Why do granted credits vary so much between finished work-units (and seem to be uncorrelated to the cpu-time spent)?

Paul Vleugels
Paul Vleugels
Joined: 1 Mar 05
Posts: 12
Credit: 32,281,062
RAC: 10,039
Topic 216844

In the last 3 weeks of October 2018 I noticed an extreme difference in the granted credits for finished work-units.

Showing just two of the processed work-units:

a. task id: LATeah0104R_852.0_0_0.0_1240942_0, workunit id: 374164496, run time: 2733, cpu time: 2620 => granted credit: 3465
b. task id: LATeah0042F_1288.0_271286_0.0_1, workunit id: 374163184, run time: 28353, cpu time: 18706 => granted credit: 693

How can it be that cpu times of 2620 and 18706 result in 3465 respectively 693 points? The ratio between these 2 examples is completely lost.
Why spend so much computing time when there's (almost) no credit to earn - automatically expecting that the outcome of these calculations are less or of no interest?

Is there a way to get just the credit work-units which will be highly valued only?

Richie
Richie
Joined: 7 Mar 14
Posts: 656
Credit: 1,702,989,778
RAC: 0

Paul Vleugels wrote:a. task

Paul Vleugels wrote:

a. task id: LATeah0104R_852.0_0_0.0_1240942_0, workunit id: 374164496, run time: 2733, cpu time: 2620 => granted credit: 3465

b. task id: LATeah0042F_1288.0_271286_0.0_1, workunit id: 374163184, run time: 28353, cpu time: 18706 => granted credit: 693

How can it be that cpu times of 2620 and 18706 result in 3465 respectively 693 points?

That's because the first task was run by NVIDIA GeForce GTX 950M GPU in your computer. The second task was run by only the CPU in your computer.

Those two methods are totally different in nature to each other. They are different searches also:

Gamma-ray pulsar binary search #1 on GPUs v1.20

Gamma-ray pulsar search #5 v1.0

GPU tasks are usually faster than CPU tasks these days. Computing power in modern GPU's is enormous and when the application is optimized for GPU then it will work much faster than CPU overall. GPU need some assistance from the CPU while running a task, but "Run time" is more meaningful to look at. There's no point to compare only the CPU times.

Quote:
Is there a way to get just the credit work-units which will be highly valued only?

Yes. You can go to settings and choose to run only Nvidia GPU and Gamma-ray pulsar binary search #1 on GPUs v1.20 on that host.

Go to your 'Acoount' -- 'Preferences' -- 'Project' and choose a location for a "Preference set". 

Set 'No' for "Use CPU" and 'Yes' for "Use Nvidia GPU".

On the Applications list check only the "Gamma-ray pulsar binary search #1".

Set 'No' for "Run CPU versions of applications for which GPU versions are available".

Set 'No' for "Allow non-preferred apps".

Save changes (at the bottom of the page).

 

Now, go to your host page (that one with Nvidia GPU): https://einsteinathome.org/host/12535157

At the bottom there is 'Location'. Choose same location for that host as you saved the settings earlier ('Preference set'). Restart Boinc. That host should receive only GPU tasks from now on.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5,870
Credit: 115,678,913,575
RAC: 34,669,418

Paul Vleugels wrote:In the

Paul Vleugels wrote:
In the last 3 weeks of October 2018 I noticed an extreme difference in the granted credits for finished work-units.

This situation has existed for years - not just the last three weeks! :-).

There are two entirely different searches involved.  One search (FGRP5 {your example b.}) uses a CPU core for processing,  The other (FGRPB1G {your example a.}) uses a suitable GPU (if present) to do the processing.  The CPU time you quote (2620s) is not processing time - it's basically the time that a CPU core was on standby waiting to supply support to the GPU when needed.  You can't receive tasks for this second search if you don't have a suitable GPU.

Because you have allowed both searches in your project preferences, you are receiving both types of work.  Different types of gamma-ray pulsar systems are being searched for even though the data still comes from the same source - the Large Area Telescope (LAT) on board NASA's Fermi satellite.  You can get more information about these searches and what has been discovered so far if you follow some of the links under the 'Science' menu item.

To get more information about why the credit awards are different, try reading this thread where I've answered some similar questions.  There is also a link there to a further answer in yet another thread about credit.  If these other posts don't answer any specific point you might have, please come back and ask a further question.

Paul Vleugels wrote:
How can it be that cpu times of 2620 and 18706 result in 3465 respectively 693 points? The ratio between these 2 examples is completely lost.

You are looking at the wrong ratio :-).  The 'work content' of a GPU task is five times as large as the work content of a CPU task.  So the ratio you should be looking at is 5x693=3465.  Five times the work content leads to 5 times the credit award.

Paul Vleugels wrote:
Why spend so much computing time when there's (almost) no credit to earn - automatically expecting that the outcome of these calculations are less or of no interest?

There's plenty of interest in the results that are produced from both search types.  You can't judge the 'value' of results by the credit  award.  One type of search can take advantage of the 'massively parallel' compute ability of modern GPUs whilst the other can't.  So one can complete more work in a shorter time whilst the other can't.  Both are needed.  Do you want to help the scientists process their data or don't you?

Paul Vleugels wrote:
Is there a way to get just the credit work-units which will be highly valued only?

Just use your project preferences to chose whichever searches you want to contribute to.  On your account page on the website, follow the menus and sub-menus - Account -> Preferences -> Project and scroll down to 'Applications' where you will see tick-boxes for each search type.  If you make changes, don't forget to go to the bottom of the page and click the 'save changes' link.

 

Cheers,
Gary.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.