New set of builds for Windows (r3482) available here: https://cloud.mail.ru/public/J5f8/HGuG3Vp4R

Please test if iGPU memory leak issues fixed.

Also please test how GUI responds on different GPUs with completely default options (that is, w/o any additional tuning and single task per GPU). If lags acceptable?

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Up and running.

Don't have a iGPU but not noticing any lags.

CPU usage starts 50% and slowly increases until work unit is done.

non-guppis are running about 4 min 30sec, GUPPI running almost 8 minutes

Both non-modified.

CPU time is almost identical to GPU time, varies anywhere from 9 sec to 30 sec difference.

Do you want -v 6 info or what would you like posted or linked?

This is the machine I have r3482 installed on

http://setiathome.berkeley.edu/results.php?hostid=8033686

Up and running.

Don't have a iGPU but not noticing any lags.

CPU usage starts 50% and slowly increases until work unit is done.

non-guppis are running about 4 min 30sec, GUPPI running almost 8 minutes

Both non-modified.

How this compares with r3430 ?

CPU time is almost identical to GPU time, varies anywhere from 9 sec to 30 sec difference.

add -use_sleep to see any CPU time savings.

Do you want -v 6 info or what would you like posted or linked?

Yes, single VLAR, mid-AR and VHAR outputs with -v 6 would be good to see.

http://setiathome.berkeley.edu/results.php?hostid=8033686

Thanks, will look for performance statistics.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

without -use_sleep or -v6

WU true angle range is : 203.352810
http://setiathome.berkeley.edu/result.php?resultid=5025774166

WU true angle range is : 0.740043
http://setiathome.berkeley.edu/result.php?resultid=5025858919

WU true angle range is : 0.425325
http://setiathome.berkeley.edu/result.php?resultid=5025774167

WU true angle range is : 0.008864
http://setiathome.berkeley.edu/result.php?resultid=5025858917

Now with -use_sleep and -v 6
CPU usage fluctuates. starts high at 90% and decreases down to mid 70s. Occasionally some get to 20% but 3/4 stay up in 70

-9 overflow
WU true angle range is : 0.442808
http://setiathome.berkeley.edu/result.php?resultid=5025913520


WU true angle range is : 1.265541
http://setiathome.berkeley.edu/result.php?resultid=5026007359

WU true angle range is : 1.046780
http://setiathome.berkeley.edu/result.php?resultid=5025913444
still pending validation

WU true angle range is : 0.417197
http://setiathome.berkeley.edu/result.php?resultid=5025913443

WU true angle range is : 0.011463
http://setiathome.berkeley.edu/result.php?resultid=5025956743

WU true angle range is : 0.007666
http://setiathome.berkeley.edu/result.php?resultid=5025956762

I'll keep an eye out for a high AR work unit

I can't compare r3482 to r3430 as I wasn't running single work units.

I would need to go back and downgrade to do that.

Eventually I will do that but before that, I want to try and modify the commandlines and run multiple instances so I can see how they compare to my normal processing.


edit...

The addition of -high_perf doesn't seem to do much if anything for this build. I see no change in time with it

The addition of -high_perf doesn't seem to do much if anything for this build. I see no change in time with it

Now app enables high_perf path where needed on point basis. Seems for your GPU it does it right:

Fftlength=32,pass=3:Tune: sum=34696.7(ms); min=5.576(ms); max=23.86(ms); mean=14.96(ms); s_mean=11.54; sleep=0(ms); delta=176; N=2320; usual
Fftlength=32,pass=4:Tune: sum=34696.7(ms); min=5.576(ms); max=23.86(ms); mean=14.96(ms); s_mean=11.54; sleep=0(ms); delta=176; N=2320; usual
Fftlength=32,pass=5:Tune: sum=34696.7(ms); min=5.576(ms); max=23.86(ms); mean=14.96(ms); s_mean=11.54; sleep=0(ms); delta=176; N=2320; usual
Fftlength=64,pass=3:Tune: sum=21931.8(ms); min=2.502(ms); max=17.74(ms); mean=14.61(ms); s_mean=13.87; sleep=15(ms); delta=275; N=1501; usual
Fftlength=64,pass=4:Tune: sum=21931.8(ms); min=2.502(ms); max=17.74(ms); mean=14.61(ms); s_mean=13.87; sleep=15(ms); delta=275; N=1501; usual
Fftlength=64,pass=5:Tune: sum=21931.8(ms); min=2.502(ms); max=17.74(ms); mean=14.61(ms); s_mean=13.87; sleep=15(ms); delta=275; N=1501; usual
Fftlength=128,pass=3:Tune: sum=11697.2(ms); min=1.391(ms); max=20.93(ms); mean=11.47(ms); s_mean=12.46; sleep=15(ms); delta=680; N=1020; usual
Fftlength=128,pass=4:Tune: sum=11697.2(ms); min=1.391(ms); max=20.93(ms); mean=11.47(ms); s_mean=12.46; sleep=15(ms); delta=680; N=1020; usual
Fftlength=128,pass=5:Tune: sum=11697.2(ms); min=1.391(ms); max=20.93(ms); mean=11.47(ms); s_mean=12.46; sleep=15(ms); delta=680; N=1020; usual
Fftlength=256,pass=3:Tune: sum=11239.6(ms); min=0.7013(ms); max=20.58(ms); mean=10.47(ms); s_mean=11.63; sleep=0(ms); delta=680; N=1073; usual
Fftlength=256,pass=4:Tune: sum=11239.6(ms); min=0.7013(ms); max=20.58(ms); mean=10.47(ms); s_mean=11.63; sleep=0(ms); delta=680; N=1073; usual
Fftlength=256,pass=5:Tune: sum=11239.6(ms); min=0.7013(ms); max=20.58(ms); mean=10.47(ms); s_mean=11.63; sleep=0(ms); delta=680; N=1073; usual
Fftlength=512,pass=3:Tune: sum=11053.3(ms); min=0.3725(ms); max=10.35(ms); mean=8.236(ms); s_mean=9.647; sleep=0(ms); delta=1363; N=1342; usual
Fftlength=512,pass=4:Tune: sum=11053.3(ms); min=0.3725(ms); max=10.35(ms); mean=8.236(ms); s_mean=9.647; sleep=0(ms); delta=1363; N=1342; usual
Fftlength=512,pass=5:Tune: sum=11053.3(ms); min=0.3725(ms); max=10.35(ms); mean=8.236(ms); s_mean=9.647; sleep=0(ms); delta=1363; N=1342; usual
Fftlength=1024,pass=3:Tune: sum=26917.3(ms); min=0.1883(ms); max=13.13(ms); mean=11.29(ms); s_mean=11.98; sleep=0(ms); delta=2395; N=2385; high_perf
Fftlength=1024,pass=4:Tune: sum=26917.3(ms); min=0.1883(ms); max=13.13(ms); mean=11.29(ms); s_mean=11.98; sleep=0(ms); delta=2395; N=2385; high_perf
Fftlength=1024,pass=5:Tune: sum=26917.3(ms); min=0.1883(ms); max=13.13(ms); mean=11.29(ms); s_mean=11.98; sleep=0(ms); delta=2395; N=2385; high_perf
Fftlength=2048,pass=3:Tune: sum=19290.1(ms); min=1.704(ms); max=4.652(ms); mean=4.222(ms); s_mean=4.222; sleep=0(ms); delta=1; N=4569; high_perf
Fftlength=4096,pass=3:Tune: sum=17411.2(ms); min=0.8059(ms); max=2.15(ms); mean=1.906(ms); s_mean=1.898; sleep=0(ms); delta=1; N=9137; high_perf
Fftlength=8192,pass=3:Tune: sum=21859.6(ms); min=1.033(ms); max=1.705(ms); mean=1.196(ms); s_mean=1.182; sleep=0(ms); delta=1; N=18275; usual

Also, with default sleep time of 15ms you GPU just can't provide long enough kernels to enable sleep w/o big performance loss so sleep mostly disabled.
All this tunable though but default behavior seems OK.

Well, just setup best-performance config as you did for 3430 and compare times there.
There are new options available but first step would be to see how good ap handles this new parameter space at defaults. So, same tuning line as for 3430 as start (and same number of simultaneous tasks).

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Ok, needed some more BLC to compare

For Non-guppi with old commandline in r3430, the time to complete is same. Woo Hoo!!

For GUPPI with old commandline, is about 120 seconds longer for combined work than r3430. However it is faster than r3480 so there is improvement.

Going to finish up what I have now for Main and move to Beta for test on 4bit

Temps too high outside to run during day now, so don't expect any more result until after dark.

Here is r3484 https://cloud.mail.ru/public/LqtQ/YBsLcr8nf.
Please test, especially iGPU one.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Here is new set of binaries (r3486): https://cloud.mail.ru/public/A7yG/k8yoZrFvr
Please test.
For all but iGPU only cosmetic change - no need to provide -use_sleep directly if some of options from sleep set defined.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Ok, needed some more BLC to compare

Want to try my batch file to send selected (suspended) CPU tasks to GPU queue?

---

Here is r3484 https://cloud.mail.ru/public/LqtQ/YBsLcr8nf.
Please test, especially iGPU one.

as requested in: msg 1801527

If you'd like them to be of a specific kind of vlars, I can do so since my little batch program transfers those that I have Suspended.

Both. Actually - all ARs range. But usually very VLARs cause troubles with usability.

using: NV_SoG_r3484 (no commandline)
processed: 4 vlars in sequence (1WU/GPU)
Results:

- No lag whatsoever on GTX 750 Ti with Win10

- forgot to check AR         blc3_2bit_guppi_57451_69387_HIP117559_0023.18368.0.17.26.243.vlar_2	00:26:05 (00:25:56)
- <true_angle_range>0.00823  blc3_2bit_guppi_57451_70772_HIP117779_0027.22322.831.17.26.168.vlar_1	00:25:34 (00:25:28)
- <true_angle_range>0.01140  jn10ab.24885.885.5.32.100.vlar_0	00:31:50 (00:31:45)
- <true_angle_range>0.01140  02jn10ab.24885.885.5.32.227.vlar_1	00:31:51 (00:31:45)

(before test run, non-vlar processed in ~16min +/-2)

Let me know if you'd like me to report with linked URL in the future (as Zalster did above)

As for a bigger batch (with a wider AR range), I'll wait for your modified perl script ...as I find it tedious to go digging for AR values.

Just realized that CPU usage was mostly in the 90s% as I had left 6 core running vlars ...and sometimes it reached 100% for a few secs at a time.
For future tests, I'll reduce CPU to ~80% unless specified otherwise.

[edit]Should[/e] I install SoG_r3486?

PS: I have another rig that is almost identical (except it's Win7 & 8GB ram).
It is currently setup with Lunatics v0.45 beta3 with the Cuda50 apps with 2WU/GPU.
I can also make changes to that one for testing if you'd like...or use it to compare NV_SoG to the "gold standard" for GTX 750 Ti.

Hey Boss!
Thank you for your work!
Have a nice summer :)

[quote]Here is r3484 [code]- No lag whatsoever on GTX 750 Ti with Win10

Fine!

Let me know if you'd like me to report with linked URL in the future (as Zalster did above)

yep, most representable results for midAR,VHAR,VLAR, GBT would be good to have with links to particular resuls (stderr contains additional info about ho app made choices for particular card).

As for a bigger batch (with a wider AR range), I'll wait for your modified perl script ...as I find it tedious to go digging for AR values.

soon but not today perhaps, I haven't my netbook ith me currently.

Just realized that CPU usage was mostly in the 90s% as I had left 6 core running vlars ...and sometimes it reached 100% for a few secs at a time.
For future tests, I'll reduce CPU to ~80% unless specified otherwise.

For CPU time consumption decrese try -use_sleep.

[edit]Should[/e] I install SoG_r3486?

better let it run on beta.

It is currently setup with Lunatics v0.45 beta3 with the Cuda50 apps with 2WU/GPU.
I can also make changes to that one for testing if you'd like...or use it to compare NV_SoG to the "gold standard" for GTX 750 Ti.

would be good to compare timings indeed.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Some short guidlines how to increase app performance with new params tuning (beware - GUI lags can appear in result).
1. Increase value for -tt F option (default is 15). Better to increase with step of sleep quantum (if sleep enabled). Default sleep quantum 15 (ms). If sleep not enabled no need to use such step.
2. If sleep used and CPU is Intel or AMD non-APU try to decrease sleep quantum via -sleep_quantul option (default is 15). This will make sleep more effective if real sleep quantum for particular host less than 15 ms.

I hope to write more detailed instructions with some insights in algorithm besides those options on Lunatics if will have some time for that.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

As for a bigger batch (with a wider AR range), I'll wait for your modified perl script ...as I find it tedious to go digging for AR values.

soon but not today perhaps, I haven't my netbook ith me currently.

Here it is: http://lunatics.kwsn.info/index.php?action=downloads;sa=view;down=498

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Here it is: http://lunatics.kwsn.info/index.php?action=downloads;sa=view;down=498

All I get is a few hundred identical lines of:

AR=-1ElT=0.000000Rev=0ResType=0

The file ExtractTimes_v5.7z doesn't seem to match the description "Info from client_state.xml extraction tool"

---

Here it is: http://lunatics.kwsn.info/index.php?action=downloads;sa=view;down=498

All I get is a few hundred identical lines of:

AR=-1ElT=0.000000Rev=0ResType=0

The file ExtractTimes_v5.7z doesn't seem to match the description "Info from client_state.xml extraction tool"

Times.txt content?

I checked for AVX CPU+iGPU MB task mix - it worked OK.

P.S. Also check if you have some tasks completed and still not reported (suspend network communications to accumulate such tasks). Or your client_state.xml will not contain any info to extract

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

In case anyone is interested, I made a small script to import the output into Excel:

cd /D C:\ProgramData\BOINC 
perl.exe ExtractTimes_v5.pl>null
"C:\Program Files (x86)\Microsoft Office\Office12\EXCEL.EXE" .\times.txt

Just copy those 3 lines into a file on your desktop with a filename that finishes with .cmd

@Raistmer: Here's what I am currently getting:

Task name			Result Type	Version	Parameter	 Elaps 	 CPUTime 
06jn10aj.17549.53905.6.33.198_0 	9	3484	0.374938	 1,039.1 	 1,028.3 
07jn10ab.29140.248450.3.30.124_1	9	3484	0.740539	 746.9 	 481.4 
07jn10ab.29140.248450.3.30.65_0 	9	3484	0.740539	 740.8 	 474.1 
07jn10ab.29140.248450.3.30.163_0	9	3484	0.740539	 739.8 	 474.5 
07jn10ab.29140.248450.3.30.63_0 	9	3484	0.740539	 745.9 	 479.2 
07jn10ab.29140.248450.3.30.13_0 	9	3484	0.740539	 739.9 	 474.7 
05dc10ac.29918.149181.12.39.133_1	9	3484	0.34361	 1,028.1 	 1,023.4 
05dc10ac.29918.149181.12.39.224_0	9	3484	0.34361	 987.3 	 982.0 
07jn10ab.29140.248450.3.30.69_0 	9	3484	0.740539	 745.8 	 478.9 
blc4_2bit_guppi_57451_64686_HIP117463_0009.20909.831.17.26.179.vlar_1	1	3330	0.008851	 8,724.5 	 8,654.7 

What would the perfect report look like for you?
I'm guessing you only want for: Result type = 9 (GPU?)
Do you only want for guppi? or Arecibo_VLARs also?
How many decimal points do you want for AR (Parameter), and times?
Do you want it sorted by AR?
Cheers, Rob :-)

---

Hey Raistmer!
There might be a small bug with the Perl script: ExtractTimes_v5.pl
When Boinc Client is in "Suspend network activity" as required by script above,
on 3 occasions with large batches of tasks in status: "Uploading, it was listing 1to(3?) less GPU tasks than the ones I have in status: "Uploading" ()

The last time it occured, I investigated it when 1 GPU task was missing in Times.txt.
When ordered by "Completed" times in BoincTasks' History tab, the task not in Times.txt is NOT the first one or the last one for the GPU batch.
I haven't seen the issue affecting CPU tasks.

I tested afterwards with small batches when the first task to finish is either a CPU or GPU task, but the issue didn't appear.
I will let it run another big batch to see if it happens again.
If it does, I'll stop the Boinc Client manually to see if it fixes the issue.

Let me know if there's another scenario that I didn't think of,
R :-)

---

Hey Raistmer!
There might be a small bug with the Perl script: ExtractTimes_v5.pl
...
I will let it run another big batch to see if it happens again.
If it does, I'll stop the Boinc Client manually to see if it fixes the issue

It happened again even after shutting down the Boinc Client;
1 CPU (3rd) & 1 GPU (6th) task are not showing up in Times.txt
(I rearranged the BoincTasks output so that the columns line up)

Time elapsed (CPU)	Completed		Status		Name									app
00:32:06 (00:31:56)	2016-Jul-14 8:18:22 PM	Uploading	21dc10aa.23912.8656.9.36.35.vlar_1					8.12 setiathome_v8 (opencl_nvidia_SoG)
02:27:53 (02:24:56)	2016-Jul-14 8:14:34 PM	Uploading	blc4_2bit_guppi_57451_65998_HIP117463_0013.31811.0.18.27.222.vlar_0	8.00 setiathome_v8
02:12:04 (02:09:15)	2016-Jul-14 8:02:15 PM	Uploading	blc4_2bit_guppi_57451_65342_HIP117463_0011.22853.416.18.27.24.vlar_1	8.00 setiathome_v8
02:25:28 (02:22:48)	2016-Jul-14 7:58:15 PM	Uploading	blc4_2bit_guppi_57451_65998_HIP117463_0013.31811.0.18.27.221.vlar_1	8.00 setiathome_v8
02:36:26 (02:33:32)	2016-Jul-14 7:54:15 PM	Uploading	blc4_2bit_guppi_57451_65014_HIP117463_OFF_0010.31703.831.18.27.103.vlar_1	8.00 setiathome_v8
00:31:19 (00:31:08)	2016-Jul-14 7:41:15 PM	Uploading	21dc10aa.23912.8656.9.36.17.vlar_1					8.12 setiathome_v8 (opencl_nvidia_SoG)
02:27:03 (02:24:16)	2016-Jul-14 7:35:33 PM	Uploading	blc4_2bit_guppi_57451_65670_HIP117463_OFF_0012.27821.416.18.27.169.vlar_0	8.00 setiathome_v8
00:32:04 (00:31:54)	2016-Jul-14 7:09:20 PM	Uploading	21dc10aa.23912.8656.9.36.34.vlar_0					8.12 setiathome_v8 (opencl_nvidia_SoG)
00:20:16 (00:20:03)	2016-Jul-14 6:37:03 PM	Uploading	08jn10af.16639.22971.10.37.87_1						8.12 setiathome_v8 (opencl_nvidia_SoG)

FYI, I saved a copy of my: client_state.xml & Times.txt

Also, on my rig running Cuda50, there are no GPU tasks reported in Times.txt
Don't know if that is intentional or not.

---

Hey Raistmer!
There might be a small bug with the Perl script: ExtractTimes_v5.pl
When Boinc Client is in "Suspend network activity" as required by script above,
on 3 occasions with large batches of tasks in status: "Uploading, it was listing 1to(3?) less GPU tasks than the ones I have in status: "Uploading" ()

The last time it occured, I investigated it when 1 GPU task was missing in Times.txt.
When ordered by "Completed" times in BoincTasks' History tab, the task not in Times.txt is NOT the first one or the last one for the GPU batch.
I haven't seen the issue affecting CPU tasks.

I tested afterwards with small batches when the first task to finish is either a CPU or GPU task, but the issue didn't appear.
I will let it run another big batch to see if it happens again.
If it does, I'll stop the Boinc Client manually to see if it fixes the issue.

Let me know if there's another scenario that I didn't think of,
R :-)

I never run script in BOINC's own dir. Copy xml file into another dir and run there.
Also, check if task missed in Times.txt actually presents in client_state.xml and has stored stderr content in that file.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.