Optimize your GPU. Find the value the easy way.

@Fred

Small (but too many ;) ) suggestions (mostly cosmetic) for SetiPerformance:

* (bug?) The position of the 2 windows is not saved to the registry if I move the windows in their original size and exit the program.
The position and size is saved if I also resize the main/graph windows.

I even didn't know that position/size are supposed to be remembered until I found:
HKEY_CURRENT_USER\Software\eFMer\SetiPerformance\window_position

* I don't know what the graph window is supposed to show; for me it shows only faint vertical grid on black background and counting times at bottom

* It seems that <maxAutoCount>3</maxAutoCount> is zero-based
For the AP auto-test it runs up to 4 tasks

* Show maxAutoCount in the Results, e.g.
Starting automatic test: (ap6_win_x86_sse2_opencl_ati_r1316), maxAutoCount: 3

* count vs instance vs task vs ... - which is better?:
Runtime: Device: 0, count: 1, 1948 seconds
Runtime: Device: 0, instance: 1, 1948 seconds
Runtime: Device: 0, task: 1, 1948 seconds

* Checkbox: [v] Wrap text
(some lines are long)

* Save the results to file (e.g. SetiPerformanceResults.log) automatically
(while the results are created, not at the end - in case the program is terminated prematurely for any reason)

* Can we have an option to not convert everything to lowercase?
(So the strings will be shown as are in the config.xml
I prefer MB6_win_x86_SSE3_OpenCL_ATi_HD5_r390 instead of mb6_win_x86_sse3_opencl_ati_hd5_r390)

* Checkbox: [v] Show progress
Can be done by examining the files e.g. every minute
<prog>0.xxxxx</prog> in state.sah
<fraction_done>0.yyyyy</fraction_done> in pulse.out*

(probably can also be done by shared memory but I don't know the details)

If no file still exist (at task startup) show ??% or --% or ...

The problem is how to show many running tasks.
Most people will run max 4 tasks on the single GPU they have.
But seeing how you create 256 slots (slot0 ... slot255) you are prepared for some big monster ;)


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

About ten days of crunching 3 wu per GPU. One Gtx 470 and one gtx 670.
6 gpu tasks and 6 cpu tasks now running (started cpu tasks about 4-5 days ago)

I7-920 cpu Hyperthread enable (only overclocked to 3.6ghz) GPU's not overclocked
due to different versions.

started with rac close to zero now at 23k and climbing. Will wait and see what the 60 day avg comes out 2.
Have been able to keep the system fed with wu's so far.

thanks for all the info.

Keep on crunching !!

____________

Ran the old test, worked fine....now with a new video card to test I get this:
ERROR: Unable to copy from: C:\Program Files\BOINC\Freds test software\x41g\ to: C:\Users\Jim\AppData\Roaming\eFMer\SetiPerformance, slot: 0
Aborted

wtf chuck?
____________

Try to put it NOT in C:\Program Files\


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

Using: 306.23
GPU0: 580 @833/1004/1666
GPU1: 580 @815/1004/1630
GPU2: 570 @810/975/1620

Starting automatic test: (x41g) Workunit: PG0395_v7.wu
15 September 2012 - 10:49:41 Start, devices: 3, device count: 1 (1.00) (Wu: PG0395_v7.wu)
--------------------------------------------------------------------------- Results: (PG0395_v7.wu)
Device: 0, device count: 1, average time / count: 32, average time on device: 32 Seconds (0 Minutes, 32 Seconds)
Device: 1, device count: 1, average time / count: 33, average time on device: 33 Seconds (0 Minutes, 33 Seconds)
Device: 2, device count: 1, average time / count: 36, average time on device: 36 Seconds (0 Minutes, 36 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:50:19 Start, devices: 3, device count: 2 (0.50) (Wu: PG0395_v7.wu)
--------------------------------------------------------------------------- Results: (PG0395_v7.wu)
Device: 0, device count: 2, average time / count: 50, average time on device: 25 Seconds (0 Minutes, 25 Seconds)
Device: 1, device count: 2, average time / count: 50, average time on device: 25 Seconds (0 Minutes, 25 Seconds)
Device: 2, device count: 2, average time / count: 58, average time on device: 29 Seconds (0 Minutes, 29 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:51:20 Start, devices: 3, device count: 3 (0.33) (Wu: PG0395_v7.wu)
--------------------------------------------------------------------------- Results: (PG0395_v7.wu)
Device: 0, device count: 3, average time / count: 74, average time on device: 24 Seconds (0 Minutes, 24 Seconds)
Device: 1, device count: 3, average time / count: 72, average time on device: 24 Seconds (0 Minutes, 24 Seconds)
Device: 2, device count: 3, average time / count: 84, average time on device: 28 Seconds (0 Minutes, 28 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:52:46 Start, devices: 3, device count: 4 (0.25) (Wu: PG0395_v7.wu)
--------------------------------------------------------------------------- Results: (PG0395_v7.wu)
Device: 0, device count: 4, average time / count: 96, average time on device: 24 Seconds (0 Minutes, 24 Seconds)
Device: 1, device count: 4, average time / count: 95, average time on device: 23 Seconds (0 Minutes, 23 Seconds)
Device: 2, device count: 4, average time / count: 110, average time on device: 27 Seconds (0 Minutes, 27 Seconds)
>> The best average time found: 25 Seconds (0 Minutes, 25 Seconds), with count: 0.33 (3)
--------------------------------------------------------------------------- Next
15 September 2012 - 10:54:41 Start, devices: 3, device count: 1 (1.00) (Wu: PG0444_v7.wu)
--------------------------------------------------------------------------- Results: (PG0444_v7.wu)
Device: 0, device count: 1, average time / count: 31, average time on device: 31 Seconds (0 Minutes, 31 Seconds)
Device: 1, device count: 1, average time / count: 32, average time on device: 32 Seconds (0 Minutes, 32 Seconds)
Device: 2, device count: 1, average time / count: 35, average time on device: 35 Seconds (0 Minutes, 35 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:55:19 Start, devices: 3, device count: 2 (0.50) (Wu: PG0444_v7.wu)
--------------------------------------------------------------------------- Results: (PG0444_v7.wu)
Device: 0, device count: 2, average time / count: 49, average time on device: 24 Seconds (0 Minutes, 24 Seconds)
Device: 1, device count: 2, average time / count: 49, average time on device: 24 Seconds (0 Minutes, 24 Seconds)
Device: 2, device count: 2, average time / count: 56, average time on device: 28 Seconds (0 Minutes, 28 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:56:18 Start, devices: 3, device count: 3 (0.33) (Wu: PG0444_v7.wu)
--------------------------------------------------------------------------- Results: (PG0444_v7.wu)
Device: 0, device count: 3, average time / count: 70, average time on device: 23 Seconds (0 Minutes, 23 Seconds)
Device: 1, device count: 3, average time / count: 71, average time on device: 23 Seconds (0 Minutes, 23 Seconds)
Device: 2, device count: 3, average time / count: 80, average time on device: 26 Seconds (0 Minutes, 26 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 10:57:40 Start, devices: 3, device count: 4 (0.25) (Wu: PG0444_v7.wu)
--------------------------------------------------------------------------- Results: (PG0444_v7.wu)
Device: 0, device count: 4, average time / count: 94, average time on device: 23 Seconds (0 Minutes, 23 Seconds)
Device: 1, device count: 4, average time / count: 93, average time on device: 23 Seconds (0 Minutes, 23 Seconds)
Device: 2, device count: 4, average time / count: 106, average time on device: 26 Seconds (0 Minutes, 26 Seconds)
>> The best average time found: 24 Seconds (0 Minutes, 24 Seconds), with count: 0.33 (3)
--------------------------------------------------------------------------- Next
15 September 2012 - 10:59:29 Start, devices: 3, device count: 1 (1.00) (Wu: PG1327_v7.wu)
--------------------------------------------------------------------------- Results: (PG1327_v7.wu)
Device: 0, device count: 1, average time / count: 41, average time on device: 41 Seconds (0 Minutes, 41 Seconds)
Device: 1, device count: 1, average time / count: 42, average time on device: 42 Seconds (0 Minutes, 42 Seconds)
Device: 2, device count: 1, average time / count: 45, average time on device: 45 Seconds (0 Minutes, 45 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 11:00:17 Start, devices: 3, device count: 2 (0.50) (Wu: PG1327_v7.wu)
--------------------------------------------------------------------------- Results: (PG1327_v7.wu)
Device: 0, device count: 2, average time / count: 56, average time on device: 28 Seconds (0 Minutes, 28 Seconds)
Device: 1, device count: 2, average time / count: 56, average time on device: 28 Seconds (0 Minutes, 28 Seconds)
Device: 2, device count: 2, average time / count: 65, average time on device: 32 Seconds (0 Minutes, 32 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 11:01:23 Start, devices: 3, device count: 3 (0.33) (Wu: PG1327_v7.wu)
--------------------------------------------------------------------------- Results: (PG1327_v7.wu)
Device: 0, device count: 3, average time / count: 82, average time on device: 27 Seconds (0 Minutes, 27 Seconds)
Device: 1, device count: 3, average time / count: 81, average time on device: 27 Seconds (0 Minutes, 27 Seconds)
Device: 2, device count: 3, average time / count: 95, average time on device: 31 Seconds (0 Minutes, 31 Seconds)
Next :---------------------------------------------------------------------------
15 September 2012 - 11:03:02 Start, devices: 3, device count: 4 (0.25) (Wu: PG1327_v7.wu)
--------------------------------------------------------------------------- Results: (PG1327_v7.wu)
Device: 0, device count: 4, average time / count: 109, average time on device: 27 Seconds (0 Minutes, 27 Seconds)
Device: 1, device count: 4, average time / count: 108, average time on device: 27 Seconds (0 Minutes, 27 Seconds)
Device: 2, device count: 4, average time / count: 125, average time on device: 31 Seconds (0 Minutes, 31 Seconds)
>> The best average time found: 28 Seconds (0 Minutes, 28 Seconds), with count: 0.33 (3)

____________

Ran the old test, worked fine....now with a new video card to test I get this:
ERROR: Unable to copy from: C:\Program Files\BOINC\Freds test software\x41g\ to: C:\Users\Jim\AppData\Roaming\eFMer\SetiPerformance, slot: 0
Aborted

wtf chuck?

It's the spaces that are killing you

1) HKEY_CURRENT_USER\Software\eFMer\SetiPerformance\window_position

2) I don't know what the graph window is supposed to show; for me it shows only faint vertical grid on black background and counting times at bottom

3) It seems that <maxAutoCount>3</maxAutoCount> is zero-based

4) Checkbox: [v] Wrap text(some lines are long)

5) Save the results to file (e.g. SetiPerformanceResults.log)

6) Can we have an option to not convert everything to lowercase?

7) Checkbox: [v] Show progress

8) The problem is how to show many running tasks.

1) Positions should be stored almost at once.
But on exit it's best to use use Start->Shutdown.
Pressing the power button may cause Windows to force a quick power down.
And this means some parts of the registry aren't stored.

2) The graph starts at the beginning of the text.
Or you can choose "Run continuously"
In the xml file you can adjust the colors, use the Read xml button for instant results.

3)Is intended for some cards that run out of memory, as you set the value, you should be able to remember it. Or add it to the task name

8)The number of slots is unlimited, it depends on the devices / count.

* Checkbox: [v] Show progress
Can be done by examining the files e.g. every minute
<prog>0.xxxxx</prog> in state.sah
<fraction_done>0.yyyyy</fraction_done> in pulse.out*

Doesn't write anything. stderr.txt and result.sah are not updated.
Is there a parameter to force check pointing?

* Checkbox: [v] Show progress
Can be done by examining the files e.g. every minute
<prog>0.xxxxx</prog> in state.sah
<fraction_done>0.yyyyy</fraction_done> in pulse.out*

Doesn't write anything. stderr.txt and result.sah are not updated.
Is there a parameter to force check pointing?

New package (18 MB):
All from original SetiPerformance_32_64_1_4.zip
+ 4 OpenCL tests: ATI (MB and AP) and NVIDIA AP

SetiPerformance_32_64_1_4__OpenCL_ATI_NV_Added-by-BilBg.7z
http://www.datafilehost.com/download-45c22d0e.html


OpenCL tests added:

MB6_win_x86_SSE3_OpenCL_ATi_HD5_r390
MB6_win_x86_SSE3_OpenCL_ATi_r390 (for ATI HD4xxx or to compare the speed difference to HD5_r390 app on ATI HD5xxx+ GPU)

AP6_win_x86_SSE2_OpenCL_ATI_r1316
AP6_win_x86_SSE2_OpenCL_NV_r1316 (Not tested - I don't have NVIDIA)










____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)

To Fred:

I understand that this is an 'one-time use' utility and maybe you don't want to invest much time in it
so feel free to ignore any of the following:

1) The progress indicator for AP is not updating correctly:
A few minutes after task start progress indicator is shown (at visually ~20%)
but after a few seconds it drops to 0% (empty bar) and stays this way to the end.

The files to monitor are 3:
pulse.out0
pulse.out1
pulse.out

They update in order.
Either check the last-modified file or check all 3 and show the max <fraction_done>


2) The log files (SetiPerformance.log & result.log) use strange sequence at line ends (line break):
0D 0D 0A (CR CR LF) instead of the usual for Windows 0D 0A

This makes the files difficult to edit - after Save the layout is broken.


3) <maxAutoCount>3</maxAutoCount> runs up to 4 tasks
I can test this only for the ATI AP auto-test (ATI MB error for me on 3 tasks at which time I abort - too much lag)
(I added NVIDIA AP to the package in my previous post, can you test on some of your NVIDIA? I don't know if this happens on x41g test)

My preference/suggestion will be to add drop-down list to the interface to choose maxAutoCount
(and use <maxAutoCount>X</maxAutoCount> from config.xml only as the max value that can be chosen by this drop-down list.
Probably you'll also save the last used value in the registry.
)

4) Process detection: is it possible (not too much work) to detect which program started an GPU app process?
If I forget to exit BOINC and start SetiPerformance
or if at the end of test I start BOINC and then try to exit SetiPerformance
- there is warning/ask to kill the app processes which belong to BOINC

It is in fact good to have some warning that BOINC GPU apps run (e.g. "Exit BOINC before running the tests")
but "kill the app" dialog (again: if not too much work) better not kill processes which belong to BOINC (child processes of boinc.exe)


____________



- ALF - "Find out what you don't do well ..... then don't do it!" :)