Message boards :
Number crunching :
Weird Benchmark ... !!!
Message board moderation
Author | Message |
---|---|
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
I ran across this guys Computer with a P4 2.8 that is Benchmarked ridiculously low IMO. He never claims more than 3 or 4 credits for any of his WU's & it has to be screwing other people out of their proper credits I would think ... Low Benchmark PS: Upon looking through his finished WU's it don't really look like he's costing anybody any credits but he sure isn't helping anybody either I would think ... He probably doesn't think theres anything wrong with his Benchmark because he's getting 25-35 Credits and only claiming 3 or 4 all the time... |
Arm Send message Joined: 12 Sep 03 Posts: 308 Credit: 15,584,777 RAC: 0 |
Yes, but every time he takes between 20 and 40 credits :) I saw many PCs with 0 credit claimed - they show 0 CPUs. |
John Hunt Send message Joined: 3 Apr 99 Posts: 514 Credit: 501,438 RAC: 0 |
I'm not computer literate at all, and the subject of credit claimed/given has often made me wonder...... When I download WUs, the 'time to completion' is always stated at 6hrs+ but I rarely complete one in under 10hrs. Claimed credit is usually around 100 but credit given is variable, rarely over 50. The machine I'm running (at the moment - soon to be upgraded!) has a Celeron 2.4GHz processor and 112MB available RAM (16 used for video/graphics brings that up to 128MB but is not accessible). Planned new machine will have a Pentium 4 3GHz processor and 512MB RAM. Could anyone give a 'Joe Public' explanation of how the credit system works? |
Arm Send message Joined: 12 Sep 03 Posts: 308 Credit: 15,584,777 RAC: 0 |
> > Could anyone give a 'Joe Public' explanation of how the credit system works? > A very good explanation for most of the things related to the SETI and BOINC, you can find on [url=http://boinc-doc.net/index.php]Paul D. Buck's site<a> :) |
Paul D. Buck Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0 |
> Could anyone give a 'Joe Public' explanation of how the credit system works? As Quantum said ... there is only 300 pages of material here, so you should be done reading them in about an hour of so ... :) Look in the glossary, and in the FAQ. The darker Blue FAQ icon goes to a database driven system phpMyFAQ that has a search capability (though I am not sure how well it works yet). Feel free to complain, send me an e-mail (using one of the many links therein) to ask other questions, ask here ... add a question in the FAQ, vote on your favorite FAQ questions ... I am making progress on things such as user tracking so that I can get a feel for which pages are used the most and deserve more attention. And, oh-by-the-way, if it is not understandable ... let me know about that ... You are gong to have to suffer my sense of humor ... sorry about that! Happy reading! :) |
7822531 Send message Joined: 3 Apr 99 Posts: 820 Credit: 692 RAC: 0 |
I ran across this guys Computer with a P4 2.8 that is Benchmarked ridiculously low IMO. Remove the "IMO" from your statement and look at my emulated host. It's a pesudo-PII at 524MHz W2K SP3 and gets 556/919 stones on an 867MHz host. One would expect that a real P4 2.8GHz would have to get a much higher benchmark than that! Doesn't the BOINC client automagically re-run the benchmarks on a periodic basis? |
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
One would expect that a real P4 2.8GHz would have to get a much higher benchmark than that! Doesn't the BOINC client automagically re-run the benchmarks on a periodic basis? ========== Yes on both statements NA, a P4 2.8 should be getting at least in the 1200-1300 range minimum & my Computers are Benchmarked Automatically every 5-6 days, sometimes more often than that even. This guy's showing 188 for a Benchmark, my Wrist Watch could get higher than that I think ... :/ |
7822531 Send message Joined: 3 Apr 99 Posts: 820 Credit: 692 RAC: 0 |
All I can think of is that the CPU is running as a server (dual P4 1GB RAM) and BOINC's been "bonk"-ed down to the lowest priority. But still I'd expect a lowball number from a 16MHz 68030 w/ 68882 FPU... I'm gonna try to set up my 160MHz 603e later on to see what I get from BOINC... |
NickBrownsFan Send message Joined: 28 Sep 01 Posts: 24 Credit: 1,705,461 RAC: 0 |
> I'm not computer literate at all, and the subject of credit claimed/given has > often made me wonder...... > > > Could anyone give a 'Joe Public' explanation of how the credit system works? > Humm I fall into the Joe Public realm so perhaps I can take a stab at this. :) Credit claimed is based on a bunch of numbers that most of Joe Public doesnt care about but the 2 that do matter are Measured floating point speed and Measured integer speed. These are used along with your CPU time to figure out the cliamed credit. Now this is where it gets fun. :) It seems to me that there are 3 basic types of users doing Bonic seti@home. 1) the type that must have the best (benchmarks, worktimes, etc.)and will use optimised clients overclocking etc to compete with others of the same mind set. These types tend to have the lowest claimed credit per unit between 3-18(although their times taken to finish a WU are not much faster than user type 2) 2) Those that dont really care and just fire up the client and let it run. They may have large farms but dont really tweek the comps, They just want it to run with out babysitting it. These types tend to have middle to high end claimed credit per unit ususally between the 20's and mid 40's. They are mostly effected by CPU and memory. (Their times to complete a WU will usually be very close to type 1 users although there credit claimed will be in some cases much higher) 3) This final set of users is the type that has a older comp or use their comp to do alot more than just crunch. This results in them having large "to complete" times and thus larger credit claims usually in the 50+ area. So what does this mean to us Joe Public? When the validator gets around to validating a unit it looks at how many units are a "match". If it finds that it has at least 3 valid (matched)results it will grant credit based on the following formula. Remove lowest claimed and highest claimed valid credits. Total remaining valid credits claimed divided by total claims = granted credit All remaining valid results submitted = granted credit from above Remember that this is done when the validator gets to the WU so sometimes that could be 3 units (which is usually the case) but can at times due to slow downs etc include more than 3. So you can look through your results files and get an estimate on what kind credit your going to get by just tossing out the high and low submitted claimed credit and taking the middle. If only 2 are submitted to find the middle just find where you fall into the group of 3 (as your claimed credit should be fairly consistant)and again remove the lowest and highest. In our 3 user system lets say 1 from each type of user ends up with "matched" valid results submitted. user 1 12,000 sec, claimed credit 12.50 user 2 13,500 sec, claimed credit 40.50 user 3 45,000 sec, claimed credit 60.00 granted credit would be 40.50 for all 3 plus any who submit the unit after granted credit is given. In every WU the only thing that matters is if there are 3 matched results and what the claimed credit is at the time of validation of those matched results. Well thats my "Joe Public" detailed answer that trys to explain why claimed/granted credits are all over the board :) <a href="http://www.teampicard.net"><img src="http://boinc.mundayweb.com/seti2/stats.php?userID=2205&trans=off"></a> |
Alex Plantema Send message Joined: 23 Oct 99 Posts: 35 Credit: 247,181 RAC: 0 |
What's claimed depends roughly on processing speed times processing time, so claims for the same unit shouldn't differ as much as in your example. That's the purpose of the benchmarks. There can be differences however, because: 1. the benchmark doesn't use the same instruction mix as the processing program 2. on Windows non-NT systems, clock time is measured, not the processor time spent 3. other programs running simultaneously on the same HT processor lead to longer processing times |
Keck_Komputers Send message Joined: 4 Jul 99 Posts: 1575 Credit: 4,152,111 RAC: 1 |
> What's claimed depends roughly on processing speed times processing time, so > claims for the same unit shouldn't differ as much as in your example. That's > the purpose of the benchmarks. There can be differences however, because: > 1. the benchmark doesn't use the same instruction mix as the processing > program > 2. on Windows non-NT systems, clock time is measured, not the processor time > spent This one should read: 2. on windows 9x systems, clock time is measured, not processor time spent. > 3. other programs running simultaneously on the same HT processor lead to > longer processing times > [edit] Ignore this post. I misread the post I quoted. BOINC WIKI BOINCing since 2002/12/8 |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
> > 2. on Windows non-NT systems, clock time is measured, not the processor > time > > spent > > This one should read: 2. on windows 9x systems, clock time is measured, not > processor time spent. > Actually, this is the same, except the 1st statement also means win3x and earlier doesn't have cpu-time, but since doesn't run seti anyway it's not really a difference. ;) Of course, for Joe Public the 2nd should be easy for win95 & win98-users to understand, but wouldn't really expect all winME-users would know this also is win9x. |
7822531 Send message Joined: 3 Apr 99 Posts: 820 Credit: 692 RAC: 0 |
This is one part follow-up from the lowball benchmark and one-part x86/ppc benchmark kvetch. I couldn't set up my Performa 6360 (ppc603e 160MHz) to run BOINC, so I put my TiBook through hell instead :-) The following is a benchmark comparison. I began with a reduced processor speed (667MHz), a normal processor speed (867MHz), activated the L2 and L3 caches (867+L2, 867+L3), and then benchmarked using single-user mode. It's all under OS X.3.7 (7S215) with 768MB. _____________ | | | | Dhry | Whet | ________|______|______|_____________________ | | | | | | | 667MHz | 213 | 397 | 51.449% | 31.990% | | 867MHz | 270 | 497 | 65.217% | 40.048% | | 867+L2 | 401 | 1183 | 96.859% | 95.326% | | 867+L3 | 412 | 1237 | 99.516% | 99.677% | | INIT 1 | 414 | 1241 | 100.000% | 100.000% | | W2KSP3 | 556 | 919 | 134.299% | 74.053% | |________|______|______|__________|__________|Even under the worst-case scenario (heavy multiprocessing apps) I haven't been able drop below the 667MHz benchmarks. So how this is possible is beyond me. Now as for the kvetch: Notice the benchmark for the Windows 2000 emulated environment. How is it possible for a "534MHz ConnectixCPU" emulation to out-crunch its 867MHz G4 host?? |
STE\/E Send message Joined: 29 Mar 03 Posts: 1137 Credit: 5,334,063 RAC: 0 |
So how this is possible is beyond me ========== Yup, he kinda got me on the Credits, I ended up getting only 17 credits for the WU I had with him, it could have been worse I guess. Luckily it's the only one I had with him though ... |
7822531 Send message Joined: 3 Apr 99 Posts: 820 Credit: 692 RAC: 0 |
I hope I didn't knock anyone else down... |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.