Project News - GPUGrid

[zsalter]

GPU Grid is insane, I signed up for it yesterday and it requested for some work before I could update my preferences. It grabbed one WU, and that happened to be a "Long run" unit. Which is to be performed on my poor GTX 750 TI STRIX hahahaha. Check it out.
Looks like my GPU has its work cut out for him.
How long do GPU Grid WU's take for everyone else?

[attachment deleted by admin]

[kashi]

Yes they're now very long GPU tasks indeed at GPUGrid.

Wasn't too pleased myself when I went back to GPUGrid a few weeks ago and the first task I received took over 20 hours. Not only is that longer than I like for GPU tasks because I prefer to run them without interruption but it's also getting close to the 24 hour deadline for 50% bonus credit, especially considering the slow upload after completion.

It's settled down since then and the longest tasks I'm now receiving complete in about 16 hours on my GTX 970 @ default speed with SWAN_SYNC. The PABLO_xxxxxx_x_IDP type that you're crunching take me about 11 hours and grant 252,750 credits including 50% <24 hour bonus.

For my current 22 valid tasks, here's the distribution, credit (including 50% bonus) and approx. runtimes:

3 of ADRIA_FOLDA3D01_crystal_ss_contacts_50_a3D - 137,850 credits - 6.5 hours
2 of PABLO_xxxx_adap - 157,200 credits - 7.2 hours
12 of PABLO_xxxxxx_x_IDP - 252,750 credits - 11 hours
1 of PABLO_boundxxxxxxxxx - 277,950 credits - 12.4 hours
4 of PABLO_all_data_goal - 421,800 credits - 15.3 to 16 hours

It's unfortunate that short tasks are apparently very rare now because of those 5 long types listed above, only the 2 shortest running types would finish in under 24 hours on a GTX 750 TI. Your screenshot shows that PABLO_xxxxxx_x_IDP would complete in almost exactly 24 hours but the long upload would take it over and credit would drop from 252,750 (50% bonus) for under 24 hours return to 210,625 (25% bonus) for under 48 hours return. Although possibly that's still better than some other credit stingy GPU projects I suppose.

I select only long tasks in project preferences because of the higher credit rate than short tasks. Short tasks only would definitely be the best choice for a GTX 750 TI though but current scarcity of shorties makes it a very difficult project to support for owners of GTX 750 TI and similar speed cards.

The bonus credit system combined with long tasks is a bit awkward to work with even with my GTX 970. Either have to set my cache really small to avoiding having more than one task in cache at a time for longer than a few hours or set "No new tasks" and download manually after completing each task. In the first case this limits the cache of CPU tasks of other projects, which can be inconvenient on projects where work availability is unreliable or sporadic. In the second case it means if a GPUGrid task finishes while I'm asleep the slight change in fan noise or drop in bedroom temperature doesn't always wake me so my GPU remains idle for a while.

Still, the challenge of working around the foibles and annoyances of projects I wish to support adds to the interest. If it all worked easily and required no knowledge or configuration tweaks to extract better performance then the appeal of crunching would soon wither away.

On the positive side, the tasks run smoothly for me with no errors and card temperature is low. Credit is slightly lower than it used to be last year but still quite good, about 550K-600K per day. Was going to stop about now, because don't want the winter power bill to be too high, but with 2 computers now instead of one and GPU crunching as well haven't needed a bedroom heater at all this winter, even though July had the coldest minimum temperatures here for 15 years. So GPUGrid for another week or two here I think.

[zsalter]

Hey Kashi,

Thanks for the reply!
I never knew about the bonuses for completing the task within a certain timeframe, that's something I will keep in mind. Does this apply to any other projects or is it just GPUGRID?

Woah! thanks for all of that information! Maybe I should consider upgrading my GPU shortly once I save a little more $$$
What GPU would you recommend that would be ample for crunching that is a similar price to the 750TI which I am using currently?
Something that isn't too expensive to run?
The badge for GPUGRID looks awesome so I'm keen to see what that looks like in my signature hahaha.
I agree about the annoyances of having to deal with a few problems in BOINC, add to the feeling of accomplishment.
I'm sure a few PC's crunching at 100% would definitely be sufficient as a bedroom heater. I'm pretty happy with mine at the moment it isn't pulling very much power or generating and unbearable amount of noise or heat which is good.
My GPU has the whole 0dB fan setup which I think is pretty good, but i'm not too worried about a little noise. I decided to max out my GPU fans to make sure it doesn't throttle and keeps the temperature relatively low around the 60-65 degree mark which is good.
I'm watching my rank in the team for GPUGRID and with every task I complete, which is around 1/day, I move up a lot of positions which is fun to watch and see progress.
I've also been watching the team move up a few positions in the team leaderboards which is awesome. I'm looking forward to the next AA where hopefully we can move up a few positions on the leaderboards.

Also, what other projects could I run my GPU on? I have been doing some work for milkyway which is good, and a small amount for primegrid, but that project doesn't really appeal to me very much, it seems kind of pointless compared to other projects such as WCG and POGS and things like that, but thats just my opinion.

Thanks again 

Zach

[Sean]

Also, what other projects could I run my GPU on? I have been doing some work for milkyway which is good, and a small amount for primegrid, but that project doesn't really appeal to me very much, it seems kind of pointless compared to other projects such as WCG and POGS and things like that, but thats just my opinion.

Have a look at this: https://boinc.berkeley.edu/projects.php

SETI@home and Einstein@home (among others) are popular for GPU.

[kashi]

Yes, GPUGrid has 50% bonus credit if you return (process, upload and report) tasks within 24 hours of them being downloaded. And 25% bonus if returned within 48 hours. Your first task just missed the 25% bonus by 8 minutes and 46 seconds!

Recently GPU pricing has been volatile due to high demand from Ethereum mining, so I wouldn't recommend anything just yet. When GPU market settles a bit then an Nvidia card for BOINC generally as more projects are available for them and they're also more efficient than AMD cards. Current pricing means it's probably more than you wish to spend but on the budget end of things a GTX 1060 3GB version looks good. It performs well, has a lowish power draw and excellent performance to power ratio (efficiency). Team member dyeman was very pleased with the price, performance, efficiency of his GTX 1060s. When purchasing GPUs now I'm more concerned with total running costs rather than just initial purchase price. Sounds slightly furtive but in a house shared with others I can hide the cost of hardware purchases if I wish but can't hide the electricity bill, haha.

As you know it's not just the general card model, but also the brand and specific model that is important for BOINC. Need one that runs cool and quiet, especially if your crunching room gets hot in summer. Fortunately for me when I was choosing a GTX 970 I was expertly advised by JugNut to get an MSI "Gaming 4G" model which is one of the best cards I've ever owned in terms of cool, quiet running under heavy load. With the price of power here now unless you have solar then the cost of a recent budget card is quickly recouped by its greater efficiency compared to many older cards if you do any decent amount of GPU crunching. The cheap Gainward GTX 1060s recommended by dyeman are no longer available, so I don't know which other brands/models of 1060 would be the better choice for crunching.

When considering efficiency, in theory it's better to run a more efficient card for less time and control your power usage that way, rather than running a less efficient card all the time. That's what I do now on GPU, run for a month or so, then have a month or so off. You could also only run for limited hours each day I suppose, but I'd find that a bit frustrating myself. When I'm crunching anything I prefer full speed ahead, you know: "if you start me up, if you start me up I'll never stop".

Always had AMD cards in the past, therefore had never contributed to GPUGrid before so bought my current Nvidia card specifically to contribute to GPUGrid as my preference now is to mainly support medical/biology type projects when using the extra power that GPU crunching uses. Plus other team members had been crunching up a storm on GPUGrid in the past and I had been unable to do my bit due to not having an Nvidia card. Plus of course I had no weird and wonderful looking protein badges.

Due to my GPU project type preference also did some Folding@Home on this card and would have done a lot more but the Folding application had a hardware/driver incompatibility with Maxwell series cards like mine which caused intermittent errors. So I stopped Folding@Home and haven't been back since. Will give it another go after the winter electricity bill in September. Like GPUGrid, Folding also has bonus points system for completing work within a set time, it's called Quick Return Bonus and you need to apply for a "passkey" and complete 10 tasks before you get it. Yes it's not BOINC but seems high on the "worthiness" scale if you value protein research that's hopefully contributing to the curing of diseases higher than searching for alien life, cracking codes or trying to prove mathematical hypotheses.

This is just my personal preference and recently has become more heavily influenced by my more limited ability to contribute to GPU crunching due to swift and large increases in the cost of electricity. It's not a judgement of any kind as I've done oodles of GPU crunching myself in the past on projects where any reasonable assessment would conclude that potential benefits are very dubious to put it mildly. Still like to give a little initial support to most new GPU projects whatever their area of research. Also happy to support non biomedical GPU projects when chosen for the Aussie Assaults, which they always are now that POEM@Home is finished. GPUGrid is not eligible as an AA target as there is no AMD GPU application. Can't do anything about that, seems GPUGrid just doesn't have the skills, resources, staff and/or inclination to develop an OpenCL application so AMD cards could be supported. Disappointing really, you'd think they could have come up with something by now.

If you prefer to stay within BOINC for your GPU crunching and SETI, mathematics or astrophysics type projects are not your most favoured type, then there's also XANSONS for COD which goes to the other extreme from GPUGrid and has extremely short tasks. The research relates to crystal structures of organic, inorganic, metal-organic compounds and minerals. It excludes biopolymers however so although it's a chemistry type project it's not directly a biochemistry project.

I also enjoyed "roaring up the charts" within the team on GPUGrid when I first started crunching it. My daily credit was much higher than the large number of members that had contributed in the past on the cards and Playstations that were current then. This meant whoopee, yeeha almost every day when I checked the stats and had risen up heaps of rankings. Still enjoy improving my ranking even though it takes much longer now I'm up to 22nd.

[dyeman]

Looks like there is plenty of "long task" work at the moment for anyone who's given up in frustration :-)

[kashi]

Thanks for the info. Downloaded a task and am crunching away. 

Hadn't been back for a fair while. As you said, got discouraged by the too frequent shortages of work.

[Dingo]

There is a New Test version ACEMD v2.10 (cuda100) that will run on the new Nvidia Turing type cards like GTX 1660Ti and all the RTX cards.  Until now GPUGRID did not support thse new cards.

I have two tasks running on my GTX 1660Ti's and they are almost finished so I think thy will run OK, will see in the morning.   I have run a few on my 1080Ti as well to make sure they work on the older GPUs as well.

I was still awake when the tasks completed and were valdated with credit.

https://www.gpugrid.net/show_host_detail.php?hostid=517492

As a side note the GTX1080Ti out performs the 1660Ti by almost 5000 seconds 1.38 hours per tasks.

https://www.gpugrid.net/show_host_detail.php?hostid=453402

[JugNut]

Yea that's great I was starting to think they might never support Turing cards.  It's definitely one of the best GPU projects out there.   

[Daniel]

Looks like Folding@home isn't the only project getting more support recently, yesterday the admin at GPUGrid posted ...

If I am not missing anything, GPUGRID volunteers doubled the number of active hosts in the last month. It may have been either due to the closing of seti@home, or a consequence of folding at home's covid effort.

Maybe people leaving seti@home, or maybe the Coronavirus outbreak has got people thinking more about biomedical research.
Maybe also because they have enough work at the moment - there were times in the past when I would have liked to do some GPUGrid, but had trouble getting tasks.

[tazzduke]

Well I might go back to GPUGRID, at least you can get work there.

Just wait till the end of March when the new work stops at Seti.

Cheers

[Dingo]

[GPUGrid.net] PYSCFbeta: Quantum chemistry calculations on GPU
Hello GPUGRID! We are deploying a new app "PYSCFbeta: Quantum chemistry calculations on GPU". It is currently in testing/beta stage. It is only on Linux at the moment. The app performs quantum chemistry calculations. At the moment we are using it specifically for Density Functional Theory calculations: http://en.wikipedia.org/wiki/Density_functional_theory These types of calculations allow us to accurately compute specific properties of small molecules. The current test work units have a runtime of the order 1hr (very much dependent on the GPU speed and size of molecule). Each work unit currently contains 1 molecule with ~10 configurations. The app will not work on GPUs with compute capability less than 6.0. It should not be sending them to these cards but I think at the moment this functionality is not working properly. The work-units require a lot of GPU memory. It works best if the work-unit is the only thing running on the GPU. If other programs are using significant GPU memory the work-unit might fail. Looking forward to hearing feedback from you. Steve

View article · Fri, 12 Jan 2024 13:03:21 +0000

[Dingo]

[GPUGrid.net] Discord channel for GPUGRID
Hi, I have created a discord channel for GPUGRID. It kind of duplicates this in some aspects but maybe people are more used to Discord. It is an additional channel that may help to build up the community. JOIN: https://discord.gg/abpWXawZ7v GDF

View article · Tue, 13 Feb 2024 11:29:01 +0000

[Dingo]

[GPUGrid.net] In-silico Binding Assay (ISBA/ACEMD3)
This is a message Adrià posted in our discord channel. Since he doesn't have an account in the GPUGRID forum I open the thread for him. If you want to be more up to date to the news related to this project and others please join our discord, we are usually more active there: https://discord.gg/dCMkcafPpX

Hello GPUGRID! Here Adrià. I'll be recovering the ACEMD3 application again, and sending new jobs of standard MD simulations (We've been testing it these past weeks to make sure it worked well for both Windows and Linux) The main goal of these new batch of simulations will be to validate further our capacity to predict the binding mode of ligands using simulations and adaptive sampling methods. Those of you that have been around for some time here might already be familiar with these simulations, such as the Benzamidin-Trypsin system (https://www.pnas.org/doi/abs/10.1073/pnas.1103547108) or the Dopamine D3 receptor with an antagonist ligand (https://www.nature.com/articles/s41598-018-19345-7#Ack1), which we were able to simulate thanks to GPUGRID and all your effort! Now, we are revisiting this method, which we call in-silico binding assay (ISBA). During drug discovery campaings, it's common that you know of ligands that bind to your target, but you don't know their binding mode, the exact conformation and structure that both the ligand and the protein have when bound. Knowing the binding mode is critical for further development of the molecule into a potent and usable drug. The most precise way of discovering the binding mode is with crystallization. However, that can take too much time or be directly impossible, depending on the protein. Therefore, we want to optimize and refine ISBA for binding mode prediction, so it can be usable during drug discovery projects. To summarize a bit our objectives, we want to predict binding modes for larger molecules than Benzamidin, with the same precision, but with less simulation time that was needed for the D3 receptor system. To do so, we'll be using the latest version of adaptive sampling that we developed, AdaptiveBandit (https://pubs.acs.org/doi/abs/10.1021/acs.jctc.0c00205). The objective of these new simulations I'll be sending will be to benchmark AdaptiveBandit in an ISBA scenario, improve the algorithm if required and fine-tune its hyperparameters. Let me know if there's any issue with the simulations. I'll be sending 100ns trajectories for the most part, divided in two steps.
View article · Thu, 2 May 2024 09:28:34 +0000