The Scottish Boinc Team

Mindmodeling has work! Grab it!

I'll try ....

Anyone realize MindModeling has mt (multi-threaded) app which will use all your cores? I didn't know that. Had 24 cores spend almost 3 hours on a WU for a measly 151 credits. Get 78 credits for a single threaded app running the same time.

Anyone have advice on the best apps to run or avoid?

I didn't know they had a MT app. Which sub-project is it? I want to avoid it.

I think it's all the one's with "full node" in the name

OUCH!

Hmmmm, I wonder what would happen if you ran it single threaded

With the activity I seen on the cores before I left, I would say that the windows versions only uses 1 cores worth of power anyway. Wonder if we can do the same as yafu and limit it to 2 cores max per application. As I'm away I can't test but will try and keep an eye on it. If it wastes too much then I will set NNT until I get home.

I'm trying to get some more mt WUs and see if there's hack. In the meantime, you might consider skipping the ACT-R cognitive modeling environment using Steel Bank Common Lisp WUs, as they pay about 72% the rate on average compared to Native Python v2.7 Application (Windows Only) v1.10 (sse2) WUs.

Of course, the only work they have right now is ACT-R cognitive modeling environment using Steel Bank Common Lisp.

I really hate that project! MM just sucks.

Well I'm having a ADHD attack Looking at the OcUK forums and I see they noticed TSBT creeping up on dDM so several said they will be hitting it when Enigma is over.

I've got 330k bunkered on Prima and that is enough to pass OcUK but not take the team DPC. DPC does a steady 100k/day so we wouldn't have held 2nd in the long run anyway. I'm going to go ahead and dump my bunker and then move those machines over to DDM. We have 11 hours left in the Enigma challenge so now is a good time to run it.

Here you go, an app_config.xml to use only 2 cores on the mt WUs...

<app_config>
<app_version>
<app_name>python2.7_wrap_mt_winOnly</app_name>
<plan_class>mt</plan_class>
<avg_ncpus>2</avg_ncpus>
<cmdline>--nthreads 2</cmdline>
</app_version>
<app_version>
<app_name>pypy1.9_wrap_mt</app_name>
<plan_class>mt</plan_class>
<avg_ncpus>2</avg_ncpus>
<cmdline>--nthreads 2</cmdline>
</app_version>
</app_config>

Those are the only two MM mt apps in my client_state.xml file. May be more.

Here's a preliminary look at credit with a limited number of results...

On a Xeon X5675 24 core system...
Native Pypy v1.9 Application - Full Node (Cross Platform) v3.09 (mt) running on all 24 cores at a time averages .0077 credits per sec per core
Native Pypy v1.9 Application - Full Node (Cross Platform) v3.09 (mt) running on just 2 cores at a time averages .0577 credits per sec per core

On a i7-3770k 8 core system...
ACT-R cognitive modeling environment using Clozure Common Lisp (Cross Platform) v2.35 (sse2) (1 core per WU) averages .0071 credits per sec per core

The mt WUs running on just 2 cores pay very well though I don't know if that's just a temporary thing until credit new figures it out. Still, I'd set your app_config.xml and get all the mt WUs you can.

Need to get home first then I can try this :) Still, at least I'm at work

Work available. Haven't seen any of the multi-thread WUs though.

There are some common lisp sse2 one available. Picked up a few.