PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise
1) Message boards : Number crunching : Pendings (Message 128443)
Posted 931 days ago by BarryAZProject donor
I've noticed a drop in my RAC numbers balanced by a rise in pendings over the past 10 days or so.

My RAC has dropped from 150K+ to 130K + while at the same time my pendings have increased from 52K or so to 265K.

Do I have company in this observation??
2) Message boards : Number crunching : PrimeGrid in a multi-project environment (Message 94564)
Posted 1998 days ago by BarryAZProject donor
Thanks for the reply -- I'm one of the relative long timers -- joined in early 2014.

I have taken to changing the settings to short only now, plus CPU only.

I did that a few weeks ago and likely will retain those settings. My RAC has settled into the 115K range (quite a few workstations in the farm).

3) Message boards : Number crunching : PrimeGrid in a multi-project environment (Message 94553)
Posted 1999 days ago by BarryAZProject donor
Over the months I've been working with PrimeGrid I've observed that the work units downloaded often include work units that go from a 15 hour process cycle to a 100 hour process cycle at some point. I realize that when that happens to due date is moved out as well so if the work is submitted late, one still gets the credit.

My experience though is that while this works when PrimeGrid is the only project, it can be quite problematic when other CPU projects are in the mix.

For me, those other projects include Rosetta, POEM, SETI, Einstein and World Grid.

It is a case that PrimeGrid doesn't play well with others.

My approach to this has been to limit the type of work units from Prime Grid (the available choices are extensive). As long as I limit the Prime Grid work to the short run work units, that works fine and has stabilized what I complete and the credit I get from Prime Grid.

Any comments on this?
4) Message boards : Problems and Help : Stalled downloads (Message 86248)
Posted 2314 days ago by BarryAZProject donor
Thanks -- seems to have done the trick.
5) Message boards : Problems and Help : Stalled downloads (Message 86232)
Posted 2315 days ago by BarryAZProject donor
I have encountered a batch fo stalled downloads on multiple workstations -- they actually stay stalled forever.

I've found in some cases that detatching and reattaching allows other prime grid tasks to download but there seems a specific range of the downloads that simply don't complete.

Has anyone else run into this in the past week?
6) Message boards : Number crunching : ATI video card problems (Message 77747)
Posted 2665 days ago by BarryAZProject donor
OK-- I take the problems with the AMD HD 6000 series are no surprise to folks even though the card, unlike the HD 4000 series is supported with the most recent AMD drivers, instead of the legacy drivers for the HD 4000 series
7) Message boards : Number crunching : ATI video card problems (Message 77669)
Posted 2668 days ago by BarryAZProject donor
I note the home page indicates issues with HD 4000 series cards.

Recently (in the past few weeks), I've observed a similar problem with HD 6000 series cards as well. Tasks go to computation error in seconds.

With the HD 4000 series I realize that they are legacy cards with legacy drivers.

The HD 6000 series cards are a bit more current and use the current AMD/ATI drivers.

I've encountered the problem with the two systems using these cards -- both cards are HD 6650. One is running on a Windows 7 - 64 bit system, the other on a Windows 8.1 64 bit system. I started observing the problem a couple of weeks ago, I've since updated to the current AMD drivers, stop/started BOINC (current version), reset Prime, did a workstation shut down and restart and the issue persists.

For now I simply run PrimeGrid CPU tasks on those systems and use other Projects for GPU tasks on those two systems.

Again, prior to a couple of weeks ago, I didn't see this with those two systems.

8) Message boards : Number crunching : Estimated time to complete and due dates (Message 75228)
Posted 2749 days ago by BarryAZProject donor
OK -- I've noticed that with BOINC and a couple of other projects.

For some projects, bumping up the time to network directly affects queues, sometimes to a greater extent than the actual queue setting.

In any event, for 'pure Prime' workstations, I'm not watching them as closely to see if there are any potential 'overdues'.

On those workstations where there are multiple projects along with Prime, I will keep an eye out there. But knowing that overdues (if only slightly overdue) are more likely to get credit, eases things a bit there as well.

Thanks for the frequent replies -- they are appreciated.
9) Message boards : Number crunching : Estimated time to complete and due dates (Message 75217)
Posted 2750 days ago by BarryAZProject donor
Regarding multiple projects -- I suffer a bit from being a multitasker -- so on many workstations I don't have so much as a prime and backup as multiple projects with shared support.

For most projects this isn't an issue -- seems Prime likes to be Prime a bit more -- and what I've done there is to have a handful of 'Prime' computers where Prime is either the only project running (both GPU and CPU), or it is the only project running (CPU), with others which support the HD4850 using the GPU.

My queues are typically set to 1.5 days (used to be a time when they were higher -- but not any more). The network report cycle is typically set to 1 day -- but that is something I tend to do manually periodically.

I didn't know about the 'grace' period for reporting -- good to know.



10) Message boards : Number crunching : Estimated time to complete and due dates (Message 75192)
Posted 2750 days ago by BarryAZProject donor
Fair enough -- guilty as charged here -- multiple projects and BOINC is not doing a good job of prioritizing.

I'll kill off the task.

I've been running BOINC for 10 years, only joined Prime Grid earlier this year and so typically do have multiple projects.

What you are suggesting is my approach of deleting tasks which appear to not have good chance of success, actually makes more sense on the multi-project set ups.

I'd note of course that one of the concepts behind BOINC was multi-project support. Back in the day when both SETI and Einstein had periodic issues, that made a lot of sense. Even today, project outages do happen, so mulitple project support makes sense.

I've a couple of workstations that are only running PrimeGrid -- both CPU and GPU, and a couple of others that are only running PrimeGride - CPU (these are systems with HD4850's). I will look at those for 'too late to finish' scenarios.

I'd note that when I set Prime to get new work, sometimes it seems to download too much -- it downloads a lot of short run, close in due date work which goes high priority, and pushes longer run work to the back -- in those cases, the longer run work might not get to the front of the queue soon enough even when only Prime is running. That is, BOINC is getting too much work. I should look at the queue times I set up there to affect that I suppose.


Next 10 posts
[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 4.46, 3.39, 2.92
Generated 16 Oct 2021 | 19:08:27 UTC