1) Message boards : Number crunching : Really long runs (Message 36579)
Posted 19 hours ago by Profile skgiven
In my PC is a GTX770 and runs at max boost 1150MHz at 68-69°C. I didn't make any settings or changes with a program (AfterBurner, GPUTweak etc.). And there is a GTX780Ti in it. Also made no changes, it runs at 68-70°C and starts at 1060MHz with a new WU and after a while it runs at 875.7MHz which is its base clock. Is this because it are different types of GK's?


The base clock rates differ for different types of GPU, and manufacturers can alter this, so there are Factory OverClocks (FOC) models and reference models.

The GPU is not boosting due to a bug in the driver. 337.50 might fix that, or at least improve things, or you could use Jacob's fix.

Secondly, running a GPU at max boost for 24/7 has that a (huge) influence of the longlivety of the card?


Larger cards go a long way towards looking after themselves, but I prefer to take measures to decrease temperature (usually increase the GPU fan speeds and add system fans/leave the case door off, use a water cooler for the CPU...). If the GPU is being throttled because of the temperature it's probably worse than boosting to a higher clock, staying below the temperature point where it gets throttled, and only being throttled due to the Voltage cap.
2) Message boards : Number crunching : SWAN : FATAL : Cuda driver error 3 in file 'swanlibnv2.cpp' in line 446. (Message 36576)
Posted 19 hours ago by Profile skgiven
Maybe it's been fixed:

686x-SANTI_MAR419cap310-60-84-RND9954_0 6455855 21 Apr 2014 | 1:23:59 UTC 21 Apr 2014 | 7:15:39 UTC Completed and validated 20,788.57 2,199.78 18,300.00 Short runs (2-3 hours on fastest card) v8.21 (cuda60)

1_4_9-NATHAN_CMYB_run1-18-40-RND4568_0 6450430 20 Apr 2014 | 0:37:10 UTC 20 Apr 2014 | 5:43:28 UTC Completed and validated 17,994.03 1,632.46 16,650.00 Short runs (2-3 hours on fastest card) v8.21 (cuda60)
3) Message boards : News : acemdlong application 815 updated for Maxwell (Message 36573)
Posted 20 hours ago by Profile skgiven
I had SWAN_SYNC=0 set as a User Variable, from way back, and it worked as soon as I used the 8.41 app version. It was the case that it should be set as an environmental variable and should be set to 0, but it use to work when set to other numbers including 1. However, I remember Gianni or Toni wasn't happy with it being set to other numbers - don't know why?

I found that the GPU usage rose by ~2% (to 92% and 94%) on my one W7 system (770 and 670; GK104 cards), but I am using more than 50% of the CPU's (100% now). At 50%, for me, I saw no significant difference from the drivers (actually newer drivers were slightly faster), but others found that their GK110 cards were slightly slower with the newer drivers. I wonder if that is still the case?

Is SWAN_SYNC available for Linux?

Peter_M, you completed other work with the 8.21 app, so it's probably not an app issue. It sounds like the WU is/was just running very slow. Check the GPU clocks in NVidia X server and if need be restart the application, system or even abort the WU. Are you running CPU tasks as well as GPU tasks? It doesn't look like it but if you are try to suspend the CPU work.
4) Message boards : News : acemdlong application 815 updated for Maxwell (Message 36564)
Posted 1 day ago by Profile skgiven
Try setting SWAN_SYNC to 1.
I can't test that yet because I'm running two 8.40 WU's. Maybe tomorrow I'll get some 8.41 WU's...
5) Message boards : News : Recommended driver versions are 334.21 or later (Message 36546)
Posted 1 day ago by Profile skgiven
334.21 is the minimum driver version that we are sending cuda60s to. This is the driver revision that accompanied the first pre-prelease version of the CUDA 6 toolkit that worked for us.

Matt

334.21 is for Linux and adds support for Maxwell's.
However I have successfully returned CUDA6 work using 331.49 on Ubuntu 13.10 64-bit from a Kepler:

234x-SANTI_MAR423cap310-69-84-RND5084_0 6452642 20 Apr 2014 | 10:24:50 UTC 20 Apr 2014 | 12:58:06 UTC Completed and validated 9,100.43 1,596.87 18,300.00 Short runs (2-3 hours on fastest card) v8.21 (cuda60)

Presumably that rig will stop getting CUDA6 work now, unless I manage to upgrade to the latest drivers, which I don't really need to do as I don't have a Maxwell (and I should still get CUDA42/55 work).
6) Message boards : News : Recommended driver versions are 334.21 or later (Message 36536)
Posted 1 day ago by Profile skgiven
Not quite sure what it is you are asking here. If you have driver 332.21 you shouldn't be getting the cuda60 app.

Matt

In the first post of this thread you said 334.21,
The minimum driver required for this app is 334.21, but any later version is OK.


You meant to say 332.21. So you might want to fix that typo!

CUDA 6 support was introduced with the 331.40 Beta driver way back on the 30th Sept 2013, but couldn't be used until a CUDA 6 Toolkit became available to compile with, and wasn't really needed until the Maxwell's turned up.

The latest WHQL is 335.23, though I've been testing the 337.50 Beta, and would recommend that over the 335.23 driver which downclocked and didn't boost, though there is a workaround.
7) Message boards : News : acemdlong application 815 updated for Maxwell (Message 36504)
Posted 2 days ago by Profile skgiven
Matt, thanks for the support and effort you put in, it's much appreciated.
8) Message boards : News : acemdlong application 815 updated for Maxwell (Message 36498)
Posted 2 days ago by Profile skgiven
Actually, I have to put out another revision, as I inadvertently missed out G80 support in version 820 (yes, there are still enough GTX280s out there to care!).
So, tell me what you want in the next 10 minutes, and you might well get it.

Matt

I know what you mean but to be precise the G80's (CC1.0) are no longer supported. Ditto for the next incarnations G92... (CC1.1).
It's predominantly the GT200 (and mostly 55nm) models of the high end GF200 range that still just about, occasionally work (CC1.3).
Even the CC1.2 cards no longer work AFAIK.
9) Message boards : Number crunching : Really long runs (Message 36489)
Posted 2 days ago by Profile skgiven
At present this only applies to the GIANNI_lig3 WU's. Also discussed in the 'GIANNI ligand3-0-1 bad batch?' thread.
The other 'Long' WU's are actually much shorter, but I guess it's better to put them into the Long queue than the short queue.


Indeed I saw that thread, my fault. But as mine did not error, I considered them as a "good batch" and did not post there.
____________
Greetings from TJ


That batch was withdrawn, fixed, and re-released, and it's the only truly LONG run that GPUGRID presently has, though another batch comes close-ish - probably because that research line was started when the best GPU was a lot slower. Other threads exist on this general topic but it's often the case that new threads pop-up naturally that have been previously discussed and other threads naturally overlap, topics diverge and merge, and that's fine by me; nobody wants to trawl through a 300 post thread that was first brought up 4years ago and 90% of it is irrelevant. I was just making people aware of the other thread in case they missed it and because it's related ;)
10) Message boards : Wish list : CPU project thoughts (Message 36483)
Posted 2 days ago by Profile skgiven
Maybe I misunderstood the plan but I thought there was going to be a separate project?
As anticipated there are issues with supplying both CPU and GPU work to clients attached to other projects (which is just about everyone).
Will 'Scientific Publication Badges' be awarded against CPU contribution and will there be any mutual CPU and GPU work? If so there could be a lot more plastics out there ;p...
I also agree with having a separate credit system, but again, I think it would be best done via a separate project.
Obviously the main concern is the project management overhead, so if you think it's too much to do on a different project then so be it.


Next 10