|
Here's some updates from the WCG forums on their pandemic research project:
Many people are asking about the Autodock GPU project, and why this project will use Autodock 4 instead of Autodock Vina, which is known to have a far higher performance.
|
#
?
Apr 5, 2020 00:46
|
|
|
|
| # ? Apr 23, 2024 16:18 |
|
|
Progressive JPEG posted:These are all 4GB models, but the memory usage could still be an issue since BOINC is not getting exclusive use of that full 4GB. Instead the RPis are all in a k8s cluster that's running other stuff as well. Worst case if I see any OOM errors I can cordon off a couple otherwise-idle nodes to just run Rosetta stuff. Turns out they're doing fine in terms of memory usage. Wouldn't work with a 1GB model but 4GB and possibly 2GB would be fine: code:
|
|
#
?
Apr 5, 2020 01:35
|
|
|
From what I've been seeing, on Raspberry Pi they really want you to set your BOINC Computing preferences to use up 75% of your RAM. EDIT: Also there's the issue of disk space on the Raspberry Pi because Rosetta@Home takes up a fuckton of it. Tuxide fucked around with this message at 09:12 on Apr 5, 2020 |
|
#
?
Apr 5, 2020 01:45
|
|
|
Couldn't figure out why my computers weren't getting new Seti workunits... then saw the news. Been running Seti@home on all my various PCs for 17 years, pretty sad to see it go to pasture.  Guess i gotta start making the little number the big number again. Switched to Rosetta and joined da goon team. Heres to the next 17 years... e: perhaps I can push us these guys on the team rankings! 
Cygni fucked around with this message at 08:05 on Apr 7, 2020 |
|
#
?
Apr 7, 2020 07:59
|
|
|
Cygni posted:Guess i gotta start making the little number the big number again. Switched to Rosetta and joined da goon team. Heres to the next 17 years... Welcome! I'll have you know that our cause is a noble one because we're fighting COVID-19, and also maybe I'm just being selfish but I'd rather not have some "wonder drug" with a name like derpyhooveshentaioxylene being shoved into the hips of every American. Rankings have been pretty dramatic this past week, Folding and Rosetta have been brought up a few times on the COVID-19 thread on GBS, so we're getting new goons from there too. Here's another rundown of team rankings. In Asteroids@Home, we've crossed into the Top 200 for Total credit and are now #196. Our Recent average credit score has been going down as expected due to our recent focus on COVID-19 stuff. In Folding@Home, I guess the goon team rose to #49 in rank? I'm pretty sure we were #50 last I checked. In Rosetta@Home, we continue to rise. We have risen to #71 in Recent average credit (right below Team Brony@Home) and #245 in Total credit. In TN-Grid (which doesn't have a whole lot of contributors), Team Something Awful has only existed for a week, and we are already #23 in Recent average credit and #97 in Total credit. Adding another 5,000 to the recent average credit score will get us onto the Top 20 board, but this gap has been widening. Tuxide fucked around with this message at 09:10 on Apr 7, 2020 |
|
#
?
Apr 7, 2020 09:07
|
|
|
Got back into WCG, and adding F@H so I can use my GPU too. Over 6k work units on WCG so far. Also learned that if you don't log in for long enough, you get purged from the WCG system, so I had to re-sign up. In the goon groups for both. Edit: I wish F@H had more granular control over CPU/GPU usage the way BOINC does. taiyoko fucked around with this message at 21:29 on Apr 8, 2020 |
|
#
?
Apr 8, 2020 03:14
|
|
|
TN-Grid is still offering COVID-19-related gene stuff in case anyone is skeptical as to whether they're actually contributing anything meaningful. From what it sounds like, the latest news is they added about 730k workunits covering 1242 gene isoforms towards the start of April, all related to COVID-19 in some way. Sounds like the ACE2 work was completed a couple weeks ago and handed off to the scientific community. Since then, they've expanded to include genes related to ACE2 (regardless of whether they're relevant), DPP4 and related genes, and multiple cell lines including HEK-293T and A549. If you want to know if any of your TN-Grid work is actually COVID-19-related, look for the following tokens in your workunit names:
All three of my Raspberry Pis currently have 60 COVID-19-related WUs assigned, so we'll see how much work actually gets done. I'm guessing about half of the WUs have already been assigned though, but there's still work to do here.
|
|
#
?
Apr 12, 2020 10:34
|
|
|
I've pushed out v2.5.0 of Homefarm. The new feature for this release is that the `status` and `query` subcommands can now generate JSON in addition to their existing pretty-printed report format. Usage and format is described here: https://github.com/firepear/homefarm/blob/master/docs/management.md#json-output `query` also now reports the actual names of projects, rather than using the string you used to match a project name as a proxy. This ensures consistent output, no matter what string you use as the project search parameter of your query -- and consistency is always good for tooling purposes. This isn't how I intended to spend today, but there we go.
|
|
#
?
Apr 12, 2020 22:58
|
|
|
Started up folding@home a few days ago after a bit of a break. I'm ranked 636 in the SA team. However it looks like as of this morning there are no cpu or gpu jobs. I guess I just did too much work. Is Rosetta's interface nice and simple ? I remember I did some gridcoin mining from boinc a couple years back and the interface was kinda confusing. Also I'm surprised that there is no support for integrated intel GPU's. Currently I have an i3, a ryzen 2700 and a gtx1080ti working on stuff set to medium. Edit: Set up boinc and joined the SA rosetta team. I'm running boinc set to 50% usage and folding@home set to light. Hopefully the combined two don't peg the machine at 100%. I'll do the name on my NUC units after I get back from the tire shop where I just had an unexpected $6000 bill
jonathan fucked around with this message at 15:45 on Apr 13, 2020 |
|
#
?
Apr 13, 2020 15:28
|
|
|
jonathan posted:Started up folding@home a few days ago after a bit of a break. I'm ranked 636 in the SA team. You can basically just add Rosetta and let it rip. The only thing you really need to do is set the activity settings and maybe how many days of work you want loaded up at once.
|
|
#
?
Apr 13, 2020 15:47
|
|
|
Stanley Tucheetos posted:maybe how many days of work you want loaded up at once. And leave that low. I still have mine at the default of 0.2 + 0.3 days. Don't hoard WUs for points; let them be spread around and done as fast as possible.
|
|
#
?
Apr 13, 2020 22:00
|
|
|
Some more number movements this week. In Asteroids@Home, our Total credit ranking has gone up to #195, and it'll probably continue to go up for the foreseeable future as we focus on COVID-19 stuff. In SETI@Home, now that it's in hibernation, we stopped on #201 in Total credit. That rank doesn't look like it's gonna change unless they start pumping out WUs again. In Folding@home, we remain on #49. The project as a whole is now 2.4 exaFLOPS. It was 1.5 exaFLOPS only two and a half weeks ago, so if the trend continues then it will probably be double that amount by the end of April. In Rosetta@Home, goons have risen to #62 in Recent average credit (ahead of team Brony@Home finally) and #229 in Total credit. In TN-Grid, our Recent average credit ranking has been going up and down from #22 to #24. We're #22 as time of writing, but it's attracted more BOINC users to gene research because of COVID-19. If we want the goons on the Top 20 list, we still need to close the gap by another 5000, and this is probably only going to get wider. Our Total credit ranking is sitting on #88.
|
|
#
?
Apr 14, 2020 01:25
|
|
|
Tuxide posted:Some more number movements this week. When I first checked a few days back, Team SA was reported at #46
|
|
#
?
Apr 14, 2020 02:53
|
|
|
jonathan posted:When I first checked a few days back, Team SA was reported at #46 There's a per-month ranking as well, that's probably where you're getting that from instead of the overall stats. Goons are currently #49 overall in Folding and #47 for April 2020.
|
|
#
?
Apr 14, 2020 03:03
|
|
|
I'm having trouble finding the # of work days settings in boinc (or whatever was mentioned in the responses to my questions a few days up). I was using the actual boinc manager settings within the app in my sys tray. Not the web interface. I set max storage to 5gb as I don't want it filling up my system SSD and im too lazy to redirect it to a platter drive.
|
|
#
?
Apr 14, 2020 03:38
|
|
|
Also who the gently caress is thomasteo, I'll never get to the #1 spot with that person running his hardware. Does he work at a lab somewhere ?
|
|
#
?
Apr 14, 2020 03:39
|
|
|
jonathan posted:I'm having trouble finding the # of work days settings in boinc (or whatever was mentioned in the responses to my questions a few days up). I was using the actual boinc manager settings within the app in my sys tray. Not the web interface. This is the setting we're talking about. What I have shown is fine for something like Asteroids, but I wouldn't recommend it for COVID-19 stuff. 
|
|
#
?
Apr 14, 2020 04:40
|
|
|
Tuxide posted:This is the setting we're talking about. What I have shown is fine for something like Asteroids, but I wouldn't recommend it for COVID-19 stuff. Thanks, I did .2 and .3 days for those. somehow those weren't showing before.
|
|
#
?
Apr 14, 2020 05:05
|
|
|
As an update on running Rosetta@Home on eight 4GB Raspberry Pi 4's, I found that occasionally the RAM usage would get too high and cause instability. A symptom of the prolonged low memory was the 'kblockd' kernel component freaking out, causing a bunch of IO lag, and eventually causing userspace processes to get stuck waiting on IO. This would usually happen only once all four threads were running Rosetta@Home, as opposed to some mixture of Rosetta@Home and TN-Grid, which BOINC is also configured to run due to its available arm64 support. The first thing I tried was configuring the ram_max_used_busy_pct and ram_max_used_idle_pct settings, which I set to 70%. This didn't seem to do anything where Rosetta was concerned, so I still saw instability a few hours after setting those on the Raspberry Pis that ended up running four Rosetta threads. The next thing I thought of trying was to just give Rosetta a lower priority, but this wouldn't really work if the luck of the draw led to a Raspberry Pi getting lots of Rosetta work and no TN-Grid work. Finally, I tried to just set max_ncpus_pct to 75%, so that only three of the four available cores/threads would run BOINC workloads. This would leave more headroom in terms of memory use, even if all of those were running Rosetta, but at the cost of reducing overall throughput. This seems to have done the trick - the instability has gone away. It's also leading to the Pis being a bit more responsive in terms of other things the that they are also running. With max_ncpus_pct=0.75 on each node, and the cluster otherwise fairly idle: code:- Client preferences docs - Sample XML config
|
|
#
?
Apr 14, 2020 06:11
|
|
|
jonathan posted:Also who the gently caress is thomasteo, I'll never get to the #1 spot with that person running his hardware. Does he work at a lab somewhere ? No idea, and I also don't know exactly what numbers you're looking at, but I'm sure you'll be shocked to learn that some people actually go out of their way to cheat at BOINC. I guess because unlike mobile games, you can't whale your way to the top? (Well, you could but would take money and effort. And ongoing power/cooling bills.) But also there are some crazy fuckers out there who have a lot more hardware than I do (and I'm borderline uncomfortable with the amount of hardware I have). There's also some people who have (with permission or otherwise) co-opted entire fleets of machines where they work. And then there's some "users" who clearly represent organizations rather than an actual individual, like Marist College on the WCG stats. And btw, that's the rankings for yesterday, not all-time. So maybe it's real, and maybe it isn't.
|
|
#
?
Apr 14, 2020 06:19
|
|
|
Ah I was just being pretend over aggressive because you can click users and then see the details of the hardware they're using. He (or She) is running 6 machines that are all set up for lab style number crunching or rendering or something. Nvidia Quadro gpus, Intel Xeon 16 core processors etc. I was hoping he'd pop up and say "oh yeah I work in a lab doing butt plug CNC design" or something.
|
|
#
?
Apr 14, 2020 06:47
|
|
|
Just put together that 3990x machine you've been wanting and that should do it. Probably.
|
|
#
?
Apr 14, 2020 06:54
|
|
|
I'm considering switching off Folding and over to Rosetta and/or TN because I'm not fond of how few config options there are for Folding versus BOINC, and I have to manually exit the program at night because my computer is in my bedroom, and while BOINC respects sleep mode, Folding does not and will only lock my screen instead of letting the system sleep.
|
|
#
?
Apr 14, 2020 10:25
|
|
|
taiyoko posted:I'm considering switching off Folding and over to Rosetta and/or TN because I'm not fond of how few config options there are for Folding versus BOINC, and I have to manually exit the program at night because my computer is in my bedroom, and while BOINC respects sleep mode, Folding does not and will only lock my screen instead of letting the system sleep. I've started moving some of my stuff over to BOINC because folding doesn't have any work. edit: the setup was pretty easy, just run the installer, see no projects available, wonder what the hell, close the project adder, open it again, pick the project from the list, make a login, join goons, and now it's using some cores.
|
|
#
?
Apr 14, 2020 11:58
|
|
|
Watching my numbers go up is pretty addicting. I just added in my unraid server to help me get a boost. I know I wont get the top spot with thomasteo's hardware. I made it to 4th in recent credit on one machine though! Being able to see other peoples hardware is kind of cool too.
|
|
#
?
Apr 14, 2020 14:51
|
|
|
does GPUgrid have no work ? I haven't done any work in the 24 hrs since signing up with it.
|
|
#
?
Apr 14, 2020 19:18
|
|
|
The impression I have gotten is that GPUGrid is a very small project. It's not uncommon for them to have pauses between batches of WUs. That's actually why I have my GPU projects set up like this:code:All that said, right now my GPU nodes are crunching WUs from GPUGrid, and have been very consistently for the past week. code:
|
|
#
?
Apr 14, 2020 19:56
|
|
|
Thanks, it might've been a harddrive space thing blocking it from grabbing tasks. I can't figure out how to change where it stores the tasks and completed work.
|
|
#
?
Apr 14, 2020 20:22
|
|
|
I've been running Folding for a few days now and quite impressed how nicely its playing with my computer as I've yet to notice it getting in the way of anything I've been trying to do.
|
|
#
?
Apr 15, 2020 09:27
|
|
|
jonathan posted:Also who the gently caress is thomasteo, I'll never get to the #1 spot with that person running his hardware. Does he work at a lab somewhere ? Hey. mdxi posted:There's also some people who have (with permission or otherwise) co-opted entire fleets of machines where they work. It's mostly just a small handful of machines we've just acquired to shake out any early life failures. And hopefully do something worthwhile too.
|
|
#
?
Apr 16, 2020 04:34
|
|
|
I just spun up two spare Raspberry Pis and put them to work on TN-Grid  Last time I looked at that project they were doing some sort of botany project that didn't seem too interesting to me, so I kinda lost track of them until all of this stuff started up. Seems like a lot of others did the same, because there are only a few really active teams (besides Gridcoin). We have 9 goons, not all of whom are active, and just broke the top 20. If nothing else it's worth adding for when your Rosetta WUs dry up.
|
|
#
?
Apr 16, 2020 09:46
|
|
|
Discussion Quorum posted:We have 9 goons, not all of whom are active, and just broke the top 20. If nothing else it's worth adding for when your Rosetta WUs dry up. Those "inactives" on TN-Grid are new goons, we're just waiting for the credits to start printing. Now that goons are #20 on TN-Grid, guess we need a new goal? Let's go for #19! There's a 5500 gap to close between the goons and Team arm.powered, where every single one of them is running some degree of ARM Linux. One of these nerds has BOINC running on a couple routers running OpenWrt lol
|
|
#
?
Apr 16, 2020 10:12
|
|
|
My monthly summary of WCG updates (which has the wrong month in the title!) is up. It includes the OpenPandemics infodump that I posted here back on the 4th. The only new info on that, in case it's all you're interested in, is that: 1) the team chose AutoDock over Vina because OG Autodock has a new "reactive docking protocol" that they want to use. 2) tangentially, half of FAH2's researchers are at Scripps (AKA the org running OpenPandemics) and they're dropping that work for now in order to get OP off the ground
|
|
#
?
Apr 17, 2020 05:17
|
|
|
Crossposted from the AMD thread, because I genuinely couldn't decide where this belonged: AMD is donating USD15M worth of computing gear to the COVID-19 effort. It'll be made up of Epyc-based HPC nodes to be distributed to organizations upon request, and enough Radeon Instinct M150 cards to double the peak performance of LLNL's Corona supercomputer. https://www.anandtech.com/show/15731/amd-covid19-hpc-fund-initial-15m-donation-of-epyc-and-radeon-hardware
|
|
#
?
Apr 17, 2020 15:51
|
|
|
Does anyone have any recommendations or posts elsewhere about energy/price efficient ways to setup a little bionic farm? I really like the idea of having a few machines linked together running bionic, ideally some sort of cheapish single board computer that won't be a massive energy drain. Obviously there seem to be quite a lot of people doing this with raspberry pi's but I was wondering if there are other options out in the flood of single board computers, maybe some with gpu's optimised for machine learning or whatever?
|
|
#
?
Apr 19, 2020 17:45
|
|
|
How much are you willing to spend on hardware, how much more of an electric bill are you willing to pay, and what aspect of this is most interesting to you? Are you more interested in optimizing for points/watt or hardware cost/point? Which projects do you want to run? If you just want to tinker with setting up a cluster of SBCs then get whichever one's most interesting to you and figure out how to make BOINC work on it. In the end you're making a donation so I recommend going for whatever setup is going to yield the most enjoyment and fit in your budget. I'm running jobs because I want to contribute to science and don't want to spend a penny on hardware primarily for this, so I just run it on the machines I already own for other purposes. SamDabbers fucked around with this message at 19:06 on Apr 19, 2020 |
|
#
?
Apr 19, 2020 18:54
|
|
|
I've been wondering the same thing. I doubt a farm of SBCs would be as good on a performance per dollar basis as a cheap old x64 box with plenty of RAM. Maybe not on a performance per watt metric either. Even more so for GPU-enabled apps. For reference, I have an old Android phone running a variety of projects. An Asteroids WU takes around 15-16 hours, vs 15 minutes on my GTX 1070. I also have a pair of Pi 3B+ running TN-Grid. They turn around 4 WUs every 15-18 hours, vs my Ryzen 2600X which can do 12 every 3 hours if I run it flat out. I am only running these because they were laying around and every cycle helps right now. My curiousity is more towards used corporate SFF desktops. You can get them with 8GB of RAM and an i5-4xxx for less than $150. I haven't looked but I bet a first-gen Ryzen and 350-based board would be pretty cheap as well, except for the cost of DDR4 RAM. Speaking of Pis, they finally pushed me over the edge to get my top 10% badge for TN-Grid (after sitting as the highest-ranked 25%'er for a while)  It ain't gonna get better than that so I feel free start running some Rosetta and WCG again.
Discussion Quorum fucked around with this message at 20:07 on Apr 19, 2020 |
|
#
?
Apr 19, 2020 19:06
|
|
|
For awhile for fun I was running an old pc in my work truck off an inverter down at the work yard. I was having issues when it was -40 so instead of running a heater inside I ran a F@H machine and surprisingly it kept the interior above freezing. Specifically the issues were cans of V8 juice exploding inside the fridge. Also froze a bottle of vodka solid and instead of blowing up it just pushed vodka out the sealed cap.
|
|
#
?
Apr 19, 2020 19:39
|
|
|
vodkat posted:Does anyone have any recommendations or posts elsewhere about energy/price efficient ways to setup a little bionic farm? I really like the idea of having a few machines linked together running bionic, ideally some sort of cheapish single board computer that won't be a massive energy drain. Obviously there seem to be quite a lot of people doing this with raspberry pi's but I was wondering if there are other options out in the flood of single board computers, maybe some with gpu's optimised for machine learning or whatever? Did you mean "BOINC", or is there a project out there called "bionic"? If you did mean BOINC, and if you're talking about x86 SBCs, then my Homefarm project can handle the setup and management portion of things for you. Edit: the installer might need to be made aware of new mass storage device types -- it currently only knows how to handle NVMe and SATA devices -- but I'd be happy to work on that. Irritatingly, the only thing that keeps Homefarm from working for ARM machines as well is that there are inconsistencies in package names between the x86 and ARM repos for Arch linux. Though I guess that could be worked around with (what I think would be) a single setup step on ARM. If anyone is interested, I'd be happy to work together on that. mdxi fucked around with this message at 21:40 on Apr 19, 2020 |
|
#
?
Apr 19, 2020 21:35
|
|
|
|
| # ? Apr 23, 2024 16:18 |
|
|
Number movements this past week: In Asteroids@Home, our Total credit ranking remains on #195 as goons focus on COVID-19 stuff. In Folding@home, we are still ranked #49. Folding peaked above 2.60 exaFLOPS sometime last week, then it went down to 2.58 exaFLOPS; dunno what's up with that. In Rosetta@Home, goons have been going up and down from #62 to #64 in Recent average credit (currently #63), even though the numbers themselves have been going up. We are #217 in Total credit. In TN-Grid, our Recent average credit ranking has gone up to #17, beating Team arm.powered. I guess now that we're on the Top 20, we gotta shoot for something even higher. #16 is Team SETI.USA, and there is just over a 4000 gap between goons and them. In Total credit, we have risen to #80.
|
|
#
?
Apr 21, 2020 00:23
|
|