Waiting for memory

[Duner]

Hi, I'm having some problems getting Rosetta to run properly on 1 of my rigs.

It's listed as Rig 1 in my sig. Basically a AMD 620 and 2 GTX 260s.
Boinc will only run 2 apps at once and the other ones are listed as "Waiting for memory". The problem is, BOINC will also pause my CUDA work units while waiting for memory as well.

I've already increased the available memory to 100% while idle or in use, but that didn't help. I've limited the CPU to 50%, which will only run 2 instances of Rosetta, but at least it doesn't stop my CUDA units from running.

I'm running WinXP 32bit and with my 2GB of ram and the 1.79 GB from the GTX260's, I shouldn't be running into any memory limit from the OS. This seems to have started when they started releasing minirosetta 2.00 work that use up 275MB of ram each. But even then, I show >1GB of free memory in task manager, so I'm not sure what's up. The RAC harvest starts in 1 day, so I'm hoping to get this resolved soon if possible.

Thanks

 

[drshivas]

First, go into your GPU2 config and set the priority to be "slightly higher." This will prevent the F@H GPU2 from interfering with BOINC or other DC programs. I run BOINC and GPU2 at the same time w/o issue thanks to this setting.

As for the memory, I'd say let BOINC flush itself out of those big tasks, or try aborting them and you may get tasks with a smaller memory requirement.

 

[nzaneb]

It will be different than your setup drshivas, as he's crunching Seti on the GPU's through Boinc, as opposed to F@H.

I plan on running the same setup... I'll be trying to configure it tonight, so if this isn't resolved for you by then, I might have some better insight or a solution.

 

[drshivas]

ohh, missed the SETI bit. I assumed FAH.

 

[Duner]

The thing is, I've got my Rig 2 running 4 instances of Rosetta as well as 5 instances of SETI CUDA without issues. Both rigs have 2GB of ram installed and are running the big memory workunits.

I wish I had more time to trouble shoot, but I'm just a little too busy right now.

 

[nzaneb]

The thing is, I've got my Rig 2 running 4 instances of Rosetta as well as 5 instances of SETI CUDA without issues. Both rigs have 2GB of ram installed and are running the big memory workunits.

I wish I had more time to trouble shoot, but I'm just a little too busy right now.

They're configured exactly the same? Are you on the same Boinc client version?

You might try disabling the GPU's, to see if the additional CUDA WU's are effecting the memory allocation. Just uncheck the box in the preferences labeled Use GPU while computer is in use.

What was your method for running Rosie on the CPU and Seti on the GPU's?

 

[Duner]

I'm using Boinc 6.6.38 on both.

I guess could just pause seti as well and see if the Rosetta workunits restart again.

What I did was erase everything except the CUDA stuff from my app_info.xml file for SETI, that way it only crunches SETI CUDA. Then I just setup Rosetta to run as you would normally without any changes.