Hey,

Today i ran the benchmark option using Boinc. In the morning it was:

1462 floating point MIPS (Whetstone) per cpu
4013 integer MIPS (Dhrystone) per cpu

Tonight about 12hs later i did the same benchmark just for the heck of it and this was the result:

1458 floating point MIPS (Whetstone) per cpu
2106 integer MIPS (Dhrystone) per cpu

Now this me has completely worried! Did i kill my cpu? What else can explain a drop of approx: 1900 points?

Thanks for any help.

Not to worry.....that is one of the problems with benchmarks. The numbers can vary considerably even between benchmarks run several minutes apart. Don't worry about it!

---

Boinc....Boinc....Boinc....Boinc....

Oh thank you! i was afraid that i pushed my laptop's cpu too hard! Thanks again!

A mystery in itself, maybe its something with the planets not being aligned ;)

Seriously though, if you have an extra application chewing up power while running benchmarks in can alter them, if the smallest, its perfectly normal. The only change you will see is if you use a Boinc Optimized client, the new benchmark will alter the curving of credits for a bit...

---

Human Stupidity Is Infinite...

it cuold be slowing down to save power.

thats what even my X2 does it goes all the way down to 1GHz.

---

Just to give a graphic example of how much benchmark scores can vary -

I just suspended BOINC on my Pentium-D system and ran the benchmarks twice (after making sure nothing else ran in the background first).

Results:

30.07.2006 05:53:19||Suspending computation - user request
30.07.2006 05:53:19|SETI@home|Pausing task 06my05ab.13590.11472.159650.3.170_1 (removed from memory)
30.07.2006 05:53:19|SETI@home|Pausing task 07my05aa.11335.14000.747154.3.55_1 (removed from memory)
30.07.2006 05:53:19||Suspending network activity - user request
30.07.2006 05:53:27||Running CPU benchmarks
30.07.2006 05:54:26||Benchmark results:
30.07.2006 05:54:26|| Number of CPUs: 2
30.07.2006 05:54:26|| 1935 floating point MIPS (Whetstone) per CPU
30.07.2006 05:54:26|| 1462 integer MIPS (Dhrystone) per CPU
30.07.2006 05:54:26||Finished CPU benchmarks
30.07.2006 05:55:11||Running CPU benchmarks
30.07.2006 05:56:10||Benchmark results:
30.07.2006 05:56:10|| Number of CPUs: 2
30.07.2006 05:56:10|| 1936 floating point MIPS (Whetstone) per CPU
30.07.2006 05:56:10|| 3035 integer MIPS (Dhrystone) per CPU
30.07.2006 05:56:10||Finished CPU benchmarks

So the floating point score stayed the same, but the integer score more than doubled. The runs were about 30 seconds apart.

Benchmarks are a relative thing, and as such unreliable. That's one of the reasons SETI@Home switched to fpops counting instead (but not all other projects have yet).

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Each time i run mine it varies by very little. When i see people with slower processors and double the benchs it makes absolutely no sense at all.


Using a 3200 x64 i used to get 4000/9000+ for the benchs, now i have a 4200 x2 i get 2100/3900. If this has no relevance to seti, why is it still in there?, and why isnt there a bench that is relevant?.


Cheers

---

Each time i run mine it varies by very little. When i see people with slower processors and double the benchs it makes absolutely no sense at all.


Using a 3200 x64 i used to get 4000/9000+ for the benchs, now i have a 4200 x2 i get 2100/3900. If this has no relevance to seti, why is it still in there?, and why isnt there a bench that is relevant?.


Cheers

The benchmark, with the result Duration Correction Factor (DCF) is used to calculate the estimated processing time for the units and hence the amount of work you can download. Especially important for those that run multiple projects or have large cache.

Andy

so why are they so high on one persons pc compared to another that is supposedly faster?.


Cheers

---

so why are they so high on one persons pc compared to another that is supposedly faster?.
Cheers

Off the top of my mind...

-Overclocking/Powersaving and hidden real speed. (because the speed you see in the computer info page is the default processor speed, and may not be the speed at which the CPU is currently working at. Could be more in the case of overclocking, or less if the CPU throttled down to save power in a top of the line notebook)

-Dual core / Hyperthreading. (because for instance a Pentium 4 @3Ghz with no or disabled HyperThreading will show a higher benchmark than the same processor with HT enabled. Because the benchmark shown is "per cpu")

-Differences in hardware. (because the same CPU with crappier memory and/or motherboard and/or worse temperature conditions will perform a lot worse. But in the computer info page you only see the processor. Same with CPU architecture differences, like less cache.)

-CPU was busy at the time benchmark was performed. Besides, benchmark is far from accurate or stable.

By all means feel free to add more or correct me if I'm wrong somewhere.

---

/Mav

We have lingered long enough on the shores of the cosmic ocean.
We are ready at last to set sail for the stars.
(Carl Sagan)

The 2 pcs i referred to were stock.

All other applications were disable.

Same memory used, plus same timings, all hardware the same before and after install. More cache, less performance?.

If its per cpu, its still less when added up.

---

so why are they so high on one persons pc compared to another that is supposedly faster?.


Cheers

One of three reasons(can be a combination of one or more):
1) Either the other persons pc is overclocked to the gills.
2)He's using a third party optimized boinc client.
Or
3) Your puters are linux and his are windows.

I tend to agree with 2) in Tony's list.

The only way you could have gotten benchmark scores that high using an A64 3200+ is using an optimized BOINC client. You probably updated it and the benchmark scores have been lower since then.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

so why are they so high on one persons pc compared to another that is supposedly faster?.


Cheers

Differences in processor architecture (number of pipelines, branch prediction, ALU speed, etc.)

---