Latest NV driver - 352.86

Message boards : Number crunching : Latest NV driver - 352.86
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Graham Middleton

Send message
Joined: 1 Sep 00
Posts: 1517
Credit: 86,815,638
RAC: 0
United Kingdom
Message 1681408 - Posted: 19 May 2015, 9:13:43 UTC
Last modified: 19 May 2015, 9:36:28 UTC

Hi All,

This latest NV driver released 2015/05/18.

I installed it to try on my resurrected system(ID: 7578105 )

It repeatedly restarts itself under high load - both with3 and 2 WU per GPU. I successfully ran with 3 before the system died.

The PC doesn't reboot or anything, just the drivers restart, but that looses the fan settings, so the GPUs get hot very quickly!

I would suggest that these new ones are not suitable for crunching...

I'm hoping that 350.12 is still good to go & if so I'll install that one.
Happy Crunching,

Graham

ID: 1681408 · Report as offensive
Profile Graham Middleton

Send message
Joined: 1 Sep 00
Posts: 1517
Credit: 86,815,638
RAC: 0
United Kingdom
Message 1681464 - Posted: 19 May 2015, 12:30:55 UTC

Yes, 347.88 seems to be stable on both my PCs for now. Staying with that one for the moment.
Happy Crunching,

Graham

ID: 1681464 · Report as offensive
Mike

Send message
Joined: 7 May 15
Posts: 3
Credit: 392,226
RAC: 0
Canada
Message 1681814 - Posted: 20 May 2015, 17:39:56 UTC

Yea I am having the same issue with 350.12.. Driver resets all the time when running SETI WU's. It will reset the default settings, so any OC's will default.

It says something about a memory related error. Next time it happens I'll try to see if I can find more detailed information on the cause.
ID: 1681814 · Report as offensive
Profile JaundicedEye
Avatar

Send message
Joined: 14 Mar 12
Posts: 5375
Credit: 30,870,693
RAC: 1
United States
Message 1681818 - Posted: 20 May 2015, 18:34:21 UTC

Hi Graham, I'm running 352.86 on a 750ti with no problems. I did a fresh install when upgrading as I've had problems in the past when upgrading nvidia. I've processed over 200 MB's with no issues. Can't speak for AP's as I haven't seen one in a week.....

"Sour Grapes make a bitter Whine." <(0)>
ID: 1681818 · Report as offensive
Oddbjornik Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 15 May 99
Posts: 220
Credit: 349,610,548
RAC: 1,728
Norway
Message 1681838 - Posted: 20 May 2015, 19:53:07 UTC - in response to Message 1681818.  

I've run 352.86 on my GTX 680 for a couple of days. No problems there...
ID: 1681838 · Report as offensive
Profile Graham Middleton

Send message
Joined: 1 Sep 00
Posts: 1517
Credit: 86,815,638
RAC: 0
United Kingdom
Message 1681858 - Posted: 20 May 2015, 20:56:14 UTC

Hmmm, both my PCs are very happy on 347.88, so I'll stay there for the moment.


It looks like TMMV bid time! A lovely one for someone to try to fault-find, either in Nvidia,or somewhere eelse. I'll help if I can, but I'm not that experienced with Modern Languages, Give me BASIC or Fortran any day!
Happy Crunching,

Graham

ID: 1681858 · Report as offensive
Profile JBird Project Donor
Avatar

Send message
Joined: 3 Sep 02
Posts: 297
Credit: 325,260,309
RAC: 549
United States
Message 1682249 - Posted: 21 May 2015, 20:23:24 UTC - in response to Message 1681818.  

I recall, it was a OpenCL 1.2 *bug in 350.12

Can anyone verify that 1.2 libraries or whatever it was, is *fixed in the 352.86 release? Applies to any OpenCL SETI app usage(which only APs at this time) - non-issue with other apps.
=
I understand that OpenCL 2.0 is coming soon; and CUDA <7 is already implemented/included in recent NVidia drivers; just waiting on Devs to execute/release apps - great stuff CUDA.

Combined with robust number of CUDA cores and fast, GDDR5 vram in these Maxwell chips-cards(I have a GTX 960 SC GM206-A chip - GPU)
-this thing flies already- even tho it's on a Z75 board with a PCIe 2.0 x 16 Bus and a non-Hyper threaded CPU
[until *tomorrow anyway when my upgrades get here(due today) and I get it built!]

The Hyper thread aspect unleashes Maxwell's Unified Memory for superfast I/O between CPU/GPU/RAM (and back)
which *should improve runtimes by itself
=

ID: 1682249 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1682298 - Posted: 21 May 2015, 22:12:43 UTC - in response to Message 1682249.  

350.12 should only be installed if you are comfortable downloading Raistmer's new modification and installing it into your system.

If you aren't sure how to do that then stick with 347.88.

At this point there isn't a need just yet to upgrade from this one.

If and when they decide to release new MB/AP formats that someone was talking about, then it may become necessary to upgrade.

However, if that becomes the case, then hopefully the guys at Lunatics will come out with a new installer.
ID: 1682298 · Report as offensive
Profile JBird Project Donor
Avatar

Send message
Joined: 3 Sep 02
Posts: 297
Credit: 325,260,309
RAC: 549
United States
Message 1682300 - Posted: 21 May 2015, 22:14:35 UTC - in response to Message 1681818.  

Hey dg..... have you witnessed better runtimes or throughput or increased GPU load or other improvements with this 352.86 driver? (that you can report)

My current 0.04cpu+0.5gpu config on CUDA50s already yields a 94-97% GPU load and pretty zippy at that

ID: 1682300 · Report as offensive
Profile JBird Project Donor
Avatar

Send message
Joined: 3 Sep 02
Posts: 297
Credit: 325,260,309
RAC: 549
United States
Message 1682313 - Posted: 21 May 2015, 22:49:05 UTC - in response to Message 1682298.  

Right Z.

Can avoid the thing (opencl 1.2 *bug) either by avoiding 350.12 driver or using a *fixed driver, if NVidia *fixed that in post-350.12 versions - the gist of my ? here.
=
I'm still stable and staying put with 347.52 til I hear otherwise

ID: 1682313 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1682611 - Posted: 22 May 2015, 15:14:24 UTC - in response to Message 1682313.  

Can avoid the thing (opencl 1.2 *bug) either by avoiding 350.12 driver or using a *fixed driver, if NVidia *fixed that in post-350.12 versions

You continue to talk about some "opencl 1.2 *bug" but there is no such thing.

It was only a name clash:
http://setiathome.berkeley.edu/forum_thread.php?id=77133&postid=1665102#1665102
http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2182&postid=53981#53981

Raistmer: "OpenCL 1.2 has little or no connection to this issue. AMD has OpenCL 1.2 support long ago and never complained about bool2. It's not in specification. But nVidia decided to reserve this type name."

'bool2' name was used in the SETI@home OpenCL apps but now "nVidia decided to reserve this type name"
So don't expect 'a fix' in next NVIDIA drivers as this is not a "OpenCL 1.2 bug"

P.S.
This is like:
If you make a new car in your garage and give it a name "Car2" you are free to do so.
Then some corporation decides to make its own line of cars under trade name "Car2" and reserves that name by patent/copyright

You will be forced to rename your "Car2" to something else as you can't expect the corporation to 'fix the name' in next release.

But this is not a bug.
And the issue is not because their "Car2" uses "Engine 1.1" or "Engine 1.2"
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1682611 · Report as offensive
Mike

Send message
Joined: 7 May 15
Posts: 3
Credit: 392,226
RAC: 0
Canada
Message 1683419 - Posted: 23 May 2015, 19:45:27 UTC
Last modified: 23 May 2015, 19:45:58 UTC

Yea the 352.86 drivers don't work for me either. Apparently its a big issue with quite a few people. The Nvidia forums are saturated with kernel driver crash reports. (apparently it glitches with Chrome, some webpage plug-ins, and some games.. So Nvidia will have to work on getting a fix going, if in fact it is 'fixable')

Personally, I just swapped back to 347.88. It seems rock solid. Been crunching for over 12 hrs and no crashes.
ID: 1683419 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1683460 - Posted: 23 May 2015, 21:44:13 UTC

IF You're having trouble with AP and the latest NVIDIA drivers I'll take all the blame.

Please open the offending .cl file and replace all occurences of "bool2" with "bool_2" in it.

"bool2" is a structure holding 2 boolean (true/false) variables and it was not a reserved word when I wrote the code for optimization purposes.

My Bad!

--
petri33
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1683460 · Report as offensive

Message boards : Number crunching : Latest NV driver - 352.86


 
©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.