Hello !

RE : SDOERR_ntl9evsss3-0-1-RND1620_1

Are these WUs special ? It seems they don't use the GPU's as efficiently as the other apps : lower GPU use, take longer to complete, less credit, ...

Thank You

Best Regards,

Phil1966

Can we expect an answer from scientists to whom we offer computing time? Thank you.

On a GTX 680:

Projected run time: 8.3 hours
----------------------------------------
Suggested app: long
Suggested LONG credit: 80600

So credits and runtime are according to plan. GPU utilization should be same more or less as other WUs.

Example on GTX 980:
http://www.gpugrid.net/result.php?resultid=13431508
5.2 hours runtime, 120900 credits.

You are welcome.

Dear Stefan,

Thank you for your answer.

The runtime you quote for a GTX980 is an exception. Another 980 took > 8 hours to complete tasks.(see below)

I am not running GPUGRID for "credits only", but as there is usually a "direct correlation" between runtime and credits, I am astonished to see the ratio for these WU's. (lower credit / longer runtime) => I thought there was maybe a problem or something special with these WUs.

By the way, the 3 * GTX970 (2 * G1 Gibabyte + 1 * PNY) I am currently dedicating to GPUGRID are 7° up to 10 ° cooler with these apps than with the other GPUGRID apps. But have not measured precisely GPU use.

Don't get me wrong, this is not a complaint but an open question, as I like this project very much.

Rank User name WU id Timestamp (h) GPU description
1 petebe 10338739 5.25 [3] NVIDIA GeForce GTX 980 (4095MB) driver: 344.11
2 captainjack 10391665 6.65 [2] NVIDIA GeForce GTX 970 (4095MB) driver: 343.22
3 s0m3wh4t 10388100 6.66 [2] NVIDIA GeForce GTX 780 Ti (3072MB) driver: 344.65
4 Alex 10395793 6.87 NVIDIA GeForce GTX 980 (4095MB) driver: 344.75
5 Betting Slip 10405882 6.93 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
6 JugNut 10400682 7.08 [2] NVIDIA GeForce GTX 970 (4095MB) driver: 344.48
7 Norman 10395785 7.15 NVIDIA GeForce GTX 780 Ti (3072MB) driver: 344.75
8 YannTC 10400672 7.23 NVIDIA GeForce GTX 980 (4095MB) driver: 344.11
9 [AF>Amis des Lapins] Phil1966 10400677 7.25 [2] NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
10 Astiesan 10383229 7.29 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
11 Ian Laksmana 10405880 7.33 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
12 tcrbiker 10400664 7.42 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
13 Masta-Killa 10395792 8.02 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
14 Dave GPU 10383226 8.05 [2] NVIDIA GeForce GTX 770 (2048MB) driver: 344.75
15 ERBrouwer 10383231 8.11 NVIDIA GeForce GTX 970 (4095MB) driver: 344.75
16 BruceR 10395791 8.12 [2] NVIDIA GeForce GTX 780 (3072MB) driver: 344.60
17 RaymondFO* 10391662 8.31 [2] NVIDIA GeForce GTX 980 (133656575MB)
18 Sebastian M. Bobrecki 10405885 8.32 NVIDIA GeForce GTX 680 (2047MB)
19 Viktor Svantner 10405887 8.38 [2] NVIDIA GeForce GTX TITAN Black (4095MB) driver: 344.75
20 Dennisd 10391658 8.41 NVIDIA GeForce GTX 780 (3072MB) driver: 344.65

Thank You

Kind Regards

Philippe

Dear Philippe,

You should use the "pre" tag instead of "quote" to make spreadsheets more readable, and use white spaces instead of tabulators.
Your team name (sorry about that), and the redundant "NVIDIA GeForce" text also had to go for the same reason.

                                    Time
                                    stamp
Rank User name               WU id   (h)            GPU description
   1 petebe                10338739  5.25  [3] GTX 980    (4095MB) driver: 344.11
   2 captainjack           10391665  6.65  [2] GTX 970    (4095MB) driver: 343.22
   3 s0m3wh4t              10388100  6.66  [2] GTX 780 Ti (3072MB) driver: 344.65
   4 Alex                  10395793  6.87      GTX 980    (4095MB) driver: 344.75
   5 Betting Slip          10405882  6.93      GTX 970    (4095MB) driver: 344.75
   6 JugNut                10400682  7.08  [2] GTX 970    (4095MB) driver: 344.48
   7 Norman                10395785  7.15      GTX 780 Ti (3072MB) driver: 344.75
   8 YannTC                10400672  7.23      GTX 980    (4095MB) driver: 344.11
   9 Phil1966              10400677  7.25  [2] GTX 970    (4095MB) driver: 344.75
  10 Astiesan              10383229  7.29      GTX 970    (4095MB) driver: 344.75
  11 Ian Laksmana          10405880  7.33      GTX 970    (4095MB) driver: 344.75
  12 tcrbiker              10400664  7.42      GTX 970    (4095MB) driver: 344.75
  13 Masta-Killa           10395792  8.02      GTX 970    (4095MB) driver: 344.75
  14 Dave GPU              10383226  8.05  [2] GTX 770    (2048MB) driver: 344.75
  15 ERBrouwer             10383231  8.11      GTX 970    (4095MB) driver: 344.75
  16 BruceR                10395791  8.12  [2] GTX 780    (3072MB) driver: 344.60
  17 RaymondFO*            10391662  8.31  [2] GTX 980    (133656575MB)
  18 Sebastian M. Bobrecki 10405885  8.32      GTX 680    (2047MB)
  19 Viktor Svantner       10405887  8.38  [2] GTX TITAN Black (4095MB) driver: 344.75
  20 Dennisd               10391658  8.41      GTX 780    (3072MB) driver: 344.65

About your original problem:
As you can see on the chart above, Petebe's host is much faster than any other host on this list. The reason for that this host is running Windows XP x64, which doesn't have the "Windows Display Driver Model" architecture (it's introduced in Windows Vista). This new WDDM is much safer - as a display driver crash won't let the PC to have a BSOD - in exchange it has an overhead on every GPU operation making it slower. This method taxes different workunits to different extent, as the frequency of the interaction between the CPU and the GPU is different for every type of workunit batches. This workunit batch have a higher frequency of CPU-GPU interaction, so it will run slower on Windows Vista (7, 8, 8.1 and later). That's why your GPU is colder while crunching these workunits compared to other batches.

Hello Zoltan,

Thank you for your message and your explanations.

The reference to Petebe's host was an answer to Stefan's post who took it as a kind of reference for GTX980 / runtime.

I don't insist further :)

Thank You

Best

Philippe

Yes sorry, I didn't know petebe was such a star :P I just chose the first completed WU I saw in the list, hehe.
As for the higher CPU-GPU communication on these WUs I cannot really tell why that happens. As far as I know it should not be any different because I don't have any explicit CPU calculations running, like boundary checks that we sometimes use for other systems.
So if there is a difference it has probably to do with the implementation of ACEMD which I unfortunately don't know much about.