Let me change one thing ... i said above "with the hardware i have"

I might make some small changes in hardware if it mades a big difference
in productivity

for example ... if the "work unit threads" run on individual FPOs ... well
6174s have 48 while 6276s have 32 .... so i can run more 16 more (-1 for GPU) on the 6174s ...

(yeah i know 6386s ... but they are out side of the budget)

or if the program needs 1 gig of ram per core to not use the swap file
i might move some RAM chips around or upgrade some

or if HD speed makes a big difference i might upgrade to some small SSDs

or maybe a different OS is better

some stuff like that ... small tweek, big payoff

ill listen to any suggestion ... except switching teams

CEP2 is rather write intensive so If I were building a rig specifically the crunch this I would be thinking in terms of running a significant sized ram disk. I would also be considering an ssd that is capable of big numbers over its lifetime.

Ubuntu is fine

No real knowledge of AMD to add anything useful

well SekeRob* .... i have a BUNCH of work Units that have "computational errors" in under 3 minutes
but also i have 19 running simultaneously ... on 32 cores ...
so ...
this server is not making enough Heat to make me happy

CEP2 workunits, have long intervals between checkpoints.
So, if you are running the boxes 24/7, you'll have an initial need to micromange the startup to reduce I/O, but after some time, the read/writes will settle down to a manageable level (I think) as they'll eventually start staggering.
If you shut down to computers often, with this many workunits will lead to a lot of lost computing time and constantly managing I/O.

I highly recommend a small SSD. My two big systems are not dedicated, but I bought a 32gb SSD for each to keep WCG separate from my OS drive.

As for the cores being limited, I had to add a cc_config like such


<?xml version="1.0"?>

-<cc_config>


-<options>

<ncpus>48</ncpus>

</options>

</cc_config>


have a search online and you'll find a better explanation for your needs

Since you've got plenty of RAM memory and your installed OS is Linux based I suggest you to mount BOINC data folder in RAM for bypassing I/O problem: doing so I've managed to run smoothly and without errors 7 CEP2 WUs plus 1 UGM WU at a time on a notebook equipped with 8 thread processor, 8 GB RAM and 1 GB swap partition. Otherwise I couldn't run more than one CEP2 WU simultaneously since my HDD isn't responsive enough and the system completely hung for a couple of minutes when WUs were starting simultaneously.

Each CEP2 WU needs 170 MB RAM to run plus 700 MB more for mounting the folder containing its thousands files. The downside being you'll lose all the work in progress whenever you reboot the system or a blackout occours since BOINC data folder is stored in RAM instead of HDD. But if you're crunching 24/7 it's not your case.
Surely there are some ways to set up a periodical backup system and to be able to suspend or ibernate the system without losing anything but I'm not interested in them so I can't advise you about that.

In Linux Mint I've used the following code in etc\fstab file to mount BOINC data folder in RAM:
If some WUs error out and the following error message appears in their log files in Results Status page of your account at WCG site you have to increase the size of tmpfs or the size of swap partition or swap file.