Different rsc_fpops_est needed?

Whenever my BOINC requests work from this project, it'll get way way more work in than my connect to + additional work are asking for.

E.g.
29/11/2011 22:49:33 | Albert@Home | [sched_op] Starting scheduler request
29/11/2011 22:49:33 | Albert@Home | Sending scheduler request: To fetch work.
29/11/2011 22:49:33 | Albert@Home | Requesting new tasks for ATI
29/11/2011 22:49:33 | Albert@Home | [sched_op] CPU work request: 0.00 seconds; 0.00 CPUs
29/11/2011 22:49:33 | Albert@Home | [sched_op] ATI work request: 51840.00 seconds; 1.00 CPUs
29/11/2011 22:49:36 | Albert@Home | Scheduler request completed: got 5 new tasks
29/11/2011 22:49:36 | Albert@Home | [sched_op] estimated total ATI task duration: 164372 seconds

These things run for 9+ hours on my GPU.
My connect to is 0.1 days and additional work request is 0.5 days. I should be getting one, maybe two GPU tasks, not five.
I see that the used in these tasks is exactly the same as you use them for the BRP3SSE running on a CPU. I don't think that's wise as these values depict in a high form how long BOINC thinks these tasks run for. It's also throwing off the duration correction factor for this project in a big way. The DCF for this project, which should hover around 1.0 for BOINC 6.12 and 6.13/7.0 versions, is 7.7498

I have noticed that you run these ATIOpenCL with the exact same of 140000000000000 as the BRP3SSE, but you forget that the value for both hardware sources are highly different. In my case:

    einsteinbinary_BRP4
    100
    windows_intelx86
    1.000000
    1.000000
    2776805609.309027
    BRP3SSE
    6.13.0


    einsteinbinary_BRP4

    109

    windows_intelx86

    0.200000

    1.000000

    33321667311.708328

    ATIOpenCL

    6.13.8

In the case of my CPU, this means that BOINC will think it needs (140000000000000 / 2776805609 = ) 50418 operations on this task. However, for GPU it lies different, as here it's only (140000000000000 / 33321667311 = ) 4201 operations. So the work estimates are already off from the beginning.

And it's very simply fixed, if you just increase the BRP4's value by ten fold or more. 1400000000000000 / 33321667311 = ~42015, which is already in the right direction when it comes to doing work. When you would increase it 15 fold, it would be ((140000000000000 * 15) / 33321667311 = ) 63022 operations.

Please fix it while you still can do so.

    p2030.20100913.G44.55+00.20.C.b5s0g0.00000_1280
    einsteinbinary_BRP4
    119
    140000000000000.000000
    2800000000000000.000000
    260000000.000000
    160000000.000000
    
-i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_248.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_249.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_250.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_251.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_252.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_253.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_254.binary -i p2030.20100913.G44.54-00.26.N.b5s0g0.00000_255.binary -t stochastic_full.bank -l p2030.20100913.G44.54-00.26.N.b5s0g0.00000.zap -o results.cand0 -o results.cand1 -o results.cand2 -o results.cand3 -o results.cand4 -o results.cand5 -o results.cand6 -o results.cand7 -c status.cpt -A 0.08 -P 4.0 -f 400.0 -W -z
    


    PM0062_01121.dm_396

    einsteinbinary_BRP3

    105

    140000000000000.000000

    2800000000000000.000000

    260000000.000000

    160000000.000000

    

-t stoch_PMPS.bank -l zaplist_PMPS.txt -A 0.08 -P 3.0 -f 130.0 -W -z -i PM0062_01121_396.binary -i PM0062_01121_397.binary -i PM0062_01121_398.binary -i PM0062_01121_399.binary -c status.cpt -o results.cand0 -o results.cand1 -o results.cand2 -o results.cand3

    

(terminology for the easily confused, there's no seconds used in any of the calculations or outcomes. fpops means floating point operations, while flops means floating point operations. Do not confuse either with floating point operations per second, as this will be abbreviated to FLOPs (capitals plus lower case 's').)