[mmonnin]

I wonder how Rosetta is going to handle the Pent. The project output is already several times higher in April than Feb with the Covid work. Add in a competition is going to make those servers scream for mercy.

 

[mmonnin]

Nice. A lot better than some projects that seem to run from someones closet. I wish their applications get that kind of work. It's been mentioned before it just keeps getting built upon and added to over and over again. I've always had more luck with the mini app vs main Rosetta app. I have it set for 1 hour but quite often tasks run for 5-6 hours then fail. When I do run the project, I end up aborting all Rosetta and run Rosetta mini as it can't be selected in user preferences.

 

[mmonnin]

I guess NF is technically a CPU/GPU project. When they released their GPU app the credit earned was a function of run time so running 2x tasks earned twice the credit and so forth. Then instead of a fixed credit or some calculation of flops performed the admin went to CreditNew and the project deflated.

 

[mmonnin]

It's CreditScrew so credit is low and isn't exactly determined by run time. Per watt, a CPU is probably better but if this is the only GPU project running it might as well be ran.

 

[mmonnin]

Setup multiple clients on one PC. Fill the queue. When one queue is about to complete, fire up another client. Repeat.
https://www.overclock.net/forum/180...uide-setting-up-multiple-boinc-instances.html

 

[mmonnin]

The NF tasks by default are set to require 0.957 CPU threads. When running more than 1 CPU they will add up over 1 and reduce the number of CPU tasks allowed to run. By default for every N # of NF tasks there are N-1 CPU tasks allowed to run.

 

[mmonnin]

What did you do to run more than 8? Running [email protected] on my 1700X and she wont go over 12 threads no matter what CPU% I change it to. My other rigs I see the extra cores being used but the 1700X aint happy for some reason. No other projects being run on that machine at the moment.

What's the memory usage situation on that PC and how much memory is BOINC set to use. There is a % number when idle and another when in use.

 

[mmonnin]

That's Rosetta for you. I always used to have them timeout after 6 hours (1 hour tasks). I had one this Pent that ran for over a day before I aborted it.

 

[mmonnin]

Ibercivis needs a min of BOINC client v7.9 to connect. 7.8 will not work.

 

[mmonnin]

Pent chat mentioned being out of work. I am not getting any either.

 

[mmonnin]

Thanks for this. I finally had the time to set this up. I have it working on my Win10 rig and I will try the Ubuntu rigs next. Bunkering around 600 projects for Numberfields between the 3 rigs, mostly GPU, and I will fire up the next client when those downloaded tasks are complete.

So if I understand this correctly I block the IP on all my clients. When those projects are complete, I close that instance of BOINC, start "BOINC2", allow the IP to get new projects and then block it again?

AS mentioned, suspending network activity is easier. A project like BOINC Tasks can manage multiple local/remote clients from one PC so there is no need to shut down 1 client to start another.

The Manager and Client are separate. Multiple clients ran run at once with separate data directories and gui_rpc_port IDs.