I’ve received some VLAR guppi WUs assigned to run on v8.01 (cuda50).  These appear in my cache sporadically.  Comparatively, they run 75% longer than simmilar WUs assigned to v8.12 (opencl_nvidia_sah).  I know this may not have anything to do with the app being assigned.  Is this simply a matter of the server attempting to determine which app/WU combo is more efficient on my rig?  I’ve read that the server will do this on occasion.  Does anyone know the frequency at which the server performs this action?  If so, why use an old app?

Since this is the testing ground for the apps, changes are made not only to the apps but to the data as well.

More recently, the size of the data has increased.

On main we commented about how BLC GUPPI being bigger than typical work units from Arecibo, but they were about 350mb total size. More recently here on beta they were expanded to 700mb

This change in size has to be tested on all the apps to make sure they will still work as they are supposed to before they do the same on Main.

Just recently, they were talking about the sanity check that occurs with the work units.

The next build is supposed to remove that sanity check,

Again, it will have to be tested on the apps before it goes to main.

So all the apps are rerun with the changes to make sure they work.

If you don't want a particular app version to run, you can just abort it.

I do that when we know already they will run on cuda 42 or 50 but need data on Opencl versions.

Hope that helps.

Zalster

... BLC GUPPI being bigger than typical work units from Arecibo, but they were about 350mb total size. More recently here on beta they were expanded to 700mb

I think you mean this.

Just recently, they were talking about the sanity check that occurs with the work units.

Right, I’m up to speed with SETI@home v8.12 Windows GPU applications support thread.

The next build is supposed to remove that sanity check

That would be [url+http://setiathome.berkeley.edu/forum_thread.php?id=79760&postid=1799327]r3480[/url].

If you don’t want a particular app version to run, you can just abort it.

I’m not going to become cranky about app versions like some contributors did when guppi WUs were introduced.  I understand that it is a beta project.

I do that when we know already they will run on cuda 42 or 50 but need data on Opencl versions.

Understandable, because you are more proactive with the project and communicate with Raistmer.

So all the apps are rerun with the changes to make sure they work.

That helps. I wasn’t thinking as granular as was necessary.  My participation here is running the apps in stock form, with no added options. However, I posted because I don’t recall any posts specifically mentioning a particular app and its adverse quantitative difference in runtime from its non-beta usage.

I’ve received some VLAR guppi WUs assigned to run on v8.01 (cuda50).  These appear in my cache sporadically.  Comparatively, they run 75% longer than simmilar WUs assigned to v8.12 (opencl_nvidia_sah).  I know this may not have anything to do with the app being assigned.  Is this simply a matter of the server attempting to determine which app/WU combo is more efficient on my rig?  I’ve read that the server will do this on occasion.  Does anyone know the frequency at which the server performs this action?  If so, why use an old app?

The frequency depends on relatively builds performance (as server measures it) and from number of already completed tasks. So, number of "inefficient" tasks will decrease over time.
Also, on beta worth to check all apps time to time. On main to prevent such cases worth to switch to anonymous platform (cause you are active participant and aware of changes project has time to time) and install app of desire.

---

News about SETI opt app releases: https://twitter.com/Raistmer