NVidia 436.xx and later drivers can cause very long compute times especially on Arecibo VHAR work units

Message boards : Number crunching : NVidia 436.xx and later drivers can cause very long compute times especially on Arecibo VHAR work units
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 14 · 15 · 16 · 17 · 18 · 19 · 20 · Next

AuthorMessage
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031349 - Posted: 8 Feb 2020, 3:18:50 UTC - in response to Message 2031346.  

I barely know what I'm doing. And I don't have plans to re-test. Hopefully the outputs from my runs, which took 2 days to get, are still useful.
ID: 2031349 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031395 - Posted: 8 Feb 2020, 12:36:30 UTC

Keith, how do I use the .cmd to run a CPU comparison?
ID: 2031395 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031405 - Posted: 8 Feb 2020, 13:12:34 UTC - in response to Message 2031395.  

Keith, how do I use the .cmd to run a CPU comparison?
With difficulty. Two possibilities:

1) Use the rescmpv5 tool manually, by adapting the parameterised code in mb_validate.cmd
2) Place the reference result files in ..\bench\Testdatas\ref (naming them according to convention - "ref-[appname]-[wuname].res") and running a live test on the same WU.

I'll be doing either or both of those later today on your results - many thanks for those - but I'm running slow this morning, and about to go out for lunch.
ID: 2031405 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031406 - Posted: 8 Feb 2020, 14:13:08 UTC

Okay. Thanks.
I'm not going to be redoing any testing. I did the best I could, and feel you all have valuable info if your goal is to validate the fix.
From my brief inspections, I think the fix is good :)
ID: 2031406 · Report as offensive     Reply Quote
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13164
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2031444 - Posted: 8 Feb 2020, 16:55:54 UTC - in response to Message 2031395.  

Keith, how do I use the .cmd to run a CPU comparison?

I don't like the Lunatics MBbench application. I think it has been surpassed in simplicity and capability by Rick's benchmark application benchMT.
https://github.com/Ricks-Lab/benchMT
But this tool is for Linux only unfortunately.
https://setiathome.berkeley.edu/forum_thread.php?id=83566#1965982
That is the one that I use. Already comes with preconfigured and run CPU tasks with the reference cpu application. Does both AP and MB.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2031444 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031448 - Posted: 8 Feb 2020, 17:37:19 UTC - in response to Message 2031331.  

Hi folks,

I recently went through all of my local "repro" examples ... against all 6 of my GPUs ... against both Cuda and OpenCL ... against drivers:
- 431.60 (known good NVIDIA public release)
- 431.68 (known good NVIDIA hotfix driver)
- 432.00 (known good Windows Update driver)
- 442.19 (recent NVIDIA driver with fix that looks good so far)

A .zip of the results can be found here:
https://1drv.ms/f/s!AgP0NBEuAPQRp-ZG322LD1BXy6rdAg

Richard is going to look them over, but if anyone else knows how to do that and wants to also inspect for validation, please feel free!

Regards,
Jacob Klein
OK, I'm starting to work out how best to handle those now. I think I'm going to concentrate on the results from:

* Driver 442.19
* OpenCL SoG app

(so far as I'm aware, after re-reading the first 100 or so posts in this thread, that should be the only fixed version of the only app which was causing a problem)

I've found 6 results from 28oc11aa.6787.6611.5.32.85 passing that filter (three GPUs in each of two machines). I'll need to rename the files to keep them separate.

And the same 6 results from each of

28oc11aa.13844.8247.7.34.54
28oc11aa.13844.8656.7.34.81
28oc11aa.2079.16836.10.37.50
28oc11aa.2079.18881.10.37.49
28oc11aa.2090.18472.12.39.249
28oc11aa.2090.22562.12.39.170
28oc11aa.2108.22562.13.40.187
28oc11aa.30967.11928.8.35.113
28oc11aa.30986.13155.9.36.98

in 'WUs from Old BOINC Data Copy'

I'd better make sure I've got all those WUs before I go much further...
ID: 2031448 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031451 - Posted: 8 Feb 2020, 17:59:19 UTC - in response to Message 2031448.  

I'd better make sure I've got all those WUs before I go much further...
Yup, got 'em. Consolidated 10 WUs, 60 result files - now to test my bulk renaming skills.
ID: 2031451 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031471 - Posted: 8 Feb 2020, 19:32:04 UTC

I had to leave some of the work for you :) Hope you get it all figured out!
ID: 2031471 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031474 - Posted: 8 Feb 2020, 19:43:31 UTC - in response to Message 2031471.  

I think I've got it prepped up. KNAbench relies on lots of distinct file names in a single folder: your preferred style is identical file names in separate folders. It took a while to transfer from one standard to the other, but I got there - having each set in a tight, formatted, structure helped a lot. Taking a breather to let my eyes relax in front of some telly - I'll go back and finish it off later.
ID: 2031474 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031491 - Posted: 8 Feb 2020, 21:21:16 UTC

Unfortunately, it's not doing the automatic comparisons I had hoped for, but it's generating a local set of result files. I'll either have to hack the bench script, or write my own to compare them offline. Tomorrow. [All day at home, sheltering from an expected big storm. Hope the power stays on...]
ID: 2031491 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031564 - Posted: 9 Feb 2020, 11:35:24 UTC
Last modified: 9 Feb 2020, 11:36:06 UTC

All that file renaming yesterday (and again to correct my error this morning!) was worth it. This is what I wanted to see.

All Jacob's results have been renamed as reference, with a spoof app name to indicate where they came from. The test was run on the GTX 1660 SUPER in my host 4292666, which is doing about 75% of the tasks in that machine (there's a 750 Ti in there as well). The machine itself is currently showing

State: All (1841) · In progress (167) · Validation pending (508) · Validation inconclusive (87) · Valid (1075) · Invalid (0) · Error (4)
which is pretty good. The four errors are all download errors from 28 December, so can't be blamed on the cards.

And the winner is ...

... NVidia's 442.19 driver. You'll see the occasional Q= 99.99% (probably due to floating point rounding errors - that's why we can't use direct file comparison for this sort of work), but the majority are Q= 100.0%. If we saw that level of accuracy for a new application, we'd have no hesitation in saying it was ready.

= MB Knabench 2.10 W32-W64 2012-02-18 by Kna + Simon + Joe
= mods: quick timetable, stderr, speedup/ratio, AppTimes
= /ref/ by Raistmer
= BOINC install detection by Richard Haselgrove

 10 testWU(s) found
   └─(28oc11aa.13844.8247.7.34.54.wu)
   └─(28oc11aa.13844.8656.7.34.81.wu)
   └─(28oc11aa.2079.16836.10.37.50.wu)
   └─(28oc11aa.2079.18881.10.37.49.wu)
   └─(28oc11aa.2090.18472.12.39.249.wu)
   └─(28oc11aa.2090.22562.12.39.170.wu)
   └─(28oc11aa.2108.22562.13.40.187.wu)
   └─(28oc11aa.30967.11928.8.35.113.wu)
   └─(28oc11aa.30986.13155.9.36.98.wu)
   └─(28oc11aa.6787.6611.5.32.85.wu)

 0 reference science app(s) found

 1 science app(s) found
   └─(MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0)

======================================


------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.13844.8247.7.34.54.wu
Started at  : 10:29:47.723
Ended at    : 10:35:52.465
    364.516 secs Elapsed
     24.991 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 99.99%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.13844.8247.7.34.54.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.13844.8656.7.34.81.wu
Started at  : 10:35:55.981
Ended at    : 10:41:08.505
    312.464 secs Elapsed
     52.073 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.13844.8656.7.34.81.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.2079.16836.10.37.50.wu
Started at  : 10:41:12.113
Ended at    : 10:45:42.439
    270.274 secs Elapsed
     44.148 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 99.99%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 99.99%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.2079.16836.10.37.50.wu.res
Result      : Strongly similar,  Q= 99.99%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.2079.18881.10.37.49.wu
Started at  : 10:45:45.910
Ended at    : 10:51:01.698
    315.654 secs Elapsed
     28.439 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.2079.18881.10.37.49.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.2090.18472.12.39.249.wu
Started at  : 10:51:05.141
Ended at    : 10:56:38.273
    333.060 secs Elapsed
     44.055 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.2090.18472.12.39.249.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.2090.22562.12.39.170.wu
Started at  : 10:56:41.840
Ended at    : 11:01:14.430
    272.536 secs Elapsed
     35.927 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.2090.22562.12.39.170.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.2108.22562.13.40.187.wu
Started at  : 11:01:29.902
Ended at    : 11:06:53.038
    319.665 secs Elapsed
     47.705 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.2108.22562.13.40.187.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.30967.11928.8.35.113.wu
Started at  : 11:06:56.518
Ended at    : 11:11:59.870
    303.292 secs Elapsed
     32.573 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.30967.11928.8.35.113.wu.res
Result      : Strongly similar,  Q= 99.99%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.30986.13155.9.36.98.wu
Started at  : 11:12:03.294
Ended at    : 11:16:39.078
    275.730 secs Elapsed
     48.938 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.30986.13155.9.36.98.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
Running app : MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe -v 0
with WU     : 28oc11aa.6787.6611.5.32.85.wu
Started at  : 11:16:42.492
Ended at    : 11:21:17.168
    274.629 secs Elapsed
     45.739 secs CPU time

R2: .\ref\ref-RacerX_dev_0.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_1.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-RacerX_dev_2.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_0.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_1.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%

R2: .\ref\ref-Speed_dev_2.exe-28oc11aa.6787.6611.5.32.85.wu.res
Result      : Strongly similar,  Q= 100.0%
------------
(ignore the timings - I left BOINC running in the background, so the card was doing its day job at the same time)
ID: 2031564 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031568 - Posted: 9 Feb 2020, 12:55:23 UTC - in response to Message 2031564.  
Last modified: 9 Feb 2020, 13:00:12 UTC

Just to be clear, you compared what to what? I ask, because I provided results for 6 GPUs across 4 apps across 4 drivers... And I wasn't sure if anyone did a CPU comparison.

Did your comparison just use the 442.19 OpenCL results that I provided?
And what were you benchmarking as comparison - 442.19 OpenCL, on your end?
ID: 2031568 · Report as offensive     Reply Quote
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14655
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2031571 - Posted: 9 Feb 2020, 13:19:13 UTC - in response to Message 2031568.  

The object of the exercise was to confirm that NVidia had produced a correct, working driver at v442.19 - fixing the delays and hangups in recent history (which you have confirmed), without introducing any numerical errors along the way.

With that brief, only the driver v442.19 results are significant, so I only took those.

I compared them with a live test run on my GTX 1660 SUPER, using the Windows 7 version of driver v441.12 and the MB8_win_x86_SSE3_OpenCL_NV_SoG_r3584.exe application. I have sufficient confidence in each of those separate components to judge that it was a valid test - in effect, I was using the SoG r3584 as the reference app, and your results as the test pieces, reversing the process we went through three years ago before releasing the app.

In an ideal world, we would have used a wider range of test WUs, including WUs plucked from the wild as examples of unexpected validation failures between otherwise reliable hosts. But this is good enough for now.
ID: 2031571 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031581 - Posted: 9 Feb 2020, 13:54:01 UTC - in response to Message 2031571.  

Okay. I'm glad you weren't comparing 442.19 against 442.19! Even more glad that results are good! :)

Thanks for the validation.
ID: 2031581 · Report as offensive     Reply Quote
Alfred LK Cheng

Send message
Joined: 12 Jul 99
Posts: 2
Credit: 157,511,087
RAC: 478
Hong Kong
Message 2031602 - Posted: 9 Feb 2020, 16:33:06 UTC - in response to Message 2031571.  

Actually, we are not testing a new application. We are testing a driver which is suppose to fix the problems occurred after version 436.xx. if we compare the result of 442.19 against 436.xx, we could do a result file level comparison which should be numerically 100% match?
ID: 2031602 · Report as offensive     Reply Quote
VelocityRC
Avatar

Send message
Joined: 27 Sep 19
Posts: 23
Credit: 1,421,582
RAC: 86
United States
Message 2031621 - Posted: 9 Feb 2020, 19:23:40 UTC

Since we are on the topic of GPU's. I'm considering pulling my GTX 1050 SSC 2gb and getting a 1050 ti 4gb. Looking at my production would I see much improvement ? BTW the MoBo has a PCI-e 2.0 slot so not sure where the point of diminishing return would be with these PCI-e 3.0 cards.

Thanks.
ID: 2031621 · Report as offensive     Reply Quote
Ian&Steve C.
Avatar

Send message
Joined: 28 Sep 99
Posts: 4267
Credit: 1,282,604,591
RAC: 6,640
United States
Message 2031622 - Posted: 9 Feb 2020, 19:26:03 UTC - in response to Message 2031621.  

you'll see a small increase in production.

nothing drastic. just an increase in line with the relative performance difference between those 2 cards.

PCIe 2.0 x16 is more than enough. no worries there.
Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

ID: 2031622 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031690 - Posted: 9 Feb 2020, 23:58:37 UTC

Let's get back on topic, please.

Richard, do we need any 432.00 validation? I'm not going to redo any of my testing.
ID: 2031690 · Report as offensive     Reply Quote
Alfred LK Cheng

Send message
Joined: 12 Jul 99
Posts: 2
Credit: 157,511,087
RAC: 478
Hong Kong
Message 2031747 - Posted: 10 Feb 2020, 4:00:48 UTC - in response to Message 2031602.  

Actually, we are not testing a new application. We are testing a driver which is suppose to fix the problems occurred after version 436.xx. if we compare the result of 442.19 against 436.xx, we could do a result file level comparison which should be numerically 100% match?


The result of Richard gives us the confidence that the new driver does indeed fix the problem with the previous version. I am not suggesting further test. I am suggesting this alternative to consider in the future so that it might be easier to match the result.
ID: 2031747 · Report as offensive     Reply Quote
Jacob Klein
Volunteer tester

Send message
Joined: 15 Apr 11
Posts: 149
Credit: 9,783,406
RAC: 9
United States
Message 2031748 - Posted: 10 Feb 2020, 4:06:35 UTC - in response to Message 2031747.  
Last modified: 10 Feb 2020, 4:08:30 UTC

If you find my .zip file from a few posts back, you'll see it has results from 431.60, 431.68, 432.00, and 442.19.
Here's a link: https://setiathome.berkeley.edu/forum_thread.php?id=84694&postid=2031331
You are welcome to unzip that and do comparisons.

Offhand, I think there are slight floating point discrepancies on each run, that are usually considered "normal" and "within tolerance". But I'm no expert, especially with SETI.
ID: 2031748 · Report as offensive     Reply Quote
Previous · 1 . . . 14 · 15 · 16 · 17 · 18 · 19 · 20 · Next

Message boards : Number crunching : NVidia 436.xx and later drivers can cause very long compute times especially on Arecibo VHAR work units


 
©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.