I have it closely watched over the past week and the Q6600 with Vista x86 took between 16.000 and 18.000 seconds to complete a S5R5 WU with BOINC 6.4.7.
The i7 with Vista x64 took between 21.000 and 24.000 seconds to complete S5R5 WU’s with BOINC 6.6.38.
Both systems ran 24/7 without any throttling, at manufactures (dell) settings.
To me this means that the Q6600 is more efficient in calculating and/or BOINC 6.6.38 has a slowing influence however I do know that it is not actually doing the crunching, it is only a manager.
You are making allowance for the hyperthreading on the i7, aren't you?
I have it closely watched over the past week and the Q6600 with Vista x86 took between 16.000 and 18.000 seconds to complete a S5R5 WU with BOINC 6.4.7.
All my hosts and all others I have inspected so far, show differences in the range of 50% in time for WUs with about the same credit. Your hosts are hidden, so I can't verify. And also time does not matter for comparison, but Credits/hour.
Quote:
The i7 with Vista x64 took between 21.000 and 24.000 seconds to complete S5R5 WU’s with BOINC 6.6.38.
Same here. Difference is far to low.
Quote:
Both systems ran 24/7 without any throttling, at manufactures (dell) settings.
To me this means that the Q6600 is more efficient in calculating and/or BOINC 6.6.38 has a slowing influence however I do know that it is not actually doing the crunching, it is only a manager.
Your Q6600 runs 4 WUs concurrently, while the i7 runs 8 WUs.
Two examples from my i7 920 running 64-bit Linux:
15,687.26 sec for 149.40 credits(S5R5)
24,940.75 sec for 149.76 credits(S5R5)
Given this high fluctuation in times for the S5R5 WUs, I think it is better to compare times with ABP1 WUs. These are pretty constant in running times.
Right now my i7 920 does 8 S5R5 WUs:
[pre]WU-name % done cl. cr. cr / h
h1_0990.55_S5R4__479_S5R5a_0 73.50 44.17 24.64
h1_0990.55_S5R4__478_S5R5a_0 54.60 44.12 20.75
h1_0990.55_S5R4__477_S5R5a_0 54.60 44.07 26.81
h1_0990.55_S5R4__476_S5R5a_0 50.50 44.02 31.74
h1_0990.55_S5R4__475_S5R5a_0 28.42 43.97 21.03
h1_0990.55_S5R4__474_S5R5a_0 11.70 43.92 21.18
h1_0990.55_S5R4__473_S5R5a_0 15.80 43.87 35.11
h1_0990.55_S5R4__472_S5R5a_0 2.14 43.82 24.46
[/pre]These WUs are all close together and will claim almost equal credits, but show huge differences in cr/h.
Right now my i7 920 does 8 S5R5 WUs:
[pre]WU-name % done cl. cr. cr / h
h1_0990.55_S5R4__479_S5R5a_0 73.50 44.17 24.64
h1_0990.55_S5R4__478_S5R5a_0 54.60 44.12 20.75
h1_0990.55_S5R4__477_S5R5a_0 54.60 44.07 26.81
h1_0990.55_S5R4__476_S5R5a_0 50.50 44.02 31.74
h1_0990.55_S5R4__475_S5R5a_0 28.42 43.97 21.03
h1_0990.55_S5R4__474_S5R5a_0 11.70 43.92 21.18
h1_0990.55_S5R4__473_S5R5a_0 15.80 43.87 35.11
h1_0990.55_S5R4__472_S5R5a_0 2.14 43.82 24.46
[/pre]These WUs are all close together and will claim almost equal credits, but show huge differences in cr/h.
cu,
Michael
The i7 has 4 cores with Hyperthreading which gives the illusion of two CPUs per core, or 8 total in this case. Hyperthreading uses some fancy footwork to accomplish this and this is one of the reasons you see the variation in processing time. THink of them as "virtual" CPUs ... The tasks are "bumping into" each other and competing for the same on chip resources... so, one of the tasks loses out and takes longer ...
This is one area where we who have had hyperthreading CPUs have requested UCB be a little more creative on scheduling the CPUs ... there is even an open Trak ticket ... the i7 will do "best" if you mix and match tasks on the system ... ideally running multiple projects with different resource needs ...
Sadly, UCB has been compromising variety for the sake of, I don't quite know what, and so this is more and more difficult. Which is really sad as Intel is increasing the number of processors with HT across the product line ...
Bottom line, the i7 and its brethren do best with a mix of computing needs, the more in the pot the better it will do ...
I think I know how HT works. ;-)
If you read my other post, you will find that line talking about all types of CPUs showing these differences in crunching times.
Taken from my AMD X2 5000:
[pre]CPU time (sec) granted credit
21,178.87 151.70
14,733.62 151.50[/pre]The differences in running times are also well documented by the project and imho still the best way to give server side credits at the moment.
I have coded a small app that is communicating with my boinc clients through RPC-calls and generating web pages for the running apps. So I can(and I do) monitor very closely what is happening.
My i7 does not show any runtime differences when crunching 8 ABP1 apps. A mixed set of S6R5 and ABP1 apps does not change the times too, so here at E@H I would say you have a perfect field running apps with HT.
With other apps it still might be a good decision to mix projects.
You are making allowance for the hyperthreading on the i7, aren't you?
Yes I am Richard.
But I did that too with several versions o BOINC 6.x.x. But after the upgrade to 6.6.38 the WU' S5R5, the ABP, and the ABP CUDA, are taking more time to complete. So the only things change in the configuration was the upgrade.
But I'll stop nagging about this and will upgrade to 6.10.3 as all currently running jobs are done.
RE: I have it closely
)
You are making allowance for the hyperthreading on the i7, aren't you?
RE: I have it closely
)
All my hosts and all others I have inspected so far, show differences in the range of 50% in time for WUs with about the same credit. Your hosts are hidden, so I can't verify. And also time does not matter for comparison, but Credits/hour.
Same here. Difference is far to low.
Your Q6600 runs 4 WUs concurrently, while the i7 runs 8 WUs.
Two examples from my i7 920 running 64-bit Linux:
15,687.26 sec for 149.40 credits(S5R5)
24,940.75 sec for 149.76 credits(S5R5)
Given this high fluctuation in times for the S5R5 WUs, I think it is better to compare times with ABP1 WUs. These are pretty constant in running times.
cu,
Michael
Right now my i7 920 does 8
)
Right now my i7 920 does 8 S5R5 WUs:
[pre]WU-name % done cl. cr. cr / h
h1_0990.55_S5R4__479_S5R5a_0 73.50 44.17 24.64
h1_0990.55_S5R4__478_S5R5a_0 54.60 44.12 20.75
h1_0990.55_S5R4__477_S5R5a_0 54.60 44.07 26.81
h1_0990.55_S5R4__476_S5R5a_0 50.50 44.02 31.74
h1_0990.55_S5R4__475_S5R5a_0 28.42 43.97 21.03
h1_0990.55_S5R4__474_S5R5a_0 11.70 43.92 21.18
h1_0990.55_S5R4__473_S5R5a_0 15.80 43.87 35.11
h1_0990.55_S5R4__472_S5R5a_0 2.14 43.82 24.46
[/pre]These WUs are all close together and will claim almost equal credits, but show huge differences in cr/h.
cu,
Michael
RE: Right now my i7 920
)
The i7 has 4 cores with Hyperthreading which gives the illusion of two CPUs per core, or 8 total in this case. Hyperthreading uses some fancy footwork to accomplish this and this is one of the reasons you see the variation in processing time. THink of them as "virtual" CPUs ... The tasks are "bumping into" each other and competing for the same on chip resources... so, one of the tasks loses out and takes longer ...
This is one area where we who have had hyperthreading CPUs have requested UCB be a little more creative on scheduling the CPUs ... there is even an open Trak ticket ... the i7 will do "best" if you mix and match tasks on the system ... ideally running multiple projects with different resource needs ...
Sadly, UCB has been compromising variety for the sake of, I don't quite know what, and so this is more and more difficult. Which is really sad as Intel is increasing the number of processors with HT across the product line ...
Bottom line, the i7 and its brethren do best with a mix of computing needs, the more in the pot the better it will do ...
I think I know how HT works.
)
I think I know how HT works. ;-)
If you read my other post, you will find that line talking about all types of CPUs showing these differences in crunching times.
Taken from my AMD X2 5000:
[pre]CPU time (sec) granted credit
21,178.87 151.70
14,733.62 151.50[/pre]The differences in running times are also well documented by the project and imho still the best way to give server side credits at the moment.
I have coded a small app that is communicating with my boinc clients through RPC-calls and generating web pages for the running apps. So I can(and I do) monitor very closely what is happening.
My i7 does not show any runtime differences when crunching 8 ABP1 apps. A mixed set of S6R5 and ABP1 apps does not change the times too, so here at E@H I would say you have a perfect field running apps with HT.
With other apps it still might be a good decision to mix projects.
cu,
Michael
RE: You are making
)
Yes I am Richard.
But I did that too with several versions o BOINC 6.x.x. But after the upgrade to 6.6.38 the WU' S5R5, the ABP, and the ABP CUDA, are taking more time to complete. So the only things change in the configuration was the upgrade.
But I'll stop nagging about this and will upgrade to 6.10.3 as all currently running jobs are done.
Thanks for the suggestions.
Greetings from
TJ