Author |
Message |
|
I'm new to this project but not entirely new to BOINC. I'm switching from a GT430 to a GTX570 and having a little problem. After BOINC is up and running, and the timing appears to be right about when GPUGRID would be starting up, my display goes blank, and the computer is non-responsive (even via VNC remote). If I shut down BOINC after bootup, the system runs fine....it's only after BOINC starts up.
I had the driver already loaded for the GT430. It auto loaded for the GTX 570 during which I prevented BIONC from starting up. I reboot after the drivers are loaded (as instructed). Again, I make sure the driver is running and reflecting the correct video card before I fire off the BOINC manager.
I did not do a reinstallation of the driver. GPUGRID is in the middle of a task (I did not wait between tasks to make the switch). Also, I'm only running a 450 Watt PSU. I was thinking of maybe dropping a 600 watt in there. I've never used one of these super video cards, so I wasn't sure what the symptoms would be if there wasn't enough power.
BOINC 7.0.28
Windows XP SP3
GTX 570, Nvidia WHQL 306.23 drivers
The machine ID is 133536
I'm hoping this is familiar to someone, and can tell me "on yeah, just change this setting <here>, and you'll be fine". I'm really excited about upgrading my video card for some better crunching.
Thanks!
-MichaelMac |
|
|
|
I just went through upgrading video cards on 2 xp systems. Both had the correct drivers for the new card, but after numerous reboots I still had to reinstall the drivers fresh for the card to work.
On one machine there was a 660 Ti and simply adding a second 660 Ti still required a complete reinstall of the driver. So you should try that and then see if you still have a problem.
Also you can check in the Event Log in BOINC (ctrl+shift+E) and near the beginning of the log it will list the GPUs available for crunching. If the card is listed there with the correct driver it should work. |
|
|
|
It wouldn't be the first time drivers have to be reinstalled fresh for some reason! I'll give it a try and report back the results. I think I'll drop in a bigger power supply too.
Thanks! |
|
|
|
After trying the fresh driver install I'd make sure that the card is working normally in other 3D applications like 3D Mark (the last one which still works on XP, forgot which one). If this doesn't work the PSU is indeed a likely candidate. A good 450 W unit should be able to handle your config, but a cheap one (which may really only be able to deliver 300 W, before it catches fire) may be overwhelmed.
If you go for a new one it's not as much the maximum power you want to look at (about 500 W will be totally fine, I won't go below 450 W, though), but rather to get a quality one (Enermax, Seasonic, Be Quiet, Corsair etc.) with 80+ Gold specification. This will be more expensive, but quickly pay for itself when crunching 24/7 with a strong GPU.
MrS
____________
Scanning for our furry friends since Jan 2002 |
|
|
|
Ok. This is what I did and the results:
1. I installed the GTX570 card in the computer.
2. I booted it up, and re-ran the NVidia driver installation. I selected "clean install".
3. It rebooted.
4. Ran 3DMark, and during the second graphic test it failed the same way as before (blank screen, unresponsive computer, lights on).
5. I shut it down, and installed a Thermaltake 600 Watt PSU. I didn't see the 80+ Gold spec as suggested, but I figured it would do.
6. I booted it back up, and re-ran 3DMark which ran to completion.
7. I started up BOINC, and checked the event log. It does show the GTX570 GPU as available.
8. I started GPUGRID which ran for a while (maybe 10 minutes) then it failed the same way as before (blank screen, unresponsive computer, lights on).
9. I re-tried GPUGRID a few times with the same results.
10. I tried my Einstein project using GPU, and it's been running for several hours with no problems.
Now, is there something special that I need to do for GPUGRID that I missed? Some setting in the NVidia control panel that I need to shut down or turn off or uninstall or something?
I have another GTX570. I was going to try swapping it out to see if the results are the same. |
|
|
|
One more thing, the GTX570 card is an EVG, and the box says "Overclocked". Could overclocking cause this? If so, where do I go to change it, and what would I change it to? |
|
|
|
I've been reading some other Forum threads....could I possibly need more than a 600 watt PSU? Perhaps 850? Could power still be my problem? |
|
|
|
Sorry, that's EVGA, and it says "Superclocked" not overclocked. |
|
|
MarkJ Volunteer moderator Volunteer tester Send message
Joined: 24 Dec 08 Posts: 738 Credit: 200,909,904 RAC: 0 Level
Scientific publications
|
Well I have a couple of Palit GTX570's which I have recently removed. I have one in each machine with a Seasonic 750w power supply. Never had a problem. Maybe it's the power supply, is it a name brand one? The cheap ones usually can't actually deliver a decent amount of power on the rails, so it's always best to get a name brand. I personally use Seasonic in almost all my machines and I have one Corsair 1000w.
The other thing to try would be to get the 301.48 WHQL drivers and see if that makes any difference. There have been some problems reported with the newer 306.xx drivers.
Lastly if the EVGA cards are the reference design (have a single fan on the end of the card) then it's quite likely overheating. I had one before the Palit cards (they have a better cooling solution). You can use EVGA precision to force the fan speed and you'll be able to see how hot it's running. I have to do the same with some GTS450SP (single slot) cards as they overheat if one forgets to force the fan speed.
____________
BOINC blog |
|
|
skgivenVolunteer moderator Volunteer tester
Send message
Joined: 23 Apr 09 Posts: 3968 Credit: 1,995,359,260 RAC: 0 Level
Scientific publications
|
The problem is almost definitely the PSU; 105W TDP CPU, 219W TDP GPU - add the power requirements of the motherboard, RAM, HDD, Disk-player and you would exceed what a standard PSU could supply when running most projects. Einstein tasks utilize the GPU less so they require less power. Change it to a quality 550W PSU or better. No need to go higher than a 650W PSU, but do change it otherwise it's inevitably going to fail, and this often damages components (especially HDD, motherboard, RAM and GPU).
The TDP of the SuperClocked version is probably ~10% more, which would make it ~240W. So a task on that GPU would use ~180W running GPUGrid, possibly more, and ~140W running Einstein. Overall system power requirements would probably be ~300 to 350W, but on an 85+ PSU that's really a 350W to 410W draw, and more importantly quite a few Amps on the 12V rail to support the GPU. The PSU would probably need to continuously support >18Amps to run GPUGrid tasks, but only ~14A to run Einstein.
BTW. Just because the PSU says it supports 23A, for example, doesn't mean it will actually support that continuously. That could mean peak and only for a few seconds. I have a supposedly 500W EZCool PSU that says 34A on the +12VDC but it weighs next to nothing - from experience, there is no way it would continuously support a GPU drawing 180W.
____________
FAQ's
HOW TO:
- Opt out of Beta Tests
- Ask for Help |
|
|
|
PSU sounds likely, but he already tried the Thermaltake 600 W unit. Personally I don't regard TT products as high quality.. but they're probably not simple chinese fireworks.
Apart from trying out an even better PSU you might want to check GPU temeprature while running GPU-Grid. It shouldn't exceed 90°C (emergency shutdown may happen), 80°C should be fine and <70°C is certainly alright. If the temperature is high try to force a higher GPU fan speed. If this doesn't help clock the GPU down a bit (to test if the factory overclock is the problem).
MrS
____________
Scanning for our furry friends since Jan 2002 |
|
|
|
Ok. This is what I did and the results:
1. I installed the GTX570 card in the computer.
2. I booted it up, and re-ran the NVidia driver installation. I selected "clean install".
3. It rebooted.
4. Ran 3DMark, and during the second graphic test it failed the same way as before (blank screen, unresponsive computer, lights on).
5. I shut it down, and installed a Thermaltake 600 Watt PSU. I didn't see the 80+ Gold spec as suggested, but I figured it would do.
6. I booted it back up, and re-ran 3DMark which ran to completion.
7. I started up BOINC, and checked the event log. It does show the GTX570 GPU as available.
8. I started GPUGRID which ran for a while (maybe 10 minutes) then it failed the same way as before (blank screen, unresponsive computer, lights on).
9. I re-tried GPUGRID a few times with the same results.
10. I tried my Einstein project using GPU, and it's been running for several hours with no problems.
Now, is there something special that I need to do for GPUGRID that I missed? Some setting in the NVidia control panel that I need to shut down or turn off or uninstall or something?
I have another GTX570. I was going to try swapping it out to see if the results are the same.
When I replaced my 285 card with a 570, on an xp machine, a while back, I did the following:
1. Suspended all work units and projects within boinc.
2. Shut down the computer and unplugged it.
3. Replaced the video cards.
4. Rebooted the machine and doing a clean install.
5. I rebooted and cleaned out the register.
6. Restarted the projects and work units, and everything worked fine. Even the work unit that started on the 285 card finished on the 570 with out any errors.
7. I ran the first few work units at stock speed, before attempting any overclocking.
So, did you clean out your register? |
|
|
JeremySend message
Joined: 15 Feb 09 Posts: 55 Credit: 3,542,733 RAC: 0 Level
Scientific publications
|
The "clean install" option in the nVidia drivers is a misnomer. It's not nearly as clean as one would like to think. It only removes custom application profiles and driver settings, it doesn't actually uninstall the previous driver or any of the registry entries.
To do a fully clean driver install:
1. Uninstall the nVidia drivers via the Control Panel.
2. Reboot
3. Run a driver cleaning program to remove any remnants. Treexy Driver Fusion (formerly known as Driver Sweeper) and the free version of Driver Cleaner Pro are both very good
4. Reboot again
5. Install new drivers
450 watts should be enough for that setup if the PS is actually capable of delivering that much. If the PS is a few years old, it might not be able to deliver to juice anymore. They do start to deliver less power as they age. Cheap power supplies in particular age poorly, and they don't start out that great to begin with. |
|
|
|
Thanks to everyone for their helpful advice!
Ok, this is what I ended up doing as my next step.
1. I had not installed the EVGA Precision software so I could see what the temp was doing, so I installed this software.
2. I watched the temp and fan controls. The temp got up to 88 degrees celsius pretty quickly, and it was shortly after that the system shut down again.
3. So, I set the fan to max as suggested by MarkJ. The temp stayed around 82 degrees and ran for about an hour, no problems. It seems stable.
4. I did notice the PSU was too hot to touch. Really hot. That scared me a little, so I went ahead and switched out the 600 watt Thermaltake for an 850 watt Thermaltake (80+ Bronze).
5. I also added a feeder fan to the front of the box. I had a good rear fan blowing air out, and the PSU was circulating air out, but I was just using suction to feed air into the box. The feeder fan is blowing air into the box from the front. This set up a nice air flow of front to back (and up).
6. I powered it all up, and the GPU took a while to get up to 82 degree but is steady at that level. The PSU is hot, but not too hot to touch.
Everything seems to be running steady for a few hours now.
What I've learned; Power and Cooling are the keywords for running these video cards!
I'm going to keep an eye on it, and I'll post if anything changes. Again, thanks to EVERYONE for your help!
-MichaelMac |
|
|
|
*Cool* you've got it sorted out :)
Did you buy the PSU(s) new or did you have them laying around anyway or swapped them from another box? Because the heat you're feeling is actually your money being burnt directly in the PSU. More efficient PSU generate less waste heat, so they not only reduce your running costs, but also stay much cooler & more quiet (as there isn't as much heat to get rid of).
A few years ago I exchanged my good old Enermax (which reached almost 80+ Bronze) for a Gold unit.. it payed for itself within a year! (I was drawing 300 W 24/7 back then)
MrS
____________
Scanning for our furry friends since Jan 2002 |
|
|
|
*Cool* you've got it sorted out :)
Did you buy the PSU(s) new or did you have them laying around anyway or swapped them from another box? Because the heat you're feeling is actually your money being burnt directly in the PSU. More efficient PSU generate less waste heat, so they not only reduce your running costs, but also stay much cooler & more quiet (as there isn't as much heat to get rid of).
A few years ago I exchanged my good old Enermax (which reached almost 80+ Bronze) for a Gold unit.. it payed for itself within a year! (I was drawing 300 W 24/7 back then)
I had these laying around. It was a time, I remember with fondness, before power was that much of an issue.
Yep, I'm taking everyone's advice, and will be ordering some 80+ Gold units for these two machines. I am running these guys 24/7 so I am concerned with frying them out! |
|
|
|
Ok, a little followup.
I received my new power supplies. CoolMaster 800 watt 80+ Gold rated. WORLD OF DIFFERENCE! They run just slightly warm (barely), and temps are down and enough power for everything. So worth the extra money....and it wasn't that much more expensive.
I was stunned at the difference it made.
Thanks again for all your help!
-MichaelMac |
|
|
|
That's good to hear :)
MrS
____________
Scanning for our furry friends since Jan 2002 |
|
|