2nd task also completed and validated

http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24277297

Looks like you're onto a winner with the Intel GPU app - at least as far as my HD 530 goes. Fingers crossed it will work on Tut's iGPU too :-)

In anticipation of you getting the new builds to Eric in time to deploy them over the weekend I will now revert to stock apps :-)

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.

Edit...

Sorry, not a lot of sleep last night, I'll post result of those changes from last night.

SETI@Home Informational message -9 result_overflow
WU true angle range is : 0.429536
https://setiweb.ssl.berkeley.edu/beta//result.php?resultid=24286344


WU true angle range is : 0.438143
https://setiweb.ssl.berkeley.edu/beta//result.php?resultid=24285761


WU true angle range is : 0.008604
https://setiweb.ssl.berkeley.edu/beta//result.php?resultid=24286174

Ok, looking at r3484 vs r3430..

running 3 work units at a time,

blc guppi are running exactly the same amount of time

non-guppi are about 60 to 120 sec longer.

What does that mean? Guppi and nonguppi take exactly the same amount of time, where in r3430 a group of non guppi were faster by 2-3 minutes

Ok, looking at r3484 vs r3430..

running 3 work units at a time,

blc guppi are running exactly the same amount of time

non-guppi are about 60 to 120 sec longer.

What does that mean? Guppi and nonguppi take exactly the same amount of time, where in r3430 a group of non guppi were faster by 2-3 minutes

Seems to be some sleep going on there, even if you haven't used the sleep command.

From: http://setiweb.ssl.berkeley.edu/beta/result.php?resultid=24288206

Fftlength=8,pass=3:Tune: sum=4163(ms); min=3.54(ms); max=347.3(ms); mean=181(ms); s_mean=154.6; sleep=150(ms); delta=2190; N=23; usual
Fftlength=8,pass=4:Tune: sum=2940.5(ms); min=3.7(ms); max=250.3(ms); mean=127.8(ms); s_mean=108.1; sleep=105(ms); delta=1936; N=23; usual
Fftlength=8,pass=5:Tune: sum=2244.76(ms); min=3.986(ms); max=211.4(ms); mean=97.6(ms); s_mean=81.45; sleep=75(ms); delta=1681; N=23; usual
Fftlength=16,pass=3:Tune: sum=2581.58(ms); min=1.572(ms); max=101.9(ms); mean=52.69(ms); s_mean=47.78; sleep=45(ms); delta=887; N=49; usual
Fftlength=16,pass=4:Tune: sum=1758.8(ms); min=1.725(ms); max=71.97(ms); mean=37.42(ms); s_mean=34.64; sleep=30(ms); delta=776; N=47; usual
Fftlength=16,pass=5:Tune: sum=1310.54(ms); min=1.853(ms); max= 53(ms); mean=27.88(ms); s_mean=25.32; sleep=15(ms); delta=648; N=47; usual
Fftlength=32,pass=3:Tune: sum=1722.38(ms); min=0.7394(ms); max=35.08(ms); mean=17.76(ms); s_mean=16.43; sleep=15(ms); delta=404; N=97; usual
Fftlength=32,pass=4:Tune: sum=1251.22(ms); min=0.7958(ms); max=40.11(ms); mean=15.07(ms); s_mean=14.91; sleep=15(ms); delta=572; N=83; usual
Fftlength=32,pass=5:Tune: sum=912.687(ms); min=0.8235(ms); max= 20(ms); mean=14.49(ms); s_mean=11.43; sleep=0(ms); delta=500; N=63; usual
Fftlength=64,pass=3:Tune: sum=1607.41(ms); min=0.3656(ms); max=30.56(ms); mean=14.88(ms); s_mean=9.149; sleep=0(ms); delta=317; N=108; usual
Fftlength=64,pass=4:Tune: sum=1189.34(ms); min=10.27(ms); max=66.98(ms); mean=12.14(ms); s_mean=11.52; sleep=0(ms); delta=448; N=98; usual
Fftlength=64,pass=5:Tune: sum=829.733(ms); min=7.473(ms); max=21.34(ms); mean=8.467(ms); s_mean=8.168; sleep=0(ms); delta=448; N=98; usual
Fftlength=128,pass=3:Tune: sum=3589.43(ms); min=14.29(ms); max=27.34(ms); mean=18.22(ms); s_mean=18.06; sleep=15(ms); delta=1; N=197; usual
Fftlength=256,pass=3:Tune: sum=3848.17(ms); min=8.162(ms); max=67.33(ms); mean=9.792(ms); s_mean=12.68; sleep=15(ms); delta=1; N=393; usual
Fftlength=512,pass=3:Tune: sum=4172.92(ms); min=4.044(ms); max=133.4(ms); mean=5.302(ms); s_mean=6.138; sleep=0(ms); delta=1; N=787; usual
Fftlength=1024,pass=3:Tune: sum=1836.83(ms); min=1.018(ms); max=1.439(ms); mean=1.168(ms); s_mean=1.144; sleep=0(ms); delta=1; N=1573; usual
Fftlength=2048,pass=3:Tune: sum=1178.94(ms); min=0.336(ms); max=7.189(ms); mean=0.3746(ms); s_mean=0.3797; sleep=0(ms); delta=1; N=3147; usual


To be honest, it seems as if higher versions than r3430, gets slower and slower, at least for high end GPU's. I'll stay with r3430 on main for now.

---

WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.

hmm...I wonder if the -high_perf is causing that?

I'll remove that and see if it changes

hmm...I wonder if the -high_perf is causing that?

I'll remove that and see if it changes

I don't think so. The answer is probably in this somewhat cryptic sentence that Raistmer wrote in a message above:

"For all but iGPU only cosmetic change - no need to provide -use_sleep directly if some of options from sleep set defined."

I haven't got a clue, what he means....

Does it mean that sleep is always on, or what???

And I think also that -high_perf, is automatic with the newer versions. I think Raistmer wrote something about that somewhere.

Edit: This is becoming too complicated for this old man. I give up for now, and stay with what works for me (r3430). I can live with the autocorr bug check errors. They don't occur very often.

---

WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.

I understand, yes I only run r3430 on my Seti Main machines.

Since this is a new build machine, I can add and subtract from it without too much effort.

Raistmer, can you give us some idea if a sleep command is now on by default? What about high pref as well?

Well, since I removed the -high_pref the non_guppi speed up.. kind of weird right?

Looks like r3484 is now comparable to r3430.

Tut, what are your times like? Are they the same as r3430?

Task Postponed: Can't read CL file on opencl_atiapu_sah for version v8 8.15.

No CL delivered on initial download?

Task Postponed: Can't read CL file on opencl_atiapu_sah for version v8 8.15.

Here we go again.

No CL delivered on initial download?

It would appear so.

---

Grant
Darwin NT.

Ok, I took a chance and downloaded Raistmer new app and move the Multibeam_Kernals_r3480.cl to the beta folder and restarted Boinc.

It looks to have worked.

Will have to wait and see if they validate

here's the link to his mail.ru to download, you will need something to expand the zip file

https://cloud.mail.ru/public/4g4V/Gt278t2ku

From Zalster the last time this happened.

I'm running NVidia hardware, and getting the
Multibeam_Kernals_r3480.cl
file from Raistmer's site & putting it in the
C:\ProgramData\BOINC\projects\setiweb.ssl.berkeley.edu_beta
got things running again last time around, although it looks like we're up to r_3486 now.


If you use the ATi_APU file and get a copy of the (probably labeled) ATi_Kernals.cl file and put it in the project directory things might work for you.

Or you could wait for someone that knows what they're talking about.
:-)

---

Grant
Darwin NT.

Task Postponed: Can't read CL file on opencl_atiapu_sah for version v8 8.15.

Found it. https://cloud.mail.ru/public/A7yG/k8yoZrFvr

Has the 3486 CL file.

OK, got 8.15 for both the 980 and the iGPU. I will run without any command line options for a while, to see how total stock settings looks. 3 on the 980, and 1 on the iGPU.

So far it seems at least as if the memory leak on the iGPU is solved with 8.15.

Thanks for the morning coffee :-)

Edit: Unfortunately though, we seems to be in the beginning of another Arecibo VLAR storm, just at the start of a new version run. That's no good, it will really make it hard to compare anything.

---

WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.

Well, since I removed the -high_pref the non_guppi speed up.. kind of weird right?

Looks like r3484 is now comparable to r3430.

Tut, what are your times like? Are they the same as r3430?

I haven't got a clue Zalster, since I now run with pure stock settings, apart from running 3 at a time. I'll keep it like that for a while.

---

WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.

Edit: Unfortunately though, we seems to be in the beginning of another Arecibo VLAR storm,

There goes the neighbourhood again.

---

Grant
Darwin NT.

1. Sleep should be disabled by default for all bt low-perf path
2. "Sleep" listed in tune statistics is just how much it would sleep in case sleep would be enabled. This doesn't mean real sleep.
3. -high_perf (from performance), not -high_pref.
If you see difference with -high_pref on/off most probably you see natural deviation between tasks.
4. anyone else see issues with CL file for ATi?
5. @Zalster - tasks you posted are -v 6 ones.
Please do few runs with -sleep_quantum 3 for midAR VLAr, VHAR (w/o -v 6)

---

News about SETI opt app releases: https://twitter.com/Raistmer

I added my laptop to beta for the iGPU only. I noticed it restarts a lot on guppi but not on arecibo.

guppi

arecibo

I'm only adding the -cpu_lock switches for now. No other tunings, until this Shi* storm of Arecibo VLARs, is totally over.

So, for the GTX 980, the commandline for both opencl_nvidia_sah, and opencl_nvidia_SoG, will simply be:

-cpu_lock -high_perf -instances_per_device 3 -total_GPU_instances_num 4

And for the iGPU, it will be:

-cpu_lock -instances_per_device 1 -total_GPU_instances_num 4

Thanks for all Arecibo VLARs, but can we have something else soon?

---

WARNING!! "THIS IS A SIGNATURE", of the "IT MAY CHANGE AT ANY MOMENT" type. It may, or may not be considered insulting, all depending upon HOW SENSITIVE THE VIEWER IS, to certain inputs to/from the nervous system.

I added my laptop to beta for the iGPU only. I noticed it restarts a lot on guppi but not on arecibo.

guppi

arecibo

Hi John

It looks as though your guppi task ran with the v8.14 app - the guppi restarts are a "feature" of this version. v8.15 resolves the issue :-)