Message boards :
Number crunching :
get more ati work
Message board moderation
Author | Message |
---|---|
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
I have 3 ati gpus but i cant seem to get enugh to keep them busy all finish their work in under 2mins. Can i do anything to get more work or is this just how it works. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
You should get 12 WUs per GPU, so you should have 36 in your queue at any one time. What figure is in Tools - Preferences - Network Useage - Work Buffer? And do you have any other Project running or queued in BAM that is not Suspended ? EDIT: One suggestion not connected with the question .... looks like you are water cooled, but you will save substantial heat and power from current levels by taking memory down from your current memory setting of 1300 to at least 500, preferably 300. Use MSI Afterburner/RivaTuner to get it down that low, and save as a Preset in Catalyst. MW is not sensitive to memory needs, and has a low bandwidth requirement, anything above 250/300 just wastes heat and power to no effect. Regards Zy |
Send message Joined: 15 Jul 08 Posts: 383 Credit: 729,293,740 RAC: 0 |
Use MSI Afterburner/RivaTuner to get it down that low, Or Ray Adams ATI Tray Tools... |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
lol i get no buffer at all it seems, dont use bam and trying to use lower memory clocks does not work lol cat and msi fight a ton 1 card downclocks 2 dont 2 down clock 1 wont. So i just left them at the 3dclocks i use i didnt see much in temp differnce if any when i did get them to downclock. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
lol i get no buffer at all it seems, dont use bam and trying to use lower memory clocks does not work lol cat and msi fight a ton 1 card downclocks 2 dont 2 down clock 1 wont. So i just left them at the 3dclocks i use i didnt see much in temp differnce if any when i did get them to downclock. Assuming you are water cooled .... probably wont see much temp difference as most gets taken away by water cooling, but you will use less power. I can see your stats on BOINCStats, a fair chunk, so you must have been running WU queues before ?? If you dont use the BAM Client, what are you using to manage Project WUs? Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
lol i get no buffer at all it seems, dont use bam and trying to use lower memory clocks does not work lol cat and msi fight a ton 1 card downclocks 2 dont 2 down clock 1 wont. So i just left them at the 3dclocks i use i didnt see much in temp differnce if any when i did get them to downclock. i ment im not using the bam stuff on boincstats i just have collatz and milkyway added to my boinc manager. For whatever reason collatz and milkyway wont give me much work but primegrid will and so did dnetc when it was up. Yea im watercooled a i7@4300 with my 5870 and 5970 both under water with a rx360 and rx240 rad. gpus dont pass 40c and vrms on 5970 gpu1 and 5870 dont hit 50c the vrms on 5970 gpu2 hover between 70-80c on milkyway cause thers no active channels for them. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
There shouldnt be a queue problem then .... the buffer length you set is compared against the total number of WUs held in your Client from Projects who are not suspended. The fact that they are waiting to be crunched because of date pre-emption from the running project, can get forgotten. So if the total time (as shown in the Client, never mind it maybe a wrong time) of WU's queued is less than the Buffer you should get more - well you will ...:) Project limits can overide BAM settings, but not to the extent you are showing. (Collatz limit is 150 WUs per machine, and MW is 12 WUs per GPU) There have been a few problems recently at MW servers, but thats over with, right now supply is fine. Are you having a queue problem with other projects? Whats the buffer set for (and the "time between connection buffer" - two separate figures) ? Regards Zy |
Send message Joined: 28 Feb 10 Posts: 120 Credit: 109,840,492 RAC: 0 |
I use Boinc Ver. 6.12.22. I think this client has no problem with getting WU's till the Limit. Even it's a beta it works without any problems. |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
There shouldnt be a queue problem then .... the buffer length you set is compared against the total number of WUs held in your Client from Projects who are not suspended. The fact that they are waiting to be crunched because of date pre-emption from the running project, can get forgotten. So if the total time (as shown in the Client, never mind it maybe a wrong time) of WU's queued is less than the Buffer you should get more - well you will ...:) Ah i fixed i had put in connect every 320 days lol changed that to 0 and now just picked up 19 work units on milkyway thanks a ton. lol 6mins of work units 3 gpus really run thru work lol. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Good Stuff :) Happy Crunching :) (You should get 36 with three GPUs, so watch the right hand buffer value) Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
One thing i noticed is im using the 0.57 version did they revert again to fix the new client. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
No - you need to use 0.62 - dont use 0.57, life too short for the excitement:) Have you got/had an app_info file lurking? Maybe leftover from that. A default clean MW install will auto supply 0.62 Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
yea it was a appinfo i had downloaded a optimized 0.62 to try to get more work but it would only get 0.57 work all fixed now lol got alot of work to keep my cards busy. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Excellent, well done - happy crunching :) Regards Zy |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Just thought - a suggestion .... You have a well tuned setup that will eat the MW cache. If there are server issues for example, and they take longer than 15mins, you are toast with no WUs. Frustrating when you wake up to look at a machine thats been empty for hours still running server imposed back off. Try the BOINC Project backup facility. To use that, set the resource share of the backup project to Zero. What will then happen is the BOINC Client will detect that MW has zero WUs, and auto switch to your backup project (the one you set to zero resource share). It will supply WUs one by one from the backup project as a resource slot is available. Once it detects MW is back in business, it will switch back and you get full load MW again. To do it, you will need to update your Client from current 6.10.60 (I suggest 6.12.22, its a Beta version but its fine for 99.99% of uses), then go to BOINCStats, open Host machines, click on yours, and set your backup project(I suggest PrimeGrid as its also low memory project, or use Collatz if you keep your memory at 1300) to zero. Update client by using the "sync with BAM" option in Client main menu, and you are good to go. The facility does work, so if there are issues, you have a config problem somewhere. Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
Just thought - a suggestion .... Lol i tried that but kept sending me cpu work even tho i dont do cpu work on boinc and set it not to. And im not 100% fixed i get 30 units but they dont send me new ones as i finish i have to manualy force a update to get new work. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
.... Lol i tried that but kept sending me cpu work even tho i dont do cpu work on boinc and set it not to. And im not 100% fixed i get 30 units but they dont send me new ones as i finish i have to manualy force a update to get new work. PrimeGrid's preferences page is wierd to say the least - on first sight. Drove me nuts when I first used it, its a common source of complaint and errors. With that preferences mechanism, there are two distinct parts, even though it may not strictly look this way. Note the "Location" heading for the preferences you are changing, then at the top there are selections, amongst others for CPU or AMD ot NVidia. Before you do anything else on that page, clear CPU checkbox, put tick in ATI checkbox, and make sure NVidia has no checkbox. At that point, move down the page (ignoring the default checkmarks that will now have appeared against CPU projec ts despite you saying "no CPU" - they are overidden by you putting "no cpu" at the top even though they show up), and put a checkmark against the PPS Sieve ATI app. Save preferences, go to the individual PC on the account page and click "details", change the Location settings at bottom of page to the one you just ammended. (ignore the fact that it put the cpu app checkmarks in place - with the CPU box at the top cleared, it will ignore those checkboxs}. Make sure you clear the checkbox "accept apps from any other project". You should be good to go. .... And im not 100% fixed i get 30 units but they dont send me new ones as i finish i have to manualy force a update to get new work. What is the "additional work buffer" in preferences set to, and do you have any other GPU work units in your Queue for which its parent project is not set to suspended? If you leave it to itself, with a start total of 30 WUs, {roughly} how many are left over if you did no manual updating for 10mins ? Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
just have milkyway added to my boinc manager right now finished 36 and it was just sitting i hit update and it reported 36 and sent me 16 lol. I had my work buffer set to 10 days i just set it to 0 to see if it fixes it. lol with dnetc 10 days gave me like 160 work units that would last me a entire day if it was down. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Set to zero will give you no buffer at all. Just leave it at 1 day, the Project limits you to 12 per gpu no matter what you set (unless you set a lower figure like - for example - 0.03 which on yours will make a cache of around 2 or 3 probably). The server was down for around 40 mins earlier today, so with no backup project you may have been caught with a server back off. Set to one day see what its like over time - buffer set at 1 day is plenty for the current MW limit, so if still an issue there's something else in play causing it. Given a stable server supply with no downtime, you should be fine. If not post again, see what the collective mind of the Board comes up with, if we cant figure something out :) Regards Zy |
Send message Joined: 31 Dec 09 Posts: 9 Credit: 12,434,250 RAC: 0 |
Sadly im about fed up with this crap wont upload on collatz or milkyway and iv reinstalled boinc gonna give it one more shot. |
©2024 Astroinformatics Group