Message boards :
News :
Nbody WU Flush
Message board moderation
Previous · 1 . . . 12 · 13 · 14 · 15 · 16 · 17 · 18 . . . 22 · Next
Author | Message |
---|---|
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
5/10/2022 8:16:07 AM | Milkyway@Home | update requested by userYou should be getting something like this and I'm not sure why - is this on a computer without a GPU?: 10-05-2022 11:52 AM Not requesting tasks: don't need (CPU: ; AMD/ATI GPU: job cache full) You could try turning on some debugging flags so it puts more info in there. Go into Boinc Manager (can't find it in Boinctasks), options, event log options. I think sched_op_debug gives info for contacting the project. Having nothing after the : is a bug and they're looking into it. Also, what's happening at this point? Are you out of tasks to do? What's your buffer set to? What projects have you allowed new tasks from? |
![]() Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 0 ![]() ![]() |
5/10/2022 8:16:07 AM | Milkyway@Home | update requested by userYou should be getting something like this and I'm not sure why - is this on a computer without a GPU?: I should, or shouldn't be getting this? It's on 2 computers, one with a GPU and one without. Don't know where you got the job cache full message, if it was the Ryzen 9, that PC is stable and running fine. I may try debugging as a last resort. Shutting the client down and restarting did not help. Good to hear someone is looking into it. Yes, 2 computers are out of tasks. they are set to n body only. Buffer is set to 0.1 days of work with 0 days additional. They are slow computers by today's standards. The only project I run is MW. |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
I should, or shouldn't be getting this?I've had problems with daft messages like yours, and reported it to Boinc, the programmers have admitted that some circumstances cause this bug. To speak to programmers you need to get into the Boinc Github forum. They aren't in the normal forum. But you can post in there without any privileges, you just file a bug report, or respond to someone else's. This is mine: https://github.com/BOINC/boinc/issues/4738 Unfortunately I can't see anyone actually taking action on it. If you have a GPU on the machine, I would have thought the brackets would contain the words CPU and GPU. The one I pasted was from a machine with both here. The lack of info for CPU is the Boinc bug. The GPU full message is because I have lots of Primegrid on that GPU and it doesn't need to get MW. The debug option I mentioned just adds extra lines of info in the messages so you can work out WTF it's doing. It says how much work it asks for etc. In the meantime, what happens if you increase the buffer? Stick it on 5+5 days (max allowed is 10 days total). You won't get a huge amount, there's a server limit of 100 tasks on Nbody. |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
YesYES in the Pentathlon, but not in the Prime Grid run ones as it's different, and they have more than half a dozen per year each lasting from a few days to a few weeks.Ah ok, I never do those. For badges I assume? The parenthesis means it's the Team I'm on not what the stats are for, the stats are for me mikey, and YES they really do represent alot of my crunching over the years ![]() |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
5/10/2022 8:16:07 AM | Milkyway@Home | update requested by userYou should be getting something like this and I'm not sure why - is this on a computer without a GPU?: As Peter said try increasing the cache size, your 'slow' pc is only asking for 0.1 days of work but since it's a 'slow' pc you could in reality be ending up with .3 days of work or more if you actually got a task so MW can't send you any tasks. Another problem with the cache is that the cpu and gpu use the exact same cache settings meaning if you have alot of gpu tasks a project may not send you any work because 'the cache is already full', and why alot of people never crunch for the same project with both the cou and gpu on the same machine, ie machine 1 does cpu tasks from project a and gpu tasks from project b, while machine 2 does the exact opposite giving you the ability to control the cache size for your needs. |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
Another problem with the cache is that the cpu and gpu use the exact same cache settings meaning if you have alot of gpu tasks a project may not send you any work because 'the cache is already full', and why alot of people never crunch for the same project with both the cou and gpu on the same machine, ie machine 1 does cpu tasks from project a and gpu tasks from project b, while machine 2 does the exact opposite giving you the ability to control the cache size for your needs.Is this related to me not getting work and my complaint to Boinc? I had a lot of CPU but no GPU work, I don't normally see it but I was bunkering: https://github.com/BOINC/boinc/issues/4738 |
![]() Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 0 ![]() ![]() |
Truth be told, I'm trying to fine tune these rigs for best throughput using a combination of buffer size through the boinc manager options, and also the project preferences(primary default, and 3 computers labeled home, work, and school). I'm pretty sure now that I messed up on the project preferences. That is probably what caused most of the hate and malcontent. I have increased the cache size a bit for the 2 slow computers, and that seems to be nicely dialed in at this point. On the big iron, I'm still working on cache tuning, since it has quite a bit of flexibility, with nvidia and amd GPUs, and 24 threads of CPU. Another problem with the cache is that the cpu and gpu use the exact same cache settings meaning if you have alot of gpu tasks a project may not send you any work because 'the cache is already full', and why alot of people never crunch for the same project with both the cou and gpu on the same machine, ie machine 1 does cpu tasks from project a and gpu tasks from project b, while machine 2 does the exact opposite giving you the ability to control the cache size for your needs.Yes, and I find that to be exceptionally annoying! The slow NVIDIA GPU is the culprit here, since those tasks seem to be much easier to get, and they hog most of the available buffer, and are slow to get processed. I can control that by periodically toggling NVIDIA GPU on and off in the project preferences for that computer. Not all that fun to do, but it is manageable since it is such a slow producer. In consideration of the above, and many other comments on the boards, I have some proposals. Proposal #1. Have the client set up buffers for each task class. In MW right now, the classes are CPU separation, CPU n body, NVIDIA separation, AMD separation, and INTEL separation. 5 classes. Someday, the hope is that n body will move into the GPU space, and that could mean an additional 3 classes. Proposal #2 I have noticed that, on the big iron, even if I set up the buffer to 5 days plus 5 additional days, I get no where near that amount of work. It seems to be task limited to 500 or 550 tasks or so. So the second proposal is to change the buffer units from number of days, to number of tasks. This would allow one to set a very small number for a specific class, potentially even zero, in favor of a larger number of another class, or classes. Implementation of these proposals could could come in handy in case one application swamps out another one by a factor of say, 10,00 or so. Oh, wait.... |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
Boinc is a mismatched cobbled together added to over decades piece of crap. There are 10 different places to put settings, some of which might override others if it feels like it. When I want to say "do X" I actually have to say do everything except ABCDEFGHIJKLMNOPQRSTUVWYZ. |
![]() Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 0 ![]() ![]() |
Boinc is a mismatched cobbled together added to over decades piece of crap. There are 10 different places to put settings, some of which might override others if it feels like it. Yep. I sure hope we are not polishing the t*rd here! |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
I'm happy if all my chips are running flat out. If Boinc/the project/anything prevents that, I switch project or run two projects at once. As long as I see 100% usage on every CPU and GPU, I'm helping somewhere. I'm giving up on these silly competitions, bunkering tasks, finding servers can't keep up, etc. I run what I want to run. |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
Another problem with the cache is that the cpu and gpu use the exact same cache settings meaning if you have alot of gpu tasks a project may not send you any work because 'the cache is already full', and why alot of people never crunch for the same project with both the cou and gpu on the same machine, ie machine 1 does cpu tasks from project a and gpu tasks from project b, while machine 2 does the exact opposite giving you the ability to control the cache size for your needs.Is this related to me not getting work and my complaint to Boinc? I had a lot of CPU but no GPU work, I don't normally see it but I was bunkering: It could be the easy answer might be to turn off gpu work then ask for new tasks then turn the gpu tasks back on again., another option is to temporarily raise your cache size, be careful to not get tasks from unwanted projects if you do this, and then ask for new tasks and you should get them, I would combine this last part with turning off the gpu tasks too. As you know Boinc was originally designed for using just our cpu cores and gpu's were not even in the realm of thinking at that point, they were waaaay too expensive anyway, but over time they could have separated the gpu and cpu caches but they never did they just kinda attached getting gpu work into the original Boinc and 'voila it worked now don't mess with it' thinking. Gamers in fact get screwed sometimes when they try to crunch if they have 2 or more gpu's attached together for gaming as Boinc doesn't always recognize that they are separate gpu's. |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
[/quote]Truth be told, I'm trying to fine tune these rigs for best throughput using a combination of buffer size through the boinc manager options, and also the project preferences(primary default, and 3 computers labeled home, work, and school). I'm pretty sure now that I messed up on the project preferences. That is probably what caused most of the hate and malcontent. I have increased the cache size a bit for the 2 slow computers, and that seems to be nicely dialed in at this point. On the big iron, I'm still working on cache tuning, since it has quite a bit of flexibility, with nvidia and amd GPUs, and 24 threads of CPU. The slow NVIDIA GPU is the culprit here, since those tasks seem to be much easier to get, and they hog most of the available buffer, and are slow to get processed. I can control that by periodically toggling NVIDIA GPU on and off in the project preferences for that computer. Not all that fun to do, but it is manageable since it is such a slow producer. [/quote] Can you perhaps move the slower gpu to the I3 as that would enable you to manage each gpu separately? |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
Gamers in fact get screwed sometimes when they try to crunch if they have 2 or more gpu's attached together for gaming as Boinc doesn't always recognize that they are separate gpu's.Can't the task run on the dual GPU like the game does? |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
Can you perhaps move the slower gpu to the I3 as that would enable you to manage each gpu separately?I had to seperate a Baffin and a Tahiti as Boinc had a hissy fit. It created a ghost card, it thought I had two Baffins as well as the Tahiti. Nothing else thought there were two, just good old Boinc. I did complain to Boinc but got the usual "your fault for having the wrong drivers". Well I have the best driver allowed for both cards. |
![]() Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 0 ![]() ![]() |
Can you perhaps move the slower gpu to the I3 as that would enable you to manage each gpu separately?Hmmm good question. I'll look into that. |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
Gamers in fact get screwed sometimes when they try to crunch if they have 2 or more gpu's attached together for gaming as Boinc doesn't always recognize that they are separate gpu's. No, Boinc does not detect gpu's like that, I believe that was part of the 64bit total Boinc makeover that got put on hold when they lost their funding and had to go to work everyday in 'real' daytime jobs. They were envisioning a new version of Boinc that had some updates that maintained the 32bit stuff as it is and a totally rewritten totally 64bit version that did not include any 32bit stuff in it. That would enable the projects to read the full amount of the memory on the gpu's instead of the existing 32bit limit of 4gb. That alone would enable the projects to limit tasks to specific gpu's based on the memory on the gpu's, ie more than 8gb. They had a whole long list of things they were working on that all got put on hold even as they continue to search for new funding. Now updates seems to more 'okay that's now broken so lets program in a fix' and we get a new version. The Android side is mostly done by one person but they do have a group of people testing things on a variety of devices to ensure the 'fix' works as expected. |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
That would enable the projects to read the full amount of the memory on the gpu's instead of the existing 32bit limit of 4gb.I don't have such a card, but are you telling me those rich folk like Ian and Steve who have 16GB cards have 12GB sat idle?! |
![]() ![]() Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 27 ![]() ![]() ![]() |
Can you perhaps move the slower gpu to the I3 as that would enable you to manage each gpu separately? To me the question is always the size of the power supply though a Teammate of mine used to run a 2nd desktop power supply next to his cases when he didn't have a big enough psu in the machine or to put in the machine. You can buy plugs that make the psu think it's connected to a motherboard so when you plug it in it works just fine though it does need it's own power plug of course. https://www.amazon.com/outstanding-Supply-Bridging-Connector-Jumper/dp/B08B62S6Q1/ref=sr_1_51?crid=15QO8YZE685CO&keywords=turn+on+pc+power+supply+without+a+pc&qid=1652323337&sprefix=turn+on+pc+power+supply+without+a+pc%2Caps%2C136&sr=8-51 |
![]() Send message Joined: 5 Jul 11 Posts: 990 Credit: 376,143,149 RAC: 0 ![]() ![]() |
To me the question is always the size of the power supply though a Teammate of mine used to run a 2nd desktop power supply next to his cases when he didn't have a big enough psu in the machine or to put in the machine.I use these, way cheaper than PC PSUs: https://www.ebay.co.uk/itm/313454063267 You can buy plugs that make the psu think it's connected to a motherboard so when you plug it in it works just fine though it does need it's own power plug of course.Or do what I did. Take a bit of paperclip and join two of the pins together. You really don't need to buy something to connect two things. |
![]() Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 0 ![]() ![]() |
Or do what I did. Take a bit of paperclip and join two of the pins together. You really don't need to buy something to connect two things.OK sparky, you're scarin' me now! |
©2024 Astroinformatics Group