GBT ('guppi') .vlar tasks will be send to GPUs, what you think about this?

Message boards : Number crunching : GBT ('guppi') .vlar tasks will be send to GPUs, what you think about this?
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 10 · Next

AuthorMessage
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1786639 - Posted: 11 May 2016, 9:11:58 UTC - in response to Message 1786635.  
Last modified: 11 May 2016, 9:12:59 UTC

Sounds about right Petri. Gaussian Search requires a drift over a spot on the sky. i.e. pass the telescope beam over a specific direction and get a bell shaped signal from a constant source. VLAR targeted searches (unblinking eye at one spot) don't pass over a spot, but travel with it.


Yes, I know what a gaussian is and how it is formed in to the signal.

I have noticed the same negative value in test wu PG009.

I wonder if ths is a bug in cuda code base or in all sw versions. See edit in my other message.


The way I understand it, and bear in mind Joe Segur was my goto guru with this stuff, The angle range is kindof an average of the telescope possibly moving all over the place during the observation period. Sometimes he presented coordinates as though the person driving the receiver was playing space invaders. In this sense IMO, if you are looking at erratic behaviour under some conditions, you are definitely looking at noise floor and non-determinism as being factored in.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1786639 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1786648 - Posted: 11 May 2016, 9:49:31 UTC - in response to Message 1786629.  
Last modified: 11 May 2016, 9:51:55 UTC


I suspect they are not doing gaussian searching: Sigma > GaussTOffsetStop: 498 > -434

And it's just correct. Gaussian search doesn't performed on VLAR at all.
What bug and what fix then?

EDIT: w/o boundary checking your change will cause out of range error and result in exception for protected memory range.
ID: 1786648 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1786650 - Posted: 11 May 2016, 9:59:52 UTC - in response to Message 1786648.  


I suspect they are not doing gaussian searching: Sigma > GaussTOffsetStop: 498 > -434

And it's just correct. Gaussian search doesn't performed on VLAR at all.
What bug and what fix then?

EDIT: w/o boundary checking your change will cause out of range error and result in exception for protected memory range.


Yes, I know. I thought a moment and found out that if the angle rate is low enough there will be no gaussian, since the telescope movement + earth rotation is not enough in a given time window to form a gaussian.

So no error.
To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1786650 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 1786654 - Posted: 11 May 2016, 10:23:17 UTC
Last modified: 11 May 2016, 10:24:42 UTC

So far I've had only one guppi .vlar in SETI Beta on my Linux box with its AMD HD 7770. It took a far shorter time than those crunched on the faster Windows 10 PC with its GTX 750 Ti OC. But you cannot make a statistic out of a single case.
Tullio
ID: 1786654 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34381
Credit: 79,922,639
RAC: 80
Germany
Message 1786683 - Posted: 11 May 2016, 13:37:31 UTC
Last modified: 11 May 2016, 13:39:56 UTC

Have finnished my first guppi tasks here at main.

Approx 30 minutes each running 2 instances on my R9 380.

Example

http://setiathome.berkeley.edu/result.php?resultid=4923313255


With each crime and every kindness we birth our future.
ID: 1786683 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14679
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1786685 - Posted: 11 May 2016, 13:44:36 UTC - in response to Message 1786683.  
Last modified: 11 May 2016, 13:50:50 UTC

Have finnished my first guppi tasks here at main.

Approx 30 minutes each running 2 instances on my R9 380.

Example

http://setiathome.berkeley.edu/result.php?resultid=4923313255

About 1750 seconds for the guppi VLAR, against 950 seconds for mid-AR Arecibo work. Rather less than double - is that a fair summary?

Edit - by comparison, my GTX 670 - also 2 instances - is taking about 3900 seconds for guppi VLAR, 1050 for Arecibo mid-AR. A rather wider ratio.
ID: 1786685 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34381
Credit: 79,922,639
RAC: 80
Germany
Message 1786690 - Posted: 11 May 2016, 13:49:46 UTC - in response to Message 1786685.  
Last modified: 11 May 2016, 13:50:44 UTC

Have finnished my first guppi tasks here at main.

Approx 30 minutes each running 2 instances on my R9 380.

Example

http://setiathome.berkeley.edu/result.php?resultid=4923313255

About 1750 seconds for the guppi VLAR, against 950 seconds for mid-AR Arecibo work. Rather less than double - is that a fair summary?


I have finnished 10 guppis so far.
None of them took longer.

Of course i use highly optimized params.

Not a sign of any lags with period_iterations_num 40.


With each crime and every kindness we birth our future.
ID: 1786690 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1786703 - Posted: 11 May 2016, 14:14:18 UTC

So just checking some of the GUPPI from yesterday for validation.

I'm noticing a large swing in credit from 199 down to 29.

When I look at the low credit work, it looks like I'm always paired with an ATI user

They take about 1000 sec longer to process but the value is at least 50-100 credits lower when I'm paired with a non-ATI user.

Can that be right? I didn't think that was supposed to do that.

I hope I am looking at this wrong, lack of sleep and such...
ID: 1786703 · Report as offensive
Profile Mr. Kevvy Crowdfunding Project Donor*Special Project $250 donor
Volunteer moderator
Volunteer tester
Avatar

Send message
Joined: 15 May 99
Posts: 3806
Credit: 1,114,826,392
RAC: 3,319
Canada
Message 1786705 - Posted: 11 May 2016, 14:16:53 UTC - in response to Message 1786703.  
Last modified: 11 May 2016, 14:20:14 UTC

So just checking some of the GUPPI from yesterday for validation.

I'm noticing a large swing in credit from 199 down to 29.

...

Can that be right? I didn't think that was supposed to do that.

I hope I am looking at this wrong, lack of sleep and such...


It looks that way; my RAC, which was rising, has tanked 3K since yesterday, so now we have both this and CreditEw/Pew/Screw dragging it down. And since we are finally getting GUPPI Kepler candidates on our NVidia GPUs, I'm overjoyed! :^D

Edit: This may be due to the longer processing time causing a longer return time for wingies. RAC may rise again.
ID: 1786705 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34381
Credit: 79,922,639
RAC: 80
Germany
Message 1786707 - Posted: 11 May 2016, 14:17:53 UTC - in response to Message 1786703.  
Last modified: 11 May 2016, 14:18:44 UTC

So just checking some of the GUPPI from yesterday for validation.

I'm noticing a large swing in credit from 199 down to 29.

When I look at the low credit work, it looks like I'm always paired with an ATI user

They take about 1000 sec longer to process but the value is at least 50-100 credits lower when I'm paired with a non-ATI user.

Can that be right? I didn't think that was supposed to do that.

I hope I am looking at this wrong, lack of sleep and such...


I`m sure its to early for such an assumption.

This happens to me when paired to a CPU task.

We should give it a few days to level out.


With each crime and every kindness we birth our future.
ID: 1786707 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51478
Credit: 1,018,363,574
RAC: 1,004
United States
Message 1786709 - Posted: 11 May 2016, 14:25:47 UTC

So far, the kitties are not amused.
RAC down 1k already.

Meowsigh.
"Time is simply the mechanism that keeps everything from happening all at once."

ID: 1786709 · Report as offensive
Chris Adamek
Volunteer tester

Send message
Joined: 15 May 99
Posts: 251
Credit: 434,772,072
RAC: 236
United States
Message 1786711 - Posted: 11 May 2016, 14:27:26 UTC - in response to Message 1786707.  
Last modified: 11 May 2016, 14:27:35 UTC

I'm getting 3 at a time in about 30 minutes on my Mac Pro with D700 cards. Waiting for some to make it to my Windows machine with a AMD 7990. My 750Ti's in my other mac are around 23 min running 1 at a time.

Chris
ID: 1786711 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51478
Credit: 1,018,363,574
RAC: 1,004
United States
Message 1786717 - Posted: 11 May 2016, 14:52:36 UTC

These things are taking over 2 hours on a 780....slower than the CPU?
Meow?
"Time is simply the mechanism that keeps everything from happening all at once."

ID: 1786717 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1786734 - Posted: 11 May 2016, 15:11:56 UTC - in response to Message 1786703.  
Last modified: 11 May 2016, 15:15:58 UTC

Can that be right? I didn't think that was supposed to do that.


Probably right, and no it's not 'supposed to' work like that.

Probably the numbers need to settle (as much as they can) for a while (like 3 weeks), then time for autopsy will be afterwards. In the meantime could bounce back a bit, but probably will still look 'messy'
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1786734 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1786736 - Posted: 11 May 2016, 15:16:21 UTC - in response to Message 1786717.  

Mark, how many tasks are you running?

I tried 2 at a time on my 750Ti's runtime was 1:00 to 1:05.
ID: 1786736 · Report as offensive
kittyman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 9 Jul 00
Posts: 51478
Credit: 1,018,363,574
RAC: 1,004
United States
Message 1786737 - Posted: 11 May 2016, 15:17:09 UTC - in response to Message 1786736.  
Last modified: 11 May 2016, 15:18:13 UTC

Mark, how many tasks are you running?

I tried 2 at a time on my 750Ti's runtime was 1:00 to 1:05.

Running 3/per on all GPUs.
Also noticed on the Killawatts that power usage is way down as well.
"Time is simply the mechanism that keeps everything from happening all at once."

ID: 1786737 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1786740 - Posted: 11 May 2016, 15:22:50 UTC - in response to Message 1786717.  
Last modified: 11 May 2016, 15:23:05 UTC

These things are taking over 2 hours on a 780....slower than the CPU?
Meow?


Leafing through looks like the 780 in my Mac Pro was asleep after the power outages here (storms), and before that I see some Guppi-MESSIERSs at 3 second overflows

Kicked the box back up to see if it'll snag any to compare.
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1786740 · Report as offensive
Profile Brent Norman Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 1 Dec 99
Posts: 2786
Credit: 685,657,289
RAC: 835
Canada
Message 1786741 - Posted: 11 May 2016, 15:23:28 UTC - in response to Message 1786737.  

I noticed my GPU temps dropped about 4.5c, and CPU temps went up about the same, could explain the power usage change.
ID: 1786741 · Report as offensive
Profile Gianfranco Lizzio
Volunteer tester
Avatar

Send message
Joined: 5 May 99
Posts: 39
Credit: 28,049,113
RAC: 87
Italy
Message 1786742 - Posted: 11 May 2016, 15:30:05 UTC

Running 'guppi' 3 at a time on my i7 4770 using avx optimized code i take 42-45 min.
Running 1 at time on my GTX 960 using Petri's cuda code i take 17-18 min.

This means that i7 with 'guppi' is 20% more efficient than GTX 960 and uses 36% less power!

Gianfranco
I don't want to believe, I want to know!
ID: 1786742 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1786745 - Posted: 11 May 2016, 15:32:11 UTC - in response to Message 1786742.  

Running 'guppi' 3 at a time on my i7 4770 using avx optimized code i take 42-45 min.
Running 1 at time on my GTX 960 using Petri's cuda code i take 17-18 min.

This means that i7 with 'guppi' is 20% more efficient than GTX 960 and uses 36% less power!

Gianfranco


That information will help the next hard development choices a lot :D
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1786745 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 . . . 10 · Next

Message boards : Number crunching : GBT ('guppi') .vlar tasks will be send to GPUs, what you think about this?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.