3% faster in Seasonal Attribution, largely due to A being stuck with mediocre memory-speed.
In some other projects that doesn't rely so much on memory-speed, A is faster than B.


Many projects doesn't use the BOINC-benchmark at all to decide credit, example SETI, CPDN, Einstein, QMC, SIMAP. A couple has their own systems there benchmark-variations is largely averaged-away, Rosetta and WCG. The rest mosty relies on the normal quorum-rules, there granted credit is either lowest claimed, or average after highest/lowest claimed is removed...

So, depending on that projects you're running, how good/bad your computer is crunching the wu's has more influence on credit/day than whatever benchmarks your computer gets.

 

[petrusbroder]

Thanks, Rattledagger, for the info. Interesting and to the point - as usually. :beer: :sun:

 

[Rudy Toody]

Update: the rig that was lagging is now keeping up. Getting the MIPS benchmark up to where it belongs, makes a difference!