Really strange results. The credits aren't even close to proportional to the time spent on the wus.

In the last 2 days, I've seen the following, all on the same box with the same cards:

# Time per step: 44.911 ms
# Approximate elapsed time for entire WU: 22455.469 s

Granted Credit: 1887.96527777778

-------------------------------------------------

# Time per step: 34.907 ms
# Approximate elapsed time for entire WU: 17453.613 s
called boinc_finish


granted credit 2435.94444444444

-------------------------------------------------


# Time per step: 27.683 ms
# Approximate elapsed time for entire WU: 23530.125 s

granted credit: 3232.06365740741

---

I have seen the same at my WU's. Maybe new credit regulations??

---

They have ever been proportional to time, it's fixed credit for every WU.

gdf

How are they granted?? Before for me it looked always as you get the same credit (3,232.06) for each WU.

I don't see how they are granted when i look to my last WU's.

Time per step: 56.057 ms
Approximate elapsed time for entire WU: 28028.304 s
CPU-time: 15,081.44 s
Granted credit: 1,887.97

Time per step: 33.296 ms
Approximate elapsed time for entire WU: 16648.047 s
CPU-time: 12,795.25 s
Granted credit: 2,435.94

Time per step: 45.582 ms
Approximate elapsed time for entire WU: 38744.812 s
CPU-time: 23,842.38 s
Granted credit: 3,232.06

---

There are at minimum 3 different type of WUs. Look at the run time, it's also different. Your PCs are hidden, so I can' look at your run time for the CPU. Otherwise I would calculate the runtime from the value of

# Approximate elapsed time for entire WU: 18088.121 s

in your task data. This WU from my GTX260 was running 18088.121 seconds and I got 2435.94 credits for it.

2435.94 / 18088.121 * 3600 = 484.81 cr/h.

Thid is the way to compare the WU, not by compare the credits directly.

---

I have also been discussing this in another thread (Curious Performance Difference) under Number crunching. Don't know if PS3's have the same problem or if it's just Nvidia cards but my lower granted credits started on the 10th of Dec and about the same time all these DCF and no cell processor/no work problems started.

Kokomiko: If I calculate my credit for two very different WU I get these answers,

Task 127339 - 2435.94 / 5822.16 * 3600 = 1506.20 cr/h (GTX280)

Task 126687 - 2435.94 / 23082.91 * 3600 = 379.9 cr/h (8800GT)

GDF has explained that the credit issued is fixed therefore no matter what time it takes you should still be awarded 3232.06 for the WU.

Just another problem to fix. Good luck GDF

GDF: Maybe should confine this problem to one thread before it gets out of hand?

Cheers

Pat

Task 127339 - 2435.94 / 5822.16 * 3600 = 1506.20 cr/h (GTX280)

I think you take the wrong value. Can't believe, that this WU was worn out in 5822.16 seconds, thats only 1:37h. The normal runtime for this type of WU is on a GTX280 5:09h.

---

The credit is fixed on the workunit, but it depends of the workunit. So, there are wu which are longer and shorter and give more or less credits. All the number here are fine. A 280 gets 4 times more credits than a 8800 in average.

We try to split work depending on the molecular system such that it takes 12H on a 8800GT approximatively.

gdf

Hey GDF,

I'd like to believe that but I have 2x280's and 2x8800GT's and I have compared CPU time to granted credit and even elapse WU time to granted credit and I just don't get it. How can a WU take 40102 sec and be granted 2435 credit and then a WU that takes 17497 sec be granted 1887 compared to another WU that takes 39580 sec and gets 3232. ALL are in elapsed time because I was told CPU time is not the way to judge. My 280's normally process a WU in 5-6hrs consistently, whereas my 8800GT's normally process a WU in 11-13hrs consistantly. I am still doing the same number of WU in a day but am getting less credit when technically if I have shorter WU I should be doing more WU's and getting the same RAC as before. Logically of course. Is there a limitation on the amount of WU issued to one system per day?

Pat

Is there a limitation on the amount of WU issued to one system per day?

Pat

Yes, 15 per day. It's to protect the project server if a client is bad and call one WU after the other and crash them. You can control this for your box here. In the line

Maximum daily WU quota per CPU 15/day

you find this value for your X9650.

---

Sweet!

Thanks

You can always learn something new every day!!

Pat

From the different times per step it seems like the systems differ in complexity and number of steps. And the credits seem granted according to number of steps.

# Time per step: 44.911 ms
# Approximate elapsed time for entire WU: 22455.469 s

Granted Credit: 1887.96527777778

Few long steps.

# Time per step: 27.683 ms
# Approximate elapsed time for entire WU: 23530.125 s

granted credit: 3232.06365740741

About twice as many steps, system as complex as in the previous runs.

MrS

---

Scanning for our furry friends since Jan 2002

Can't say that I'm fond of this new credit method. It only started on the 11th of December. and in the last 70-80 WU's under half where full credit and my rigs still complete the same number of WU/day with less credit.

Pat

This should not be the case. If it is there is an mistake.
The credit system has not changed it is just that now we have several different molecular structures on going with more or less atoms, for more or less iterations.

It should produce approximately the same credit/day irrespective of the type of WU on the same PC.

gdf

Could it be, that the entry

# Approximate elapsed time for entire WU: 18305.484 s

is no longer the real time for elapsed WU but rather the CPU time since the new application is running?

---

is no longer the real time for elapsed WU but rather the CPU time since the new application is running?

CPU time is still different than reported elapsed time.

GDF,

could it be that the complexitiy of the model is omitted in the credit calculation and only number of steps is taken into account?

MrS

---

Scanning for our furry friends since Jan 2002

For the same elapsed time on the same card they should give similar credits irrespective of the WU. Provide that the estimates are accurate.
The time per step is computed on the last run only, so it cannot count if you were playing games before. The total elapsed time simply multiplies by the number of iterations.

Now, this seems not to be the case there. How many GPUs has the PC and what are the names of the WUs?

gdf

Task 170388 - SM21632-SH2_USPME-1-40-SH2_USPME2020000_0 has a approximated time of 39646.065 s under the 6.55, task 170472 - hg19408-SH2_USPME-1-40-SH2_USPME1400000_0 18391.686 s under the 6.54. Both are running on the same box, no games while running and the same credits.

---

This should be do the client version.

gdf

All WUs with 6.55 and running 4+1, no gaming or heavy use

3232 credits
# Time per step: 58.789 ms
# Approximate elapsed time for entire WU: 49970.359 s

2436 credits
# Time per step: 69.752 ms
# Approximate elapsed time for entire WU: 34875.875 s

1888 credits
# Time per step: 68.344 ms
# Approximate elapsed time for entire WU: 34171.750 s

Scaling for the first 2 kinds is OK and I have 4 more WUs with approximately similar times, but the one for 1888 credits doesn't fit.

MrS

---

Scanning for our furry friends since Jan 2002