We are losing CPU Volunteers, can anyone help?

If admins want CPU support they'll make an app w/o bugs. No effort by them, no effort from us. They get paid, we pay for the PCs and we'd rather not have it aimlessly wasted.

We are losing CPU Volunteers, can anyone help?

If admins want CPU support they'll make an app w/o bugs. No effort by them, no effort from us.

+1

I would put up with the bugs if I could. But I get only crashes on QC now, so I am out of business.

I hope they make a big announcement when it is fixed, since I won't know about it otherwise. Maybe they are spending their time on the Windows version?

Maybe they are more interested to the results of the CRUNCHATHLON competition. You can not compete with a dead horse.
Tullio

You can not compete with a dead horse.
Tullio

Not at the Palio either.

Please see the Number Crunching forum.

Well, I gave QM a try, but I really didn't like it, for a few reasons.

First, the setup doesn't work when running more than one WU at a time. You need a "lock file" to ensure only one thread does the setup. If that thread gets aborted you might have problems, but it's better than the current situation.

Second, you know what I hate? I hate BOINC apps that download more data themselves. (Such as LHC@Home Atlas.) All necessary data should be coming from your BOINC server. At least your app only seems to do that downloading once.

Third, you know what else I hate? Bloated apps. Your "miniconda" includes such things as TK (not needed unless there's a screen saver?) and man pages. Maybe if you streamlined it you could fit whatever extra data it downloads in the initial download, and then you wouldn't need the networking libraries either.

But, I finally did get QM working. What made me give up on it was the credit. It started around 600 credits/WU, but it seemed to get cut in half every few sets of WUs. It was down around 50 when I gave up.

But, I finally did get QM working. What made me give up on it was the credit. It started around 600 credits/WU, but it seemed to get cut in half every few sets of WUs. It was down around 50 when I gave up.

I don't care about credits themselves, but I do wonder about the science that is being accomplished. If it is the same per work unit, that is OK, but if it is being cut in half periodically also, then that is a problem. I wonder what causes it?

But, I finally did get QM working. What made me give up on it was the credit. It started around 600 credits/WU, but it seemed to get cut in half every few sets of WUs. It was down around 50 when I gave up.

The credits are down to around 50 right now because the Work Units are extremely short due to testing.

I'm not sure what everyone's fantasy is with credit. I personally couldn't care less as long as what I am doing is benefiting science. The credit itself is worth nothing so I'm not sure I get the point. I guess some people just need something more.

Indeed as papalito said these WUs are muuuuuch shorter than the 600 credits ones. They scale by computation time automatically. I miscalculated a bit yesterday how fast they are so we ran out but I'm going to submit more in an hour or so.

The downloading stuff I understand but we depend unfortunately on external software (psi4) so we cannot control everything. As you said it only downloads once. But yes conda is bloated in general but we kept it down to the bare minimum packages.

@Jim the WUs that I send contain varying molecules of different number of atoms. The ones I sent yesterday had very few atoms so they completed super fast. But "every molecule is sacred" as per Monty Python. I guess I could mix all molecules together but it would become an organizational chaos for me to keep track of what I have already calculated, so for now I will continue submitting them with increasing molecule size.

Stefan, Can you reduce free disk-space requirement of 4768.37 MB to something like 4000MB or less? Your QM WUs do not fit on my USB 16 GB Stick anymore! (After Lubuntu Up-Grade from 17.10 to 18.04)

Sorry klepel, I don't think I can :( Most of it is taken up by miniconda and the required software, not the workunits themselves.

@Jim the WUs that I send contain varying molecules of different number of atoms. The ones I sent yesterday had very few atoms so they completed super fast. But "every molecule is sacred" as per Monty Python. I guess I could mix all molecules together but it would become an organizational chaos for me to keep track of what I have already calculated, so for now I will continue submitting them with increasing molecule size.

No problem at all. Keep doing what you have to do.

@Stefan can you quickly describe for us what the Quantum Chemistry Work Units are doing and what you are learning from these work units?

When can we expect some info about scientific results created with this app?

Months away, we have just started.

Some months have gone.
We are crunching for....? Cancer research? Any preliminary result?
Do you plan, togheter with windows app, also a gpu version?

@Stefan can you quickly describe for us what the Quantum Chemistry Work Units are doing and what you are learning from these work units?

+1

+ another 1. I have 7+7+3 cores 100% dedicated to QC and 4+4+4 cores 50% dedicated to QC and WCG, all 24/7. Along with 3 GTX 1060's, makes for a very warm office, especially now in summer. Would be interesting to find out what the project goals are :).

@Stefan can you quickly describe for us what the Quantum Chemistry Work Units are doing and what you are learning from these work units?

See the "New Student and QMML Project" thread for a clue.

@Stefan can you quickly describe for us what the Quantum Chemistry Work Units are doing and what you are learning from these work units?

See the "New Student and QMML Project" thread for a clue.

Really a "clue".
"We are simulating molecules". Ok, very precise :-P

Yeah sorry. I've developed a partially healthy paranoia over the last years due to some researchers being unhealthily competitive. In this case I'm trying to steer away from what others do to avoid problems but I'll be a bit vague anyway just to be safe.

Practically we are trying to teach a neural network to calculate molecular energies and forces. QM calculations are horribly slow and scale quadratically to the number of atoms. But a network trained on QM data is orders of magnitude faster, scales linearly with the numbers of atoms and achieves decently good accuracy. We believe these networks are the future for molecular simulations so we try to work with them and see what problems we can apply them to. At the moment they are still slower than usual MD simulations which we used to do but they should be much more accurate given enough training data. This training data is what is critical and what we are trying to produce currently.

Currently there are already three or so groups working on such networks and they have shown great results so we try to mostly collaborate with them to avoid duplication of effort and clash of research topics.
If you want to read up on some great projects that inspired us, check out ANI1 https://arxiv.org/abs/1610.08935 TensorMol https://arxiv.org/abs/1711.06385 DeepMD https://arxiv.org/abs/1712.03641

On applications to more biological research and implications to disease related research you will have to wait for my publication :)