Nucleon farm detail

[kladner]

Originally Posted by nucleon

It wasn't like the brochure. :)

The network map looks a lot better. :)

3x of the PCs

Pictured here:
mc - left
gauss00 - top right
gauss03 - bottom right

Turnoff white fan - mc reboots, the hottest GPU on gauss00 goes up 5degC
Turnoff black fan - gauss03 reboots

:)

-- Craig

Well that is amazing. I guess dual high-end GPUs make it kind of hard to keep things cool, even in what look like pretty well ventilated cases. Still, I would take it as a challenge to get the air flow to work closed case. But that's just one of my personal obsessions.

Are these all running on stock CPU coolers?

[Dubslow]

Crap. That's what I get for waiting an hour.

@kladner: No way a stock heatsink can boost a 2600k to 4.5 GHz. At stock speeds I was running 80C full load on stock heatsink. I get about 60C full load at 3.9 GHz on the 212+.

[retina]

Originally Posted by kladner

Are these all running on stock CPU coolers?

Did you not see the fans in the pictures?

[kladner]

Originally Posted by retina

Did you not see the fans in the pictures?

Indeed I did. Part of my comment was aimed at the fact that two of those cases had pretty serious ventilation potential, at least on the exhaust side.

@Dubslow: At least the "master control" machine has what could be a stock cooler.

EDIT: Can't see the heatsinks on the Gauss boxes. You're right that stock cooling is unlikely for those. The other two are running at stock frequency, so stock cooling is a possibility.

[Dubslow]

How many hours do you spend feeding assignments to 20 different instances of mfaktc?

[nucleon]

GPU cooling is all stock.

CPU Coolers:
gauss00 - corsair H80 (sealed water)
gauss03 - corsair H80 (sealed water)
seth - stock
mc - stock


-- Craig

[nucleon]

Originally Posted by Dubslow

How many hours do you spend feeding assignments to 20 different instances of mfaktc?

And hence that's the problem.

I have a script on each machine to distribute work evenly across the instances (see below). So that cuts down the work from 20 down to 4. But there's no guarantee that each instance does the same amount of work - so one has to shift even the worktodo.txt files occasionally. Less overall work though.

I have scripts to get work from primenet but not gpu-2-72.

-- Craig

#!/bin/bash
#usage: cat workfile | dist.sh
#evenly distributes work across i0 - i3 sub directories
#adjust max if this directory structure changes

MAX=4
n=0
while read line
 do
  echo $line >> i$(( n % MAX ))/worktodo.txt 
  n=$(( n + 1 ))
 done

[aaronhaviland]

Originally Posted by nucleon

I have a script on each machine to distribute work evenly across the instances

I do something similar, but with sed:

#!/bin/bash
BASE_DIR=~/gimps

# Distributes contents of pullwork file to each worker
SOURCE=work-source.txt

# Number of work dirs
COUNT=6

for i in $(seq 0 $(( $COUNT - 1 ))); do
        eval sed -n '${i}~${COUNT}p' ${BASE_DIR}/${SOURCE} >> ${BASE_DIR}/mfakto-${i}/worktodo.txt
done

[nucleon]

As an update.

gauss03 CPU above is out of the farm for the moment and replaced with 6core i7-3930k @4.2GHz/8GB ram@1066/Win8.

1066ram was what I had spare. I figured it's quad chan so ram upgrade could possibly wait for a while.

4cores driving GTX580 averaging 380+ GHz-days/day for last 6 days, 2 cores doing P-1. P-1 is averaging 15.8GHz-days/day.

I can't recommend intel i7-3930K highly enough.

-- Craig

[Xyzzy]

[Dubslow]

The 3930 has the H80 on it? What temps do you get at 4.2 GHz?