Estimating ratio for CPU and GPU

Started by Tim Norton, July 27, 2010, 06:14:35 PM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Tim Norton

Fred

we have the option to put in ratios for cpu and gpu

whats the basis for these numbers so i can put the correct values in

you thread only show 0.5 to 1.5 for both

how do we decide what the values should be?

Thanks

Tim
Thanks

Tim

Tim Norton

Not sure why you moved this - but no problem

i was referring to reschedule not Boinc tasks ?
Thanks

Tim

fred

Quote from: Tim Norton on July 27, 2010, 06:38:50 PM
Not sure why you moved this - but no problem

i was referring to reschedule not Boinc tasks ?
Sorry about that I just moved it back, was a bit confused.

rsc_fpops_est is the value that determines the estimate speed of the CPU or GPU that runs the task.

The scheduler calculates the value from the workunit data file the same way the Server does. Depending on the AR.
The new server program will adjust this value depending on the speed. To adjust the runtime to a proper value and move the duration factor to 1.

The ratio is the difference between the calculated value and the value the Server sends.
At the moment the ratio is 1.
I use a ratio of 4 for the CPU and GPU at the moment for testing and it moves the duration factor from 0.01 close to 1 on my computer.
So higher is a faster computer.
The rsc_fpops_bound is the max run time for the task, beyond that it gives a -177 error.
So when you change the ratio, better check the the "limit  rsc_fpops_bound" to avoid any -177 errors.

<workunit>
    <name>18mr10aa.6205.9883.16.10.69</name>
    <app_name>setiathome_enhanced</app_name>
    <version_num>528</version_num>
    <rsc_fpops_est>161122762245845.000000</rsc_fpops_est>
    <rsc_fpops_bound>1611227622458450.000000</rsc_fpops_bound>
    <rsc_memory_bound>33554432.000000</rsc_memory_bound>
    <rsc_disk_bound>33554432.000000</rsc_disk_bound>
    <file_ref>
        <file_name>18mr10aa.6205.9883.16.10.69</file_name>
        <open_name>work_unit.sah</open_name>
    </file_ref>
</workunit>

Tim Norton

Ah Ok i think i follow you

looks like i need to read up on rsc_flops est and bound before i change anything

checking up it appears for Seti the bound is 10 times the estimate - milkyway looks like 100 times

one thing do you mean your ratio of 4 increases with faster computers - just out of interest how fast are your cpu and gpu so i can gauge mine etc

mine are gtx 260 and i5 655k @ 3.3 (at the moment)

Thanks again (and i will try and get my posts in the correct place first next time  ;D)
Thanks

Tim

fred

Quote from: Tim Norton on July 27, 2010, 07:19:49 PM
Ah Ok i think i follow you

looks like i need to read up on rsc_flops est and bound before i change anything

checking up it appears for Seti the bound is 10 times the estimate - milkyway looks like 100 times

one thing do you mean your ratio of 4 increases with faster computers - just out of interest how fast are your cpu and gpu so i can gauge mine etc

mine are gtx 260 and i5 655k @ 3.3 (at the moment)

Thanks again (and i will try and get my posts in the correct place first next time  ;D)
I7 + GTX 295
The bound value at Seti is *10 I use *20 while rescheduling, when setting the -177 check the value is way higher.
But nothing dramatic happens when you change the est ratio, just the estimate times change, but not linear.
As long as you keep the -177 check set when you are doing that.
Because the DCF may move a bit and can cause the -177 timeout error.

And your post was in the correct location, I was just sleeping. ;D