GPU computing causing excessive hard drive activity

Questions and Answers : GPU applications : GPU computing causing excessive hard drive activity
Message board moderation

To post messages, you must log in.

AuthorMessage
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2023541 - Posted: 18 Dec 2019, 6:28:22 UTC

Hi guys, I'm having a bit of an issue with Boinc that I'm hoping you can help. I've been running Seti for years, first classic and then Boinc, but in the last couple of months I've noticed that enabling GPU computing causes near-steady hard drive activity. (I can't actually say how long this has been going on, but that's when I noticed.) Fairly simple situation: With GPU enabled, hard drive activity doesn't stop; suspending it from the activity dropdown, and it stops. Is this some kind of known issue, and is there some way to stop it aside from just not using the GPU?

Relevant info: Running Boinc 7.14.2 (x64), exclusively Seti@home. I have had other projects running but they've all been suspended for months. Running on an i5-6600K, all four cores enabled, not overclocked, 16GB RAM, GTX 980Ti GPU, Windows 10 Pro 64-bit.

Any info or ideas are welcome. Thanks!
ID: 2023541 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 2023558 - Posted: 18 Dec 2019, 8:48:21 UTC

Not sure why you consider it to be a problem, however
The more work you do, the more Seti has to read & write to the HDD. If you have your AV programme set to monitor the BOINC folders, then that would result in a lot of HDD activity. Whitelisting the BOINC programme & data folders & their sub folders would stop that from occurring.

However I find it odd that running the video card significantly increases the drive activity, as CPU crunching would actually result in more disk activity as you have 4 threads on your CPU, so if all are being used, that's 4 times the disk activity compared to the 1 video card.


To find out what is actually doing all the disk activity, start up Task Manager, select the Performance tab, and down the bottom is "Open Resource Monitor", click on that, then click on the Disk tab to see what's going on there.
Grant
Darwin NT
ID: 2023558 · Report as offensive
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2023587 - Posted: 18 Dec 2019, 15:51:03 UTC - in response to Message 2023541.  

when you start and complete a WU, there is some disk reading and writing that happens and it records the results, writes some info about the WU, and then the files get deleted when you send them back to the project, and files created when you download more.

CPUs are relatively slow, so the amount of disk activity is rather low. but GPUs are several orders of magnitude faster so all of those previous activities just happen more often, leading to more disk activity.

try have several very fast GPUs on single system with highly optimized apps. I have all this happening every few seconds lol.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2023587 · Report as offensive
Profile Mr. Kevvy Crowdfunding Project Donor*Special Project $250 donor
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 15 May 99
Posts: 3776
Credit: 1,114,826,392
RAC: 3,319
Canada
Message 2023591 - Posted: 18 Dec 2019, 17:20:44 UTC
Last modified: 18 Dec 2019, 17:21:40 UTC

I would try in BOINC client Options > Computing preferences > Computing tab setting "Request tasks to checkpoint at most every [x] seconds" to a value longer than the longest GPU work unit completion time (not Astropulse.) I see from your computer that the longest was 602 seconds, so I would set it to 700 (or the highest BOINC allows.) Then the GPU tasks won't write to disk midway, or at least will do so minimally.
ID: 2023591 · Report as offensive
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2023673 - Posted: 19 Dec 2019, 6:13:27 UTC - in response to Message 2023558.  

My only concern is that it's putting undue stress on the drive, and I'd rather avoid any unnecessary wear and tear. Interesting idea about whitelisting the folders - I had to do that quite a long time ago with my Thunderbird folder, I forget exactly why but it was really hammering performance, and I can see where this could be a similar issue so I've whitelisted. Got the resource monitor running too, so we'll see what that reveals. Thanks!
ID: 2023673 · Report as offensive
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2023675 - Posted: 19 Dec 2019, 6:16:37 UTC - in response to Message 2023591.  

I'm learning things! I'll give this a try as well, thanks for the suggestion.

I'll report back in a few days, hopefully with good news.
ID: 2023675 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 2023683 - Posted: 19 Dec 2019, 7:45:45 UTC - in response to Message 2023673.  

My only concern is that it's putting undue stress on the drive, and I'd rather avoid any unnecessary wear and tear.

If you were running database queries, that would be working the drive hard, but with the reading and writing that Seti does, even when running a dozen GPUs & 30+ CPU cores, it really is a very, very, very low workload for a HDD.
Grant
Darwin NT
ID: 2023683 · Report as offensive
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2024852 - Posted: 25 Dec 2019, 3:26:11 UTC - in response to Message 2023683.  

Okay, I'll take your word for it. I tried running it with the data directory moved to a flash drive, to alleviate the load on the HDD, which seemed to work at first - but I very quickly started getting computation errors on units being worked out by the GPU. Issues with reading/writing to the slower media? In any event, I've moved it back to the HDD so we'll see how it goes. I may be back with followup questions. :)

Thanks to everyone for all your help!
ID: 2024852 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 2024856 - Posted: 25 Dec 2019, 3:47:44 UTC - in response to Message 2024852.  

I tried running it with the data directory moved to a flash drive, to alleviate the load on the HDD
HDDs are built for lots of activity, USB flash drive's aren't.


Your driver version (v441.41) will result in errors on some Arecibo WUs. The last driver that worked with all WUs was v431.60
Grant
Darwin NT
ID: 2024856 · Report as offensive
Darrell Wilcox Project Donor
Volunteer tester

Send message
Joined: 11 Nov 99
Posts: 303
Credit: 180,954,940
RAC: 118
Vietnam
Message 2025128 - Posted: 27 Dec 2019, 0:31:27 UTC - in response to Message 2023541.  

@ Andy Chalk
Has any other change been made that has resulted in over-committing RAM? Page thrashing could result.
ID: 2025128 · Report as offensive
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2025899 - Posted: 1 Jan 2020, 16:35:09 UTC - in response to Message 2025128.  

Not that I'm aware of, unless it was something that could have changed in a Win10 update. But no hardware changes (in years) and I haven't noticed any other performance changes. I think the great likelihood is that it's always been like this (or at least for quite a long time) but I only just noticed recently.

Moved Boinc back to the HDD and the frequent reads/writes have resumed, but it's not doing anything "alarming" I guess. My only concern was that it's putting undue strain on the drive (plus a little surprise that GPU computing would be so much more R/W intensive than standard CPU computing) but the general consensus is that it's perfectly normal, nothing to worry about, so that's where I'll leave it.
ID: 2025899 · Report as offensive
Andy Chalk

Send message
Joined: 19 May 99
Posts: 10
Credit: 16,867,222
RAC: 18
Canada
Message 2025901 - Posted: 1 Jan 2020, 16:35:50 UTC - in response to Message 2024856.  

True, but flash drives cost five bucks and take two seconds to swap out. :)
ID: 2025901 · Report as offensive

Questions and Answers : GPU applications : GPU computing causing excessive hard drive activity


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.