Re: [Dorset] New BOINC project that does computing for COVID-19
On Fri, 15 May 2020 at 10:29, Hamish McIntyre-Bhatty wrote: > I'm going to have one last go at getting GPU computing to work for > Folding soon, but if I can't make it work, does anyone think it's worth > doing CPU folding, or are the contributions so small as to be pointless? The CPU folding is important. Some of the simulations are only made available for CPU work. > Tim, I'd be interested to see how your KVM computing is going - you said > it was very slow over the call, but hoping it will speed up. I probably > won't do it, but I'm interested to see how well it works. Well, it's hard to say because of the estimated time remaining being so broken. But some tasks have completed so we can compare. Here is the 'bare metal' version, actually running inside a podman container: https://boinc.ibercivis.es/ibercivis/show_host_detail.php?hostid=2288 The measured floating point and integer speeds are 6.53b and 136.5b ops/s, and the tasks generally take about 30,000s for v0.01 and 3,500s for v0.02. Here's a VM I was using: https://boinc.ibercivis.es/ibercivis/show_host_detail.php?hostid=3505 and the measured floating point and integer speeds don't seem to have been filled in yet; they are still the default you get before you've even submitted a completed task. The tasks generally take about the same amount of time! But the credit received for them is much lower -- I have no idea why that is. I reconnected that VM using the correct project URL (oops) so now it shows up here: https://boinc.ibercivis.es/ibercivis/show_host_detail.php?hostid=4223 I'll leave it going for a bit to compare some more. Tim. */ -- Next meeting: Online, Jitsi, Tuesday, 2020-06-02 20:00 Check to whom you are replying Meetings, mailing list, IRC, ... http://dorset.lug.org.uk New thread, don't hijack: mailto:dorset@mailman.lug.org.uk
Re: [Dorset] New BOINC project that does computing for COVID-19
On Friday, 15 May 2020 10:28:36 BST Hamish McIntyre-Bhatty wrote: > Aw that's a shame. I was just going to re-add this to my Ryzen system to > contribute some more work. Hamish, Checkpointing is working; it's just ETA computation that isn't. All my Tasks from yesterday survived shutdown and completed withing an hour of startup this morning. The ETA computation issue is an annoyance, nothing more. There has been some discussion on the Ibercivis Message Board about Checkpointing. There have been some complaints that users can't see checkpointing happen: https://boinc.ibercivis.es/ibercivis/forum_thread.php?id=49&postid=300 The response has been that they are using a wrapper to allow their Tasks to run under BOINC and this wrapper doesn't support checkpointing so they've instigated at workround. Perhaps the wrapper inhibits the ETA computation too. -- Terry Coles -- Next meeting: Online, Jitsi, Tuesday, 2020-06-02 20:00 Check to whom you are replying Meetings, mailing list, IRC, ... http://dorset.lug.org.uk New thread, don't hijack: mailto:dorset@mailman.lug.org.uk
Re: [Dorset] New BOINC project that does computing for COVID-19
On 15/05/2020 10:22, Terry Coles wrote: > On Friday, 15 May 2020 08:47:08 BST Terry Coles wrote: >> The ETA computation seems to be fixed as well. The client is now reporting >> fractions of a percentage complete and shows no sign of freezing at any >> point. One of yesterday's Tasks is now 'Ready to Report' and of the rest, >> six out of seven have 35 minutes to go. > It only 'seemed' to be fixed. Those 35 min Tasks ran for another hour or so > and then suddenly reported. > > The eight Tasks currently running are all at exactly 20.000% or 60.000% > exactly and their ETA has increased to 13 hours and one hour respectively ;-( > Aw that's a shame. I was just going to re-add this to my Ryzen system to contribute some more work. At the moment I only have it on my old laptop with Rosetta, because that's quiet and can stay on 24/7, but it's also very slow. I might ask if they're going to add support for ARM, because if so I could computer for them on the Pi 3 too. Rosetta now supports ARM, but my Pi 3 doesn't really have enough RAM for most tasks sadly. I'm going to have one last go at getting GPU computing to work for Folding soon, but if I can't make it work, does anyone think it's worth doing CPU folding, or are the contributions so small as to be pointless? Tim, I'd be interested to see how your KVM computing is going - you said it was very slow over the call, but hoping it will speed up. I probably won't do it, but I'm interested to see how well it works. Hamish signature.asc Description: OpenPGP digital signature -- Next meeting: Online, Jitsi, Tuesday, 2020-06-02 20:00 Check to whom you are replying Meetings, mailing list, IRC, ... http://dorset.lug.org.uk New thread, don't hijack: mailto:dorset@mailman.lug.org.uk
Re: [Dorset] New BOINC project that does computing for COVID-19
On Friday, 15 May 2020 08:47:08 BST Terry Coles wrote: > The ETA computation seems to be fixed as well. The client is now reporting > fractions of a percentage complete and shows no sign of freezing at any > point. One of yesterday's Tasks is now 'Ready to Report' and of the rest, > six out of seven have 35 minutes to go. It only 'seemed' to be fixed. Those 35 min Tasks ran for another hour or so and then suddenly reported. The eight Tasks currently running are all at exactly 20.000% or 60.000% exactly and their ETA has increased to 13 hours and one hour respectively ;-( -- Terry Coles -- Next meeting: Online, Jitsi, Tuesday, 2020-06-02 20:00 Check to whom you are replying Meetings, mailing list, IRC, ... http://dorset.lug.org.uk New thread, don't hijack: mailto:dorset@mailman.lug.org.uk
Re: [Dorset] New BOINC project that does computing for COVID-19
On Thursday, 14 May 2020 10:15:31 BST Terry Coles wrote: > 'We are doing checkpointing after each of these lines (in version 0.02 and > in workunits like 13052020 or higher)' I can report that this is working. I shut down last night instead of Suspending and the Ibercivis Tasks all resumed this morning. > BTW all my earlier Tasks have shot up to exactly 60% of Progress with around > 8 hours to complete. The ETA computation seems to be fixed as well. The client is now reporting fractions of a percentage complete and shows no sign of freezing at any point. One of yesterday's Tasks is now 'Ready to Report' and of the rest, six out of seven have 35 minutes to go. -- Terry Coles -- Next meeting: Online, Jitsi, Tuesday, 2020-06-02 20:00 Check to whom you are replying Meetings, mailing list, IRC, ... http://dorset.lug.org.uk New thread, don't hijack: mailto:dorset@mailman.lug.org.uk
