UT theoretical chemistry code forum

With a runtime of almost 5.000 seconds I get around 2,5 points using 3.04 on Windows 7
When I look at the top computer I see about 3.500 seconds runtime and around 18 points. What am I doing wrong?

I posted my experience in regard of credits in the other thread already, but here's now with the 3.04 update:

I too am receiving shortened work unit points values, There seems to be a decline in the value of the points given in the last 12 hours. Also I am running an AMD processor and the intel units running are returning work units in 1/2 the time and receiving almost twice the points values!

I switched to Linux today, still 3.04. The loop time and CPU time are a bit different now.
For a runtime of around 5.000 I now get around 50 points, about 23x times as much. Uhm...

I think something is really wrong with the credit calculation of the workunit since v3.04!
Okay maybe Saenger has no differences but others have differences for sure.
So it would be no fault for the project admins to take a deeper look at this issue.

Not that the credit is valued higher than the science, but it is necessary for a fair competition.
I can not believe that the runtime of an Intel M740 is valued higher than an Intel Q6600.

Maybe some of the project admins could explain their point of view or their credit-algorithm.

I have a Windows 7 64-bit machine, several Linux machines and a Mac working part-time on this project. The Windows 7 one is producing considerably less points than the others for the amount of CPU time, so it would appear the credit calculation issue is with the Windows version. I have set it to no new tasks for now as it's pretty much a waste of time. I'll let the others continue to crunch on it.

I am seeing a readjustment in the points values, and a more equal operating field of values dealing with INtel and AMD based systems. While the correction is not great, it is much better to see 3 hours of work performing and receiving value in respect to pre-long-form work units, then dismal credit values irritating the masses. 'Thank You

I would just like to comment that BOINC's credit system is very complicated and doesn't always seem fair to me. I would think that credit should look something like the product of the CPU time spent on the work unit and the floating point operations per second that the cpu is benchmarked at.

We don't adjust or modify anything to do with the credit system. In fact, I'm not sure how we would without modifying the server software's source code.

Still, the points difference between Windows and Linux is way too huge, just for the same runtime.

Stef42 wrote:Still, the points difference between Windows and Linux is way too huge, just for the same runtime.

I'd be interested in seeing the difference in the benchmark on windows and linux for the same computer. I remember reading the the BOINC client benchmark isn't as optimized on linux as it is on windows.

Well, I can say that on the same clockspeed Linux gives me around 10% higher floating points. Around 3100 vs. 2600.
The integer speed is à big difference, 6000 on windows 7 and about 18500 on linux. Big difference!

The credit is getting lower on my windows machine. Same runtime, but only half of the credits i had before (and this was not even much as in older version)

Since a few hours the updated 3.05 went online, runtime's are cut by around 50% for me.
The overall credit has dropped a bit, around 10%.

I was receiving fairly good points per hour over the last few months so this was a set and forget type of project.
Or so I thought.
Did a check a week or so ago and found that one of my computers (only one out of four), had started to give really low credit even though it was my fastest computer (5 cr/h).
I stopped that one as the other three were still giving good credit.
Now two other computers have started to give the same low credit as well (between 5 and 9 cr/h).

Now with every new work unit giving "Maximum Disk Usage Error" on all computers and every work unit has an expected run time of over 100 hours (only takes 45 minutes), things have gone from good to average to bad.
This is happening on both Linux and Windows.

I have set the project to NNW . I am putting in lots of hours but getting very little back, I was hoping to get past 300,000 but that will now have to wait till things have been fixed.

Conan

My Observations have now got a rotten core as the points values are decreasing again. This time across both platforms both intel and AMD are suffering. Plus there are a lot of "Error while computing" on the intel machine and the AMD is just barely seeing (12 Point) for 1 hour and 40 minutes of task. (The AMD used to see 18 points an hour) That's dreadful, you should be ashamed of the work our electrons are doing to hand out such low points values. CHANGE is needed Gentlemen!