Seti Performance

[fred]

A Tool to test the best multiple instances of a GPU task.
In other words the setting in <count>, like 1, 0.5, 0.33, 0.25.

This tool finds the best setting automatically.

Useful for NVIDIA  Gpu's in the 4xxx, 5xxx, 6xxx series.
Make sure you read the instructions first.
A lot of additional info can be found in config.xml

Download SETI Performance version 1.8

Compare the results with others: Gpu list

Pitfalls:

Place the files in this zip file to a temp location.
Some users use: C:\Program Files, but this may not work for you.
C:\Program Files is a restricted folder, so you need full access rights to the folder you put the files in.

In the config.xml all spaces are striped away, to use spaces use %20
To get x41z cuda 32 use x41z%20cuda%2032

Together with the exe you NEED to supply a workunit named: work_unit.sah.
This isn't needed when you check "Use all xx workunits" in that case the workunits are copied from the folder: workunits.

SetiPerformance64.exe is for 64 bit systems.
SetiPerformance.exe is for 32 and 64 bit systems.

Progress indicator:

If you need a progress indicator add the supplied file: init_data.xml
The checkpoint time is the progress indicator update frequency.

Logging:

Loggings are placed in: C:\Users\username\AppData\Roaming\eFMer\SetiPerformance\log
SetiPerformance.log : Startup info.
result.log : A copy of the logging displayed in the dialog.

Now that you ran the test:

Now that I found out that a count of 2 (0.5) is the best setting, what now?

Install this one: Lunatics installer
Make sure you have the minimum driver mentioned in the installer and choose the 32 or 64 bit version.
Next go here: C:\ProgramData\BOINC\projects\setiathome.berkeley.edu
Open app_info.xml with notepad.exe.

Look for these 2 lines together:
<type>CUDA</type>
<count>1</count>

Change the 1 in count into 0.33 or 0.5
Make sure you use a dot . after the 0! NOT a ,

Version:

V 1.4
Add log file, C:\Users\username\AppData\Roaming\eFMer\SetiPerformance\log
Keep case in name, like: MB6_win_x86_SSE3_OpenCL_ATi_HD5_r390
Added: progress indicator, needs init_data.xml. This works for MB and AP.
The window position wasn't always stored.
Some small bug fixes.

V 1.5
In the logging file line ends where \r\r\n now \r\n.
ERROR: Unable to copy from: didn't show the actual file it failed to copy, now shows: ERROR: Unable to copy file:.
Devices, Count now drop down boxes.
Added: Max count drop down box.
Fixed: Progress indicator dropped back to 0.

V 1.6
The automatic test goes one beyond the Max drop down box setting.
The drop down boxes Devices, Count and Max are now disabled, while testing.
The progress indicator didn't work correctly with AP tasks.

V1.7
Didn't automatically stop when the best result was found.

V1.8
Add Run x number of CPU tasks while running the GPU test.

[archae86]

And where are the instructions to be found?

[fred]

And where are the instructions to be found?

When you start the program, you see them.

[szopler]

Will be such app for AMD/ATI GPUs?

[fred]

Will be such app for AMD/ATI GPUs?

Nothing will hold you back.
Look in the config.xml.
You can modify or add a new <data></data> entry.

Add a new folder just below the exe as the one present.

It should work on any exe that can be run on BOINC.

[Dirk]

(...)
Download SETI Performance version 1.8
(...)

The URL don't work ...
Where I could download 'Seti Performance'?

Thanks.

[fred]

Where I could download 'Seti Performance'?

OK The link is working again.

[Dirk]

Finally I made the test with the tool:

NV GT730 (PCIe 2.0 x1 speed)

------------------------------------
Starting automatic test: (x41zc_cuda50)
21 June 2015 - 14:33:18 Start, devices: 1, device count: 1 (1.00)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 1, average time / count: 749, average time on device: 749 Seconds (12 Minutes, 29 Seconds)
Next :---------------------------------------------------------------------------
21 June 2015 - 14:45:53 Start, devices: 1, device count: 2 (0.50)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 2, average time / count: 1326, average time on device: 663 Seconds (11 Minutes, 3 Seconds)
Next :---------------------------------------------------------------------------
21 June 2015 - 15:08:03 Start, devices: 1, device count: 3 (0.33)
--------------------------------------------------------------------------- Results:
Device: 0, device count: 3, average time / count: 1967, average time on device: 655 Seconds (10 Minutes, 55 Seconds)
Next :---------------------------------------------------------------------------
21 June 2015 - 15:40:57 Start, devices: 1, device count: 4 (0.25)
21 June 2015 - 16:00:20 ERROR: Device: 0, Count: 0, finished with an error, check stderr.txt .
21 June 2015 - 16:00:20 ERROR: Device: 0, Count: 1, finished with an error, check stderr.txt .
--------------------------------------------------------------------------- Results:
Device: 0, device count: 4, average time / count: 1530, average time on device: 382 Seconds (6 Minutes, 22 Seconds)
Aborted, detected an error, results may be invalid.
------------------------------------

Does this mean that 3 (0.33) WUs/GPU is the fastest and best solution?


Then I tested with CPU WUs, but ...
------------------------------------
Starting automatic test: (x41zc_cuda50)
21 June 2015 - 16:22:12 Start, devices: 1, device count: 1 (1.00)
21 June 2015 - 16:22:12 Start, CPU count: 4
--------------------------------------------------------------------------- Results:
Device: 0, device count: 1, average time / count: 732, average time on device: 732 Seconds (12 Minutes, 12 Seconds)
Next :---------------------------------------------------------------------------
21 June 2015 - 16:34:30 Start, devices: 1, device count: 2 (0.50)
21 June 2015 - 16:34:30 Start, CPU count: 4
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
SYTEM ERROR: Please report this in: Device: 15, Device count 39
------------------------------------

(strange, with CPU WUs the GPU was/is faster?)

The progress bar was long time at ~98% and nothing happened.
So I aborted this test.

What does this mean, that this test with CPU WUs didn't worked on my system?

Intel Celeron CPU J1900 (Quad-Core, up to SSE4.2, incl. Intel iGPU) + NV GT730, Win8.1 x64.

Thanks.

[fred]

(strange, with CPU WUs the GPU was/is faster?)

The progress bar was long time at ~98% and nothing happened.
So I aborted this test.

What does this mean, that this test with CPU WUs didn't worked on my system?

Intel Celeron CPU J1900 (Quad-Core, up to SSE4.2, incl. Intel iGPU) + NV GT730, Win8.1 x64.

Thanks.

The second run the GPU crashed.
Most likely it ran too many instances.
It may be that some where still running from the previous run, It ended in an error: finished with an error, check stderr.txt
You need to complete the test without errors.

Shut down any BOINC tasks before starting the test.
Don't run a new test on errors, check if there are still tasks running.....

[Dirk]

I added the cuda42 app for to make a speed test (cuda50 vs. cuda42) on my NV GT730. And it worked.
The cuda50 app is faster.
This tool, and the one WU, is OK for to make a speed comparison test?

I wanted to add the Intel iGPU app, but it worked not.
I couldn't.

Maybe for other which have less experiences than me, and for me, maybe you could add all apps of Lunatics Installer v0.43b?

SETI@home : Message boards : Number crunching : Optimised Applications and Other Binaries - refresh: Windows Installer v0.43b

Currently I prepare a new PC build with four 'AMD Radeon Fury X' VGA cards.
If I compare with the HD7970 (3 WUs/GPU) then maybe on the 'Fury X' 6 WUs/GPU would be possible...
It would be nice if I could test it with your great tool how much simultaneously WUs give highest performance.

Thanks.

[fred]

I wanted to add the Intel iGPU app, but it worked not.

I didn't test it but adding something like this in config.xml

   <data>
      <name>iGPU</name>
      <folder>igpu</folder>
      <exe>intel.exe</exe>
      <parameter></parameter>
   </data>

Simply copy the xcpu folder and rename it igpu
copy the intel GPU exe to the folder and and make sure you have the right name for the exe above in the       <exe>intel.exe</exe>
Including the exe

I don't have the time to test it right now, otherwise post it in the Seti forum I know some who did something similar.

[Dirk]

I found the thread in NC: SetiPerformance 1.8 + Added apps from Lunatics v0.41
That's with very old apps.

BilBg's last post was at 5 Jun 2015 (~ 2 months ago).
His PCs had the last time at 2 Jun 2015 contact to the SETI server.
I guess he have vacation, or left SETI.

He added also the AP apps (and an AP WU - I guess - otherwise this wouldn't work ).
I have no idea which AP WU would be OK...

There is no hurry... - if you like and you have time - it would be super if you could add all (SETI and AP) apps of v0.43b.

I guess the new PC is finished in around one week (I need to wait that I get a new Windows OS DVD - RMA - the DVD which I got was not bootable )...

Thanks.