@MetalWarrior: I have seen machines where the variation was about 10%. I didn't make a statistical analysis of that, 5% on average was just my educated guess. Maybe the machines with larger variations have a problem with CPU time measuring, maybe it depends on how often the App needs to be restarted, and maybe it had gotten better with the BOINC code in the latest Apps or the Core Clients that are used by now. My impression was still from the beginning of S5R1. Anyway, thanks for the report, and sorry for my sloppiness.
BM
I have a server which is used as an internet router with an 'Average CPU efficiency' of 0.996479.
The cpu variation is definitively below 2%!
For CPUs with a larger variation I think that variation is related to hyperthreading processors.
Average CPU Efficiency
New for v4.70 and later.
The fraction of CPU Time compared to real time a Science Application runs, used for calculating how much work to give out.
If for example it takes 2 wall-hours to get 1 hour of CPU Time (as displayed in BOINC Manager), this will be close to 0.5.
On computers only used for web-surfing, checking emails, and the occasional file-download and listening to music, it should be close to 1.
If for example you only run the BOINC Client Software at the same time you run CPU-demanding games, it can be close to 0.
I do more of the former ^^ since games tend to be too "demanding" for my system anyway... not much fun gaming with 5 fps, which is probably good for crunching as my CPU efficiency is close to 0.9... Thanks a lot for your answer Guido, it was really helpful.
If for example it takes 2 wall-hours to get 1 hour of CPU Time (as displayed in BOINC Manager), this will be close to 0.5.
I think this measurement must be skewed when used with a HT machine. Even if my P4/HT runs nothing else the number only approaches 0.5, it's 0.6 plus or minus 0.05 usually. Note that I've only allowed Boinc to use 1 processor for power and heat considerations and the fact that 2 CPUs only gives about 15% increase in throughput. The 1 CPU may be the reason that number is only approaching 0.5, or it could be something totally different. I'd hate to think my computer is only 50% efficient, although it's possible.
I think this measurement must be skewed when used with a HT machine. Even if my P4/HT runs nothing else the number only approaches 0.5, it's 0.6 plus or minus 0.05 usually. Note that I've only allowed Boinc to use 1 processor for power and heat considerations and the fact that 2 CPUs only gives about 15% increase in throughput. The 1 CPU may be the reason that number is only approaching 0.5, or it could be something totally different. I'd hate to think my computer is only 50% efficient, although it's possible.
restricting Boinc (the science app in fact) to only 1 CPU reduces the max. CPU Efficiency to 0.5!
I think the term 'CPU Efficiency' is not totally correct here. It should say 'Host Efficiency' - but for single CPU systems this does not matter.
Akos' Apps got their speedup from many different things, not all of them being related to SSE2 or SSE3 instructions, even if the Apps ran faster on CPUs capable of these.(SNIP)
BM
Thanks alot Bernd for the detailed answer. So, I assume, the only chance of getting the S5 run to end more quickly is raw CPU force. Let's see what can be done on that side :-)
Cheers
+bb
:
your thoughts - the ways :: the knowledge - your space
:
These two number tell a lot about how well your computer is doing
Average CPU efficiency 0.995156
Result duration correction factor 0.677028
Cpu is spending 99.5% of the time actually doing work on the einstein units
DCF of 1 would be considered standard time that a unit takes to make result. The more the number goes below 1 says how well your cpu is at actually doing the crunching. If it goes above one then you are not doing quite so well and you may need to tune your system or find another project that your computer is better at crunching.
But don't get me wrong here, I'm not telling anyone to leave but you can use this info when deciding how to split up your resource shares among the projects that you do. Devoting a little more to the projects that your computer is best suited for is a good idea if efficency is important to you. However I like to base my decision mostly by what projects interest me the most.
These two number tell a lot about how well your computer is doing
Average CPU efficiency 0.995156
Result duration correction factor 0.677028
Cpu is spending 99.5% of the time actually doing work on the einstein units
DCF of 1 would be considered standard time that a unit takes to make result. The more the number goes below 1 says how well your cpu is at actually doing the crunching. If it goes above one then you are not doing quite so well and you may need to tune your system or find another project that your computer is better at crunching.
But don't get me wrong here, I'm not telling anyone to leave but you can use this info when deciding how to split up your resource shares among the projects that you do. Devoting a little more to the projects that your computer is best suited for is a good idea if efficency is important to you. However I like to base my decision mostly by what projects interest me the most.
Steve
DCF also depends on how accurate the fpops estimated set by the project is. If they way overestimate, your DCF will be less than one. If they way overestimate, then it will be over 1. I've got a project where DCFs ranged from 10E-3 to 10E3 at one point (they subsequently changed the app so that's not the case anymore).
These two number tell a lot about how well your computer is doing
Average CPU efficiency 0.995156
Result duration correction factor 0.677028
Cpu is spending 99.5% of the time actually doing work on the einstein units
DCF of 1 would be considered standard time that a unit takes to make result. The more the number goes below 1 says how well your cpu is at actually doing the crunching. If it goes above one then you are not doing quite so well and you may need to tune your system or find another project that your computer is better at crunching.
But don't get me wrong here, I'm not telling anyone to leave but you can use this info when deciding how to split up your resource shares among the projects that you do. Devoting a little more to the projects that your computer is best suited for is a good idea if efficency is important to you. However I like to base my decision mostly by what projects interest me the most.
Steve
I have a Pentium II (no SSE) running Linux and BOINC 5.4.9. My ACE is about 0.81 on both SETI and Einstein, while DCF is 0.82 on SETI app 5.12 and 0.59 on Einstein app 4.16. Does this mean that Einstein runs better than SETI on my CPU? I gave a 3:1 relative share in favour of Einstein.
Tullio
RE: @MetalWarrior: I have
)
I have a server which is used as an internet router with an 'Average CPU efficiency' of 0.996479.
The cpu variation is definitively below 2%!
For CPUs with a larger variation I think that variation is related to hyperthreading processors.
Udo
Udo
Which leads me to a somewhat
)
Which leads me to a somewhat nooblike question I've been wanting to ask for a while now: What exactly does the "average CPU efficiency" tell us?
RE: Which leads me to a
)
Extracted from BOINC-Wiki:
Udo
[edit: changed layout]
Udo
I do more of the former ^^
)
I do more of the former ^^ since games tend to be too "demanding" for my system anyway... not much fun gaming with 5 fps, which is probably good for crunching as my CPU efficiency is close to 0.9... Thanks a lot for your answer Guido, it was really helpful.
RE: If for example it
)
I think this measurement must be skewed when used with a HT machine. Even if my P4/HT runs nothing else the number only approaches 0.5, it's 0.6 plus or minus 0.05 usually. Note that I've only allowed Boinc to use 1 processor for power and heat considerations and the fact that 2 CPUs only gives about 15% increase in throughput. The 1 CPU may be the reason that number is only approaching 0.5, or it could be something totally different. I'd hate to think my computer is only 50% efficient, although it's possible.
RE: I think this
)
restricting Boinc (the science app in fact) to only 1 CPU reduces the max. CPU Efficiency to 0.5!
I think the term 'CPU Efficiency' is not totally correct here. It should say 'Host Efficiency' - but for single CPU systems this does not matter.
Udo
Udo
RE: Akos' Apps got their
)
Thanks alot Bernd for the detailed answer. So, I assume, the only chance of getting the S5 run to end more quickly is raw CPU force. Let's see what can be done on that side :-)
Cheers
+bb
:
your thoughts - the ways :: the knowledge - your space
:
These two number tell a lot
)
These two number tell a lot about how well your computer is doing
Average CPU efficiency 0.995156
Result duration correction factor 0.677028
Cpu is spending 99.5% of the time actually doing work on the einstein units
DCF of 1 would be considered standard time that a unit takes to make result. The more the number goes below 1 says how well your cpu is at actually doing the crunching. If it goes above one then you are not doing quite so well and you may need to tune your system or find another project that your computer is better at crunching.
But don't get me wrong here, I'm not telling anyone to leave but you can use this info when deciding how to split up your resource shares among the projects that you do. Devoting a little more to the projects that your computer is best suited for is a good idea if efficency is important to you. However I like to base my decision mostly by what projects interest me the most.
Steve
98SE XP2500+ @ 2.1 GHz Boinc v5.8.8
RE: These two number tell a
)
DCF also depends on how accurate the fpops estimated set by the project is. If they way overestimate, your DCF will be less than one. If they way overestimate, then it will be over 1. I've got a project where DCFs ranged from 10E-3 to 10E3 at one point (they subsequently changed the app so that's not the case anymore).
Kathryn :o)
Einstein@Home Moderator
RE: These two number tell a
)
I have a Pentium II (no SSE) running Linux and BOINC 5.4.9. My ACE is about 0.81 on both SETI and Einstein, while DCF is 0.82 on SETI app 5.12 and 0.59 on Einstein app 4.16. Does this mean that Einstein runs better than SETI on my CPU? I gave a 3:1 relative share in favour of Einstein.
Tullio