Stefan, I see 4 plus one which says psi.30091.clean
Tullio

Im fixing an issue with SELE2 so I cancelled them and will send out SELE3 in a bit

Im fixing an issue with SELE2 so I cancelled them and will send out SELE3 in a bit

Is this related to the 'upload failure - file size too big' problem reported for SELE2 last week? Whether or not, please double-check the <max_nbytes> value for the new batch.

I had to add WCG to this 48-thread beast because it isn't using all of the threads @ 100% when running GPUGRID only. I'd wager it's because of scratch space bottleneck (it's running a SSD tho, 200 MB/s according to hdparm)... ?

---

No, the issue was with an old version of psi4 giving wrong results on large molecules when using the scratch space. This is fixed in the latest version now.

No, the issue was with an old version of psi4 giving wrong results on large molecules when using the scratch space. This is fixed in the latest version now.

I'll set WCG to don't allow new work and I'll report back!

---

I am running 3.31 SELE6.
Tullio

Something is wrong. The BOINC Manager says it is running but python does not appear in the "top" console.

---

I just had about 20 of these fly through before they corrected and started to run correctly.

<core_client_version>7.8.3</core_client_version>
<![CDATA[
<message>
process exited with code 195 (0xc3, -61)</message>
<stderr_txt>
10:09:04 (14352): wrapper (7.7.26016): starting
10:09:04 (14352): wrapper (7.7.26016): starting
10:09:04 (14352): wrapper: running /usr/bin/flock (/home/zalster/Desktop/BOINC/projects/www.gpugrid.net/miniconda.lock -c "/bin/bash ./miniconda-installer.sh -b -u -p /home/zalster/Desktop/BOINC/projects/www.gpugrid.net/miniconda &&
/home/zalster/Desktop/BOINC/projects/www.gpugrid.net/miniconda/bin/conda install -m -y -n qmml2 --override-channels -c defaults -c gpugrid --file requirements.txt ")
Python 3.6.5 :: Anaconda, Inc.

PackagesNotFoundError: The following packages are not available from current channels:

- psi4==1.2.1

Current channels:

- https://repo.anaconda.com/pkgs/main/linux-64
- https://repo.anaconda.com/pkgs/main/noarch
- https://repo.anaconda.com/pkgs/free/linux-64
- https://repo.anaconda.com/pkgs/free/noarch
- https://repo.anaconda.com/pkgs/r/linux-64
- https://repo.anaconda.com/pkgs/r/noarch
- https://repo.anaconda.com/pkgs/pro/linux-64
- https://repo.anaconda.com/pkgs/pro/noarch
- https://conda.anaconda.org/gpugrid/linux-64
- https://conda.anaconda.org/gpugrid/noarch

To search for alternate channels that may provide the conda package you're
looking for, navigate to

https://anaconda.org

and use the search bar at the top of the page.


10:09:21 (14352): /usr/bin/flock exited; CPU time 11.828434
10:09:21 (14352): app exit status: 0x1
10:09:21 (14352): called boinc_finish(195)

</stderr_txt>
]]>

Yes we had to do some testing with SELE3-5. SELE6 ought to work fine though. 1741/88 success/fail ratio

Things seem rather stable for SELE6. For further discussion let's please go to the multicore forum.

I'm trying out the AMD EPYC trial from Packet, it runs 48 QM WUs at a time... all valid.

To everybody using hyper-threaded CPUs for crunching:
You should test how well the given app scales with HT on or off on your system. The other approach is leave HT on, but lower the percentage of the usable CPUs in BOINC manager (down to 50%). Too many simultaneous memory intensive apps would cause too many cache misses, resulting in degraded combined performance. With HT off (or by setting the usable CPUs to 50%) calculation time should be halved (due that two threads have one FPU). If it's more than a half, then the number of usable CPUs could be increased, while the RAC has risen accordingly (= in a direct ratio).
I can't test it myself until the Windows app has been released, but I'm interested.
A simultaneous GPU task also could degrade the performance of the CPU tasks and vice versa.

Zoltan, I think you have a great point here. I am noticing much higher CPU utilization and half of the RAM usage when I switched to 50% CPU in BOINC on these new QC WUs. I think it's mostly to due to the much lower Hard drive bandwidth required and perhaps also the cache on the CPU is more efficiently allocated.

I'm trying out the AMD EPYC trial from Packet, it runs 48 QM WUs at a time... all valid.

To everybody using hyper-threaded CPUs for crunching:
You should test how well the given app scales with HT on or off on your system. The other approach is leave HT on, but lower the percentage of the usable CPUs in BOINC manager (down to 50%). Too many simultaneous memory intensive apps would cause too many cache misses, resulting in degraded combined performance. With HT off (or by setting the usable CPUs to 50%) calculation time should be halved (due that two threads have one FPU). If it's more than a half, then the number of usable CPUs could be increased, while the RAC has risen accordingly (= in a direct ratio).
I can't test it myself until the Windows app has been released, but I'm interested.
A simultaneous GPU task also could degrade the performance of the CPU tasks and vice versa.

Zoltan, I think you have a great point here. I am noticing much higher CPU utilization and half of the RAM usage when I switched to 50% CPU in BOINC on these new QC WUs. I think it's mostly to due to the much lower Hard drive bandwidth required and perhaps also the cache on the CPU is more efficiently allocated.

Yup, I added Rosetta and WCG to the mix and the few GPUGRID WU run constantly @ 400%

---

Do you have any tips for getting higher utilization out of these new large molecule QC WUs? I am already running 4 WUs on a 16 core system which is 50% usage in BOINC but the utilization is all over the place. It's using up to 23gb of ram (I have 32gb) with only 4 WUs and I have plenty of space on the SSD.

CPU tasks - unsent: 44,723; in progress: 848; users in last 24hrs: 76

Quantum Chemistry unsent: 13,191 in progress: 866

Looks like we are cutting that number down to size quickly...

---

QC WUs are almost out, less than 600 to send out.

They may be waiting until the 3.31 jobs finish before introducing the new 3.32 version. I expect they have plenty more.

Hopefully. We are officially out of cpu work.

---

I am running two resends. One of them failed with "file too big" error. The other is running.
Tullio

I submitted some WUs but I am warning you :P This batch will use lots of scratch space.