Questions and Answers :
GPU applications :
Upgrade from GTX275 to GTX 295 Question
Message board moderation
Previous · 1 · 2 · 3 · 4 · Next
Author | Message |
---|---|
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
O.K. Will try that first. A straight change of Boinc version should not affect work in-flight with the Science App. F. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Well Fred gone back to 6.6.26 and still only the 2 GPU's. Well today is another day so will try something else later. Let you know how things go tonight. Dave |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Well I have to take the "Bull by the Horns" and try and solve this unseen 295 problem. Going back to Boinc 6.6.26 did not work,and am now back on 6.6.36. I will remove both cards and put an older type card in the system, and remove the drivers,then try a new install of the cards and drivers. At least I have not lost any ground on seti stats as the one 295 that is running is working as well as the 2*275's that were in here. To anyone else reading this. If you have any ideas please feel free to chip in. Dave |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Well I have to take the "Bull by the Horns" and try and solve this unseen 295 problem. try this. we had trouble getting most versions of boinc to properly recognize and use 2 physical devices... 6.6.11 worked well except it had other issues mostly dealing with elapsed time reporting... i am now using 6.9.0 from svn source and it works perfectly with 2 devices and time reporting.. i suspect this may be your problem with 6.6.36. can't hurt.. give 6.6.11 or 6.9.0 a try and i would bet it sees both devices. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Thank's for the info Chuck but can you direct me to the svn source? to d/load files from. Dave |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Thank's for the info Chuck but can you direct me to the svn source? to d/load files from. make sure svn is installed and then just do this to get the current source svn co http://boinc.berkeley.edu/svn/trunk/boinc there are also links for windows svn clients.. this is the main page to check out http://boinc.berkeley.edu/trac/wiki/SourceCode i suspect the version is 6.10.x now.... |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
Thank's for the info Chuck but can you direct me to the svn source? to d/load files from. Without svn you can get any version you require from here. F. |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Thank's for the info Chuck but can you direct me to the svn source? to d/load files from. true but that page doesn't help me on my stuff since the highest version they have for 64bit linux is 6.6.37 which can't handle multiple devices properly. at least svn has the source for fixed multiple devices. i see 6.10.x out there but not for linux 64 only for linux 32. thats the only problem with getting precompiled stuff. have to wait till someone decides to make a distribution for my arch. then it is also generically compiled to run on the most machines which is necessary for a binary distro. i like to compile specifically for each machine. since he has windows though he can get the 6.10.x from there and see how it goes for him without the extra work of compiling. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Well I have now updated Boinc to 6.10 and rebooted the rig and still not seeing the second card. Using Everest by Lavalys I can see that the temp of the gpu's Are: GPU1 59c GPU2 70c GPU3 63c and GPU4 49c. CPU temp is 48c 100% loaded The 2 GPU's that are running are returning 608's in around 3.5 mins to 13.5 mins so they are working ok. I think the only option left to me is as i posted earlier. Dave |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Well I have now updated Boinc to 6.10 and rebooted the rig and still not seeing the second card. odd. wonder what is causing this. i have seen system configs in the stats running 4 295 devices for 8 cuda all working properly.. wonder what they do differently... im at a loss. sorry. i covered everything i knew about multiple devices in getting my gtx285 and tesla working at the same time. there is probably some little thing being missed causing this but as far as i can tell it has all been covered. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Last night I did some testing of various options with the GPU cards. I found that in the Nvidia control panel there is an Enable Multi GPU Mode, and a Do Not Use Multi GPU Mode. This was not there when I had the two 295's in the rig. There was only the Enable Quad SLI Mode, and Do Not Use SLI. I know that I tried it in both modes when I had both cards in, and was in Do Not Use SLI Mode when the problems were happening. I have now been running this i7 rig over night with 1 BFG GTX295 in it with both GPU1 and GPU2 running ok with not a problem, no errors on the work completed. I will now install the second 295 and hope that Boinc sees it. This rig is Vista 64bit and Boinc 6.10 nothing overclocked. Dave [Edit] Just fitted the second card and in Do not use sli mode and Boinc still only seeing the first card and crunching on it. I am now wondering if there is a way to force boinc to see all gpus? |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
Only the cc_config file, as far as I am aware, and I assume you still have that in place with the <use_all_gpus> flag set? And, while this may be a little "grandmother / egg-sucking", you are sure the cc_config file is in the right folder and is being seen by Boinc? (Set one of the log flags in there and check the effect in your BM log after re-reading it). Really all I can think of ATM. F. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Hi Fred yes done that and here are the few lines with it set to 0 and then set to 1. 30/08/2009 10:14:58 SETI@home Starting task 27mr08ag.24858.18477.16.10.150_0 using setiathome_enhanced version 608 30/08/2009 10:22:02 Re-reading cc_config.xml 30/08/2009 10:22:02 Re-read config file 30/08/2009 10:22:02 log flags: task, file_xfer, sched_ops 30/08/2009 10:23:19 Re-reading cc_config.xml 30/08/2009 10:23:19 Re-read config file 30/08/2009 10:23:19 Configured to use all coprocessors 30/08/2009 10:23:19 log flags: task, file_xfer, sched_ops I am now lost as what to do and am a bit frustrated. I know that boinc has issues with reporting the actual cpu details and all I can think of is that it may have a problem getting details of this new version of the card,althought it sees and is crunching on the first 1 ok. Dave |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Hi Fred yes done that and here are the few lines with it set to 0 and then set to 1. i just thought of something... in the project directory make sure a file named "number_of_gpus" exists. typically it contains a 1 or 2. maybe you should put a 4 in there. can't hurt to try it. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Hi Fred Good thought looked in the project folder and there isn't a file called that! Looked on the other rig and there is I will copy the file from that rig to this and try as it is then change to 4 if it doesn't work.Here's hopeing Thank's Something to try scratching my head at the moment |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Fred No luck just the same with a 1 or a 4 in the number of gpus file. After lunch I am going to take the cards out of this rig and throw them in the river,(Only joking). I am going to put them in my AMD rig to see if it is a rig problem. The AMD rig is still Vista but 32 bit as opposed to this one which is 64bit,it has the to GTX275's in at the moment with the same Nvidia drivers and is running 6.6.36. I may have a power supply issue but if I get the message 4 GPU's then that will suffice,I don't need to run the task's. Let you know how I get on later,as there is a F1 race at 1pm. Dave Thank's for your help so far it is appreciated. |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Fred No luck just the same with a 1 or a 4 in the number of gpus file. hmm number of gpus should have controlled what was being looked at. and it was in the seti project folder living with the apps, app info file etc... should be a file with 1 character in it on the first line. you have a point. if there is not sufficient power sometimes motherboards hide resources and shut them down. is there a power connector on the mobo for pci-e? mine has one that is to be used if more then 1 device is plugged in. i suppose it is possible that at startup since all is at idle everything is happy and wants to run, and then when faced with cuda processing the cards discover not enough power, panic and begin shutdowns in order to continue running at least in single gpu mode. what size power supply is in that? is it multi-rail? and are different rails feeding different cards? might want to compare the specs to be sure a rail can actually supply the amount of current the 295 wants. for 2 of those cards plus an i7 and associated gear an 850w would be extremely marginal. i barely have enough headroom in mine and my cards don't draw nearly as much as the 295s do. each 295 is rated at 289 watts so thats 578w plus processor/mobo/ram/drives/fans add another 150 to 200w so a maxed out supply would have to supply 778w absolute minimum just to run for a bit before it burned up not leaving much of a margin for an 850 plus considering the 'duty cycle' ratings of the psu only those psu rated 850w continuous even stand a chance. i would use absolute minimum an antec signature 850 (can run at 850 all day without breaking a sweat), preferring a 1000w. or 2 supplies a 650w for powering just the gpus and another 650w for the rest of the system. problem with that would be finding a 650 that can supply the 12v rail current requirements of the 295. might have to go 850 for that anyway. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Thanks Chuck for your input. I have an 850w dual rail active PSU and a kill-a-watt inline. At the moment running 8 cpu threads and the 2 GPU's on the GTX295 It is only showing 460 watts total power so nearly 400 watts to go. Dave |
Fred W Send message Joined: 13 Jun 99 Posts: 2524 Credit: 11,954,210 RAC: 0 |
What's confusing here is that everything else sees all 4 GPU's. Only Boinc refuses to see them so I have (so far) discounted motherboard / PSU issues. It will be interesting to see what the AMD rig makes of them. F. |
FiveHamlet Send message Joined: 5 Oct 99 Posts: 783 Credit: 32,638,578 RAC: 0 |
Yes I thought that Fred I have to go out for a little while so I will report later. The 295 that is crunching at the moment is turning the wu's round in about 13 mins on each gpu and hasn't missed a beat so 6.10 and the drivers seem to be working well. I was thinking of going on the Boinc Forum and see if anyone on there can help. I can't see a thread specific to this peroblem but you never know. Dave |
©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.