Monitoring inconclusive GBT validations and harvesting data for testing

Message boards : Number crunching : Monitoring inconclusive GBT validations and harvesting data for testing
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 . . . 36 · Next

AuthorMessage
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1811431 - Posted: 22 Aug 2016, 17:07:46 UTC - in response to Message 1811307.  

I think it would only be fair to link back to Jeff Buck's message 1810642, which sparked the whole idea off.

What particularly caught my eye in that specific WU was that each of the original 3 tasks flagged as Inconclusive showed counts that were different from each other, not just a two-against-one situation. The Triplet count, in particular, was different in each of the three. The signal summaries for the Pulses and Triplets appear to show where the main disagreements lie.

Task 5000863925 - SETI@home v8 v8.00 windows_intelx86
and my tiebreaker Task 5104834544 - SETI@home v8 Anonymous platform (NVIDIA GPU) [SSE3xj Win32 Build 3500] which supplied the detail shown
Spike count: 5
Autocorr count: 0
Pulse count: 5
Triplet count: 3
Gaussian count: 0

Pulse: peak=3.463834, time=45.84, period=6.89, d_freq=1793048198.64, score=1.04, chirp=-15.375, fft_len=512
Triplet: peak=10.94141, time=79.19, period=4.576, d_freq=1793047058.11, chirp=-18.222, fft_len=32
Triplet: peak=11.145, time=79.19, period=4.576, d_freq=1793047051.92, chirp=-27.333, fft_len=32
Pulse: peak=7.657701, time=45.9, period=20.34, d_freq=1793047378.49, score=1.009, chirp=-51.723, fft_len=2k
Pulse: peak=0.4216424, time=45.81, period=0.2394, d_freq=1793043214.91, score=1.003, chirp=-60.739, fft_len=32

Triplet: peak=11.70704, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64
Pulse: peak=2.575877, time=45.9, period=4.757, d_freq=1793047910.32, score=1.004, chirp=-74.832, fft_len=2k
Pulse: peak=5.652046, time=45.9, period=14.41, d_freq=1793053296.17, score=1.036, chirp=96.186, fft_len=2k
Best pulse: peak=3.463834, time=45.84, period=6.89, d_freq=1793048198.64, score=1.04, chirp=-15.375, fft_len=512

Best triplet: peak=11.70704, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64

Task 5000863926 - SETI@home v8 v8.00 (opencl_nvidia_mac) x86_64-apple-darwin [SSE3x OS X 64bit Build 3321]
Spike count: 5
Autocorr count: 0
Pulse count: 6
Triplet count: 2
Gaussian count: 0

Pulse: peak=3.448323, time=45.84, period=6.89, d_freq=1793048198.64, score=1.036, chirp=-15.375, fft_len=512
Triplet: peak=10.5995, time=86.75, period=4.384, d_freq=1793050046.62, chirp=-48.401, fft_len=1024
Pulse: peak=7.653532, time=45.9, period=20.34, d_freq=1793047378.49, score=1.008, chirp=-51.723, fft_len=2k
Pulse: peak=0.4235946, time=45.81, period=0.2394, d_freq=1793043214.91, score=1.008, chirp=-60.739, fft_len=32

Triplet: peak=11.62673, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64
Pulse: peak=2.582622, time=45.9, period=4.757, d_freq=1793047910.32, score=1.007, chirp=-74.832, fft_len=2k
Pulse: peak=1.258106, time=45.84, period=1.71, d_freq=1793049293.14, score=1.001, chirp=-95.853, fft_len=512
Pulse: peak=5.619142, time=45.9, period=14.41, d_freq=1793053296.17, score=1.029, chirp=96.186, fft_len=2k
Best pulse: peak=3.448323, time=45.84, period=6.89, d_freq=1793048198.64, score=1.036, chirp=-15.375, fft_len=512

Best triplet: peak=11.62673, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64

Task 5103387390 - SETI@home v8 v8.12 (opencl_intel_gpu_sah) windows_intelx86 [SSSE3xj Win32 Build 3430]
Spike count: 5
Autocorr count: 0
Pulse count: 6
Triplet count: 4
Gaussian count: 0

Pulse: peak=3.479678, time=45.84, period=6.89, d_freq=1793048198.64, score=1.045, chirp=-15.375, fft_len=512
Triplet: peak=10.85742, time=79.19, period=4.576, d_freq=1793047058.11, chirp=-18.222, fft_len=32
Triplet: peak=11.05782, time=79.19, period=4.576, d_freq=1793047051.92, chirp=-27.333, fft_len=32

Pulse: peak=7.675513, time=45.9, period=20.34, d_freq=1793047378.49, score=1.011, chirp=-51.723, fft_len=2k
Triplet: peak=10.98131, time=60.22, period=19.37, d_freq=1793046782.23, chirp=51.817, fft_len=1024
Pulse: peak=4.355305, time=45.9, period=11.04, d_freq=1793048395.46, score=1.002, chirp=60.881, fft_len=2k
Triplet: peak=12.00244, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64
Pulse: peak=2.598966, time=45.9, period=4.757, d_freq=1793047910.32, score=1.013, chirp=-74.832, fft_len=2k
Pulse: peak=5.616441, time=45.9, period=14.41, d_freq=1793053296.17, score=1.029, chirp=96.186, fft_len=2k
Pulse: peak=1.084218, time=45.82, period=1.296, d_freq=1793043223.34, score=1, chirp=99.459, fft_len=128
Best pulse: peak=3.479678, time=45.84, period=6.89, d_freq=1793048198.64, score=1.045, chirp=-15.375, fft_len=512

Best triplet: peak=12.00244, time=59.39, period=21.47, d_freq=1793054369.21, chirp=-63.776, fft_len=64

So, does any of that point to something that would help identify the root of the inconsistencies?
ID: 1811431 · 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 1811448 - Posted: 22 Aug 2016, 17:47:13 UTC - in response to Message 1811431.  
Last modified: 22 Aug 2016, 17:51:02 UTC

So, does any of that point to something that would help identify the root of the inconsistencies?


It does help, though the Cuda app isn't represented here, looks familiar. It indicates possible chirp accuracy drift, because affecting multiple signal types, and most of the signals seem to be at higher chirp rates. Also a possibility is that some floating point precision compiler option needs to be set, for the host code. In the case of Cuda apps on Windows, ms compilers, I had to set fp:precise. Comparison of Mac CPU and OpenCL gcc options with Eric's stock Windows (also gcc/DevC++), as gold reference on reliable host, could turn up something.

Depending on if those OpenCLs are nv GPUs, there are possible other considerations related to architecture generation (namely Some pre-Fermi devices don;t have double precision at all, and the float operations are not completely IEEE-754 compliant on these). That's something I've been able to manage in baseline Cuda code using special intrinsics and double emulation code, though not sure how it could be dealt with in OpenCL (if it needs to).
"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: 1811448 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1811451 - Posted: 22 Aug 2016, 17:59:39 UTC

And one other thing to consider. Both triplets and pulses need an average. To calculate average you must sum all values and divide by that value.

a) with cuda fast_math the division is an approximation. However you can specify for that single division to be fdiv_prec or something (that i'll try next when I get guppi work).

b) The sum for the average can be done sequentially, pairwise, in small batches, in a tree like parallel sum, ... all producing different 'sum' due to cumulating rounding errors and loss of precision (especially in sequential sum).

But isn't there the original data in the reported packet? The berkeley guys can run a double precision version on that. An error in the power at Nth decimal should not matter. The completely missed/extraneous pulses should trigger a third oppinion run though.
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: 1811451 · Report as offensive
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1811454 - Posted: 22 Aug 2016, 18:04:29 UTC - in response to Message 1811448.  

So, does any of that point to something that would help identify the root of the inconsistencies?


It does help, though the Cuda app isn't represented here, looks familiar.

I could've run my task w/ Cuda50 but wouldn't have gotten the signal detail that SoG provided (hint, hint).

Depending on if those OpenCLs are nv GPUs,

Ah, a detail I left out of my post. My task 5104834544 ran on a GTX 750Ti and the Mac task 5000863926 ran on a GTX 680MX.

I have the complete task detail pages, the original WU file, and my task's result file tucked away if anyone wants them for further testing or analysis.
ID: 1811454 · 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 1811455 - Posted: 22 Aug 2016, 18:06:14 UTC - in response to Message 1811431.  

So, does any of that point to something that would help identify the root of the inconsistencies?


Hmmm, possible lightbulb. Eric's reference is using 80 bit fpu settings for host code, while that's not available in the case of x64, and is possibly optimised out on the win32 build, unless fp:precise was set (presuming ms compiler).

Not sure if the OpenCL components have much/any bearing or not (maybe, maybe not), but the peak power differences do look like cumulative drift; noting that the reference values seem generally different than Eric's reference (as opposed to tiny amounts).
"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: 1811455 · 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 1811460 - Posted: 22 Aug 2016, 18:12:28 UTC - in response to Message 1811454.  
Last modified: 22 Aug 2016, 18:17:30 UTC

So, does any of that point to something that would help identify the root of the inconsistencies?


It does help, though the Cuda app isn't represented here, looks familiar.

I could've run my task w/ Cuda50 but wouldn't have gotten the signal detail that SoG provided (hint, hint).

Depending on if those OpenCLs are nv GPUs,

Ah, a detail I left out of my post. My task 5104834544 ran on a GTX 750Ti and the Mac task 5000863926 ran on a GTX 680MX.

I have the complete task detail pages, the original WU file, and my task's result file tucked away if anyone wants them for further testing or analysis.


I prefer analysing the result file by hand, with the aid of a diffing tool, having run them under lab conditions, as I don't trust Boinc's stderr, other machines, or formatting constraints that can be placed on stderr logs (Cuda needs more decimal places ;) )

Temptation is to point at result pages as evidence, then find it's some broken host.

if you are able to email me ( jason underscore groothuis at hotmail dot com ) result files, The task files, and stderrs from runs, I can certainly stick them under the microscope. The result files say a lot more than stderr ever can.

[Edit:] Entirely possible a quick bench run here would show Cuda different again, or favouring one side or another... That'd tell me more.
"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: 1811460 · Report as offensive
Profile Jeff Buck Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester

Send message
Joined: 11 Feb 00
Posts: 1441
Credit: 148,764,870
RAC: 0
United States
Message 1811462 - Posted: 22 Aug 2016, 18:17:43 UTC - in response to Message 1811460.  
Last modified: 22 Aug 2016, 18:50:33 UTC

if you are able to email me ( jason underscore groothuis at hotmail dot com ) result files, The task files, and stderrs from runs, I can certainly stick them under the microscope. The result files say a lot more than stderr ever can.

On the way....I hope.

EDIT: Just in case there's an email hitch, I've also uploaded a zip file to Amazon Cloud.
ID: 1811462 · 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 1811464 - Posted: 22 Aug 2016, 18:26:03 UTC - in response to Message 1811462.  
Last modified: 22 Aug 2016, 18:26:13 UTC

if you are able to email me ( jason underscore groothuis at hotmail dot com ) result files, The task files, and stderrs from runs, I can certainly stick them under the microscope. The result files say a lot more than stderr ever can.

On the way....I hope.


Thanks, No rush, I have to snooze and work in amongst thinking about seti stuff.

Dialling in the Cuda app to closer match Eric's v8 reference meant some difficult choices, I guess we'll find out soon enough if we need to bring in the rolling pin bearers.
"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: 1811464 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1811482 - Posted: 22 Aug 2016, 19:19:57 UTC - in response to Message 1811431.  

Adopting the same format as Jeff for my triple-inconclusive, WU 2239728586.

Task 5102978457 - SETI@home v8 v8.00 windows_intelx86
Spike count: 0
Autocorr count: 0
Pulse count: 24
Triplet count: 1
Gaussian count: 0

My tiebreaker task 5106883162 - AVXxj Win64 Build 3330
Spike count: 0
Autocorr count: 0
Pulse count: 24
Triplet count: 1
Gaussian count: 0

Pulse: peak=6.751799, time=45.99, period=17.72, d_freq=1209294052.59, score=1.093, chirp=-7.4253, fft_len=4k
Pulse: peak=4.036752, time=45.82, period=8.351, d_freq=1209302114.83, score=1.011, chirp=8.1932, fft_len=128
Pulse: peak=0.619823, time=45.82, period=0.4802, d_freq=1209300023.59, score=1.014, chirp=23.042, fft_len=128

Triplet: peak=10.17307, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024
Pulse: peak=1.450148, time=45.82, period=2.1, d_freq=1209298291.57, score=1.002, chirp=-48.902, fft_len=256
Pulse: peak=5.835546, time=45.86, period=13.87, d_freq=1209292599.49, score=1.045, chirp=50.246, fft_len=1024
Pulse: peak=10.35202, time=45.99, period=27.92, d_freq=1209299319.24, score=1.095, chirp=-50.982, fft_len=4k
Pulse: peak=7.454203, time=45.99, period=19.8, d_freq=1209299244.17, score=1.008, chirp=-52.614, fft_len=4k
Pulse: peak=9.831906, time=45.99, period=25.23, d_freq=1209299230.22, score=1.045, chirp=-52.918, fft_len=4k
Pulse: peak=9.473716, time=45.99, period=25.23, d_freq=1209299229.46, score=1.007, chirp=-52.934, fft_len=4k
Pulse: peak=10.21669, time=45.99, period=29.71, d_freq=1209299195.6, score=1.079, chirp=-53.67, fft_len=4k
Pulse: peak=9.480759, time=45.99, period=29.35, d_freq=1209299174.99, score=1.001, chirp=-54.118, fft_len=4k
Pulse: peak=7.496898, time=45.99, period=21.47, d_freq=1209299172.83, score=1.01, chirp=-54.165, fft_len=4k
Pulse: peak=7.800918, time=45.99, period=21.47, d_freq=1209299172.07, score=1.051, chirp=-54.182, fft_len=4k
Pulse: peak=7.667452, time=45.99, period=21.47, d_freq=1209299171.31, score=1.033, chirp=-54.198, fft_len=4k
Pulse: peak=8.00072, time=45.99, period=20.28, d_freq=1209299163.96, score=1.081, chirp=-54.358, fft_len=4k
Pulse: peak=9.424352, time=45.99, period=25.23, d_freq=1209299158.82, score=1.001, chirp=-54.47, fft_len=4k
Pulse: peak=6.229137, time=45.99, period=16.37, d_freq=1209299147.79, score=1.012, chirp=-54.71, fft_len=4k
Pulse: peak=4.621811, time=45.9, period=10.8, d_freq=1209300952.83, score=1.063, chirp=55.174, fft_len=2k
Pulse: peak=9.493025, time=45.99, period=25.05, d_freq=1209299112.47, score=1.009, chirp=-55.478, fft_len=4k
Pulse: peak=6.317374, time=45.99, period=16.82, d_freq=1209299095.54, score=1.024, chirp=-55.846, fft_len=4k
Pulse: peak=6.731335, time=45.99, period=16.82, d_freq=1209299094.78, score=1.091, chirp=-55.862, fft_len=4k
Pulse: peak=2.580231, time=45.99, period=5.458, d_freq=1209299074.93, score=1.024, chirp=-56.294, fft_len=4k
Pulse: peak=4.459141, time=45.99, period=10.92, d_freq=1209299074.17, score=1.051, chirp=-56.31, fft_len=4k
Pulse: peak=7.955166, time=45.9, period=18.67, d_freq=1209300658.74, score=1.051, chirp=93.322, fft_len=2k


Best pulse: peak=10.35202, time=45.99, period=27.92, d_freq=1209299319.24, score=1.095, chirp=-50.982, fft_len=4k
Best triplet: peak=10.17307, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024

Inconclusive task 5102978458 - SETI@home v8 v8.00 (opencl_nvidia_mac) x86_64-apple-darwin
Spike count: 0
Autocorr count: 0
Pulse count: 25 *** one extra
Triplet count: 1
Gaussian count: 0

Pulse: peak=6.75305, time=45.99, period=17.72, d_freq=1209294052.59, score=1.093, chirp=-7.4253, fft_len=4k
Pulse: peak=4.041894, time=45.82, period=8.351, d_freq=1209302114.83, score=1.012, chirp=8.1932, fft_len=128
Pulse: peak=0.6188955, time=45.82, period=0.4802, d_freq=1209300023.59, score=1.012, chirp=23.042, fft_len=128

Triplet: peak=10.21284, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024
(one missing at len=256)
Pulse: peak=5.902835, time=45.86, period=13.87, d_freq=1209292599.49, score=1.057, chirp=50.246, fft_len=1024
Pulse: peak=10.38408, time=45.99, period=27.92, d_freq=1209299319.24, score=1.099, chirp=-50.982, fft_len=4k
Pulse: peak=7.498811, time=45.99, period=19.8, d_freq=1209299244.17, score=1.014, chirp=-52.614, fft_len=4k
Pulse: peak=9.866991, time=45.99, period=25.23, d_freq=1209299230.22, score=1.048, chirp=-52.918, fft_len=4k
Pulse: peak=9.446428, time=45.99, period=25.23, d_freq=1209299229.46, score=1.004, chirp=-52.934, fft_len=4k
Pulse: peak=10.26235, time=45.99, period=29.71, d_freq=1209299195.6, score=1.083, chirp=-53.67, fft_len=4k
Pulse: peak=9.537156, time=45.99, period=29.35, d_freq=1209299174.99, score=1.007, chirp=-54.118, fft_len=4k
Pulse: peak=7.537107, time=45.99, period=21.47, d_freq=1209299172.83, score=1.016, chirp=-54.165, fft_len=4k
Pulse: peak=7.786188, time=45.99, period=21.47, d_freq=1209299172.07, score=1.049, chirp=-54.182, fft_len=4k
Pulse: peak=7.679629, time=45.99, period=21.47, d_freq=1209299171.31, score=1.035, chirp=-54.198, fft_len=4k
Pulse: peak=8.005, time=45.99, period=20.28, d_freq=1209299163.96, score=1.082, chirp=-54.358, fft_len=4k
Pulse: peak=9.499706, time=45.99, period=25.23, d_freq=1209299158.82, score=1.009, chirp=-54.47, fft_len=4k
Pulse: peak=4.64678, time=45.9, period=10.8, d_freq=1209300952.83, score=1.069, chirp=55.174, fft_len=2k
Pulse: peak=9.429302, time=45.99, period=25.05, d_freq=1209299112.47, score=1.002, chirp=-55.478, fft_len=4k
Pulse: peak=6.329872, time=45.99, period=16.82, d_freq=1209299095.54, score=1.026, chirp=-55.846, fft_len=4k
Pulse: peak=6.775892, time=45.99, period=16.82, d_freq=1209299094.78, score=1.099, chirp=-55.862, fft_len=4k
Pulse: peak=1.284491, time=45.82, period=1.556, d_freq=1209295817.29, score=1.005, chirp=-56.07, fft_len=256
Pulse: peak=2.553827, time=45.99, period=5.458, d_freq=1209299074.93, score=1.014, chirp=-56.294, fft_len=4k
Pulse: peak=4.492656, time=45.99, period=10.92, d_freq=1209299074.17, score=1.059, chirp=-56.31, fft_len=4k
Pulse: peak=7.644391, time=45.9, period=22.43, d_freq=1209299911.74, score=1.003, chirp=60.614, fft_len=2k
Pulse: peak=7.965314, time=45.9, period=18.67, d_freq=1209300658.74, score=1.052, chirp=93.322, fft_len=2k
Pulse: peak=5.469792, time=45.9, period=13.96, d_freq=1209300442.47, score=1.003, chirp=95.306, fft_len=2k


Best pulse: peak=10.38408, time=45.99, period=27.92, d_freq=1209299319.24, score=1.099, chirp=-50.982, fft_len=4k
Best triplet: peak=10.21284, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024

Inconclusive task 5104606775 - x41p_zi3d, Cuda 7.50 special
Spike count: 0
Autocorr count: 0
Pulse count: 25 *** one extra
Triplet count: 1
Gaussian count: 0

Pulse: peak=6.461686, time=45.99, period=17.81, d_freq=1209294052.59, score=1.046, chirp=-7.4253, fft_len=4k
Pulse: peak=4.036749, time=45.82, period=8.351, d_freq=1209302114.83, score=1.011, chirp=8.1932, fft_len=128
Pulse: peak=0.6198232, time=45.82, period=0.4802, d_freq=1209300023.59, score=1.014, chirp=23.042, fft_len=128

Triplet: peak=10.17307, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024
Pulse: peak=1.450149, time=45.82, period=2.1, d_freq=1209298291.57, score=1.002, chirp=-48.902, fft_len=256
Pulse: peak=5.835545, time=45.86, period=13.87, d_freq=1209292599.49, score=1.045, chirp=50.246, fft_len=1024
Pulse: peak=10.35202, time=45.99, period=27.92, d_freq=1209299319.24, score=1.095, chirp=-50.982, fft_len=4k
Pulse: peak=7.454203, time=45.99, period=19.8, d_freq=1209299244.17, score=1.008, chirp=-52.614, fft_len=4k
Pulse: peak=9.831906, time=45.99, period=25.23, d_freq=1209299230.22, score=1.045, chirp=-52.918, fft_len=4k
Pulse: peak=9.473717, time=45.99, period=25.23, d_freq=1209299229.46, score=1.007, chirp=-52.934, fft_len=4k
Pulse: peak=6.689241, time=45.99, period=16.64, d_freq=1209299220.64, score=1.085, chirp=-53.126, fft_len=4k *** extra?
Pulse: peak=10.2167, time=45.99, period=29.71, d_freq=1209299195.6, score=1.079, chirp=-53.67, fft_len=4k
Pulse: peak=9.480759, time=45.99, period=29.35, d_freq=1209299174.99, score=1.001, chirp=-54.118, fft_len=4k
Pulse: peak=7.496899, time=45.99, period=21.47, d_freq=1209299172.83, score=1.01, chirp=-54.165, fft_len=4k
Pulse: peak=7.800919, time=45.99, period=21.47, d_freq=1209299172.07, score=1.051, chirp=-54.182, fft_len=4k
Pulse: peak=7.667452, time=45.99, period=21.47, d_freq=1209299171.31, score=1.033, chirp=-54.198, fft_len=4k
Pulse: peak=8.00072, time=45.99, period=20.28, d_freq=1209299163.96, score=1.081, chirp=-54.358, fft_len=4k
Pulse: peak=9.424352, time=45.99, period=25.23, d_freq=1209299158.82, score=1.001, chirp=-54.47, fft_len=4k
Pulse: peak=6.229138, time=45.99, period=16.37, d_freq=1209299147.79, score=1.012, chirp=-54.71, fft_len=4k
Pulse: peak=4.62181, time=45.9, period=10.8, d_freq=1209300952.83, score=1.063, chirp=55.174, fft_len=2k
Pulse: peak=9.493025, time=45.99, period=25.05, d_freq=1209299112.47, score=1.009, chirp=-55.478, fft_len=4k
Pulse: peak=6.317374, time=45.99, period=16.82, d_freq=1209299095.54, score=1.024, chirp=-55.846, fft_len=4k
Pulse: peak=6.731335, time=45.99, period=16.82, d_freq=1209299094.78, score=1.091, chirp=-55.862, fft_len=4k
Pulse: peak=2.580231, time=45.99, period=5.458, d_freq=1209299074.93, score=1.024, chirp=-56.294, fft_len=4k
Pulse: peak=4.459141, time=45.99, period=10.92, d_freq=1209299074.17, score=1.051, chirp=-56.31, fft_len=4k
Pulse: peak=7.955166, time=45.9, period=18.67, d_freq=1209300658.74, score=1.051, chirp=93.322, fft_len=2k


Best pulse: peak=10.35202, time=45.99, period=27.92, d_freq=1209299319.24, score=1.095, chirp=-50.982, fft_len=4k
Best triplet: peak=10.17307, time=51.72, period=9.351, d_freq=1209300748.83, chirp=29.251, fft_len=1024

Which reveals that although the raw signal counts for both inconclusives show '+1 pulse', it's far more complicated than that - the first one obviously has a pulse missing, so there must be two extras - but when I tried to locate them, I got completely out of sync. The second inconclusive has a more obvious candidate for the extra pulse.

I can generate result files for automatic diffing from any application I can run here - but they all produce valid answers. We do need some of the people running the suspect builds to buy into this process, and provide result files for the builds which are producing the inconclusives in the first place - otherwise the published summaries, as copied here, are the only data we have to go on.
ID: 1811482 · 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 1811483 - Posted: 22 Aug 2016, 19:23:29 UTC - in response to Message 1811482.  

Nice, please email tasks, results and stderrs similarly. I know Petri will be especially interested in the findings.
"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: 1811483 · 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 1811484 - Posted: 22 Aug 2016, 19:25:30 UTC - in response to Message 1811482.  

otherwise the published summaries, as copied here, are the only data we have to go on.


Exactly. Printing data in stderr is feelgood happy times, but the real meat is in the result files.
"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: 1811484 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1811492 - Posted: 22 Aug 2016, 19:36:10 UTC - in response to Message 1811483.  

Nice, please email tasks, results and stderrs similarly. I know Petri will be especially interested in the findings.

Will have to re-run them offline to get result files - been a busy day for the real world to intrude today, so beer is calling next. I should have more time to devote to this through the rest of the week, and I'm also planning to draw up a medal table for the number of times different apps show up as one half of an inconclusive, but go (weakly) valid in the end. More to come in this thread.
ID: 1811492 · 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 1811494 - Posted: 22 Aug 2016, 19:37:37 UTC - in response to Message 1811492.  

Nice, please email tasks, results and stderrs similarly. I know Petri will be especially interested in the findings.

Will have to re-run them offline to get result files - been a busy day for the real world to intrude today, so beer is calling next. I should have more time to devote to this through the rest of the week, and I'm also planning to draw up a medal table for the number of times different apps show up as one half of an inconclusive, but go (weakly) valid in the end. More to come in this thread.


oooh, nice, yep grabbing a beer
"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: 1811494 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1811556 - Posted: 22 Aug 2016, 22:22:34 UTC
Last modified: 22 Aug 2016, 22:25:02 UTC

Hi,

I've got now dozens of guppi tasks in the queue so soon they'll start flowing in to the database.

I'll be waiting the results. (A guess: 1-4 bit errors in the lsb of the mantissa of a float 'peak' snr/thresh)

EDIT: And now they are coming: (Those 164 seconds/task ar 0.07)
http://setiathome.berkeley.edu/results.php?hostid=7475713&offset=0&show_names=0&state=2&appid=29

Petri
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: 1811556 · Report as offensive
Kiska
Volunteer tester

Send message
Joined: 31 Mar 12
Posts: 302
Credit: 3,067,762
RAC: 0
Australia
Message 1811621 - Posted: 23 Aug 2016, 0:58:29 UTC - in response to Message 1811482.  

I think these are the extras you might be looking for:

extras
ID: 1811621 · 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 1811643 - Posted: 23 Aug 2016, 1:39:59 UTC
Last modified: 23 Aug 2016, 2:02:28 UTC

Someone have a download/link for current Windows stock x86 CPU multibeam (including libraries) ? Mine appears to have gone AWOL from bench, and would like to generate a reference for comparison of Jeff's WU result, Cuda, and later Petri's code.

[Edit:] nevermind, got it --> directed Lunatics installer to install said app in a blank folder. [Test running]
"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: 1811643 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1811689 - Posted: 23 Aug 2016, 3:16:26 UTC
Last modified: 23 Aug 2016, 3:37:09 UTC

I've found a nasty Autocorr error that seems to have developed over the past few days in the CUDA Special App. Looking at past results it appears to have materialized around the 16-17th. That would be about the time I started building Apps from r3516. I did notice a sharp increase in Inconclusive results about that time but just passed it off as a Flood of GUPPIs. I have reverted the build to exclude that change, we'll see how it goes.

The problem shows up as an unreal Autocorr: peak.
Autocorr: peak=131072, time=6.711, delay=0.004096, d_freq=1419599801.58, chirp=28.641, fft_len=128k
I have found the problem on my OSX machine, my Linux machine, and Petri's machine.
Hopefully change 3515 was the problem.

In other news my otherwise well preforming ATI Linux machine is being challenged by the iGPUs again. Yes, a Mac is again involved;
http://setiathome.berkeley.edu/workunit.php?wuid=2233419202
SETI@home v8 v8.12 (opencl_intel_gpu_sah) windows_intelx86
Spike count: 9
Autocorr count: 0
Pulse count: 7
Triplet count: 1
Gaussian count: 0

SETI@home v8 v8.00 (opencl_intel_gpu_sah) x86_64-apple-darwin
Spike count: 9
Autocorr count: 0
Pulse count: 4
Triplet count: 0
Gaussian count: 0

SETI@home v8 Anonymous platform (ATI GPU)
Spike count: 9
Autocorr count: 0
Pulse count: 6
Triplet count: 0
Gaussian count: 0

I'm kinda busy trying to track down this Autocorr error to fool with iGPUs.

crap...there it is again;
Autocorr: peak=71016.57, time=6.711, delay=0.0016384, d_freq=1421220765.4, chirp=9.2796, fft_len=128k
Autocorr: peak=130882.8, time=6.711, delay=0.0013312, d_freq=1421220843.11, chirp=20.859, fft_len=128k
Autocorr: peak=67537.06, time=6.711, delay=0.0013312, d_freq=1421220553.87, chirp=-22.24, fft_len=128k
Autocorr: peak=69187.27, time=6.711, delay=0.0013312, d_freq=1421220861.26, chirp=23.564, fft_len=128k
http://setiathome.berkeley.edu/result.php?resultid=5111325626
4 in one task? WTH?
ID: 1811689 · 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 1811733 - Posted: 23 Aug 2016, 5:55:51 UTC - in response to Message 1811689.  
Last modified: 23 Aug 2016, 6:01:08 UTC

A problem with autocorrelations in the special code, should crop up as I ramp toward special alpha builds. Fortunately once chirps, 128k FFTs and powerspectra are confirmed accurate enough or fixed, then that only leaves the straightforward Autocorrelation change Petri's made. Since I crafted the original GPU autocorrelation in Matlab myself, plugging his modified variant that uses smaller transforms through Matlab myself should be relatively trivial (being the thorough, long route)

With testing Jeff Buck's example, Stock Win32 CPU, and Win32 Cuda 50 are confirmed Strongly similar, with Q=99.74%, which just confirms what we already knew, the other builds in the online set have some issues (if not the systems themselves). Reference results say:
Spike count: 5
Autocorr count: 0
Pulse count: 5
Triplet count: 3
Gaussian count: 0


So no sixth reportable pulse according to stock reference build and the totally different Cuda50 app.

Will probably try to work out a way to check chirp accuracy under bench, or by hand inspection, and recall I may have a WU from Joe a long time ago that could be updated to v8. Will dig for that later, and possibly see If I can collect some bench and applications for the Mac, to attempt to reproduce the variations under controlled conditions.
"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: 1811733 · Report as offensive
TBar
Volunteer tester

Send message
Joined: 22 May 99
Posts: 5204
Credit: 840,779,836
RAC: 2,768
United States
Message 1811760 - Posted: 23 Aug 2016, 7:01:58 UTC - in response to Message 1811733.  

I may have found it. It could be the autocorrelation blocking sync added a few days ago. I thought it started before then, but, maybe not. I removed it from the AC section and for now it seems alright, but, it's working GUPPIs now, and the problem seems to occur on the Arecibo tasks AFAIK.
For now the settings below seem to be working;
cudaEventCreateWithFlags(&chirpDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&fftDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&summaxDoneEvent, cudaEventDisableTiming|(blockingSync ? cudaEventBlockingSync : 0));
cudaEventCreateWithFlags(&powerspectrumDoneEvent, cudaEventDisableTiming);

cudaEventCreateWithFlags(&autocorrelationDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&autocorrelationRepackDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&ac_reduce_partialEvent, cudaEventDisableTiming);
      
cudaEventCreateWithFlags(&tripletsDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&tripletsDoneEvent1, cudaEventDisableTiming|cudaEventBlockingSync);
cudaEventCreateWithFlags(&pulseDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&pulseDoneEvent1, cudaEventDisableTiming|cudaEventBlockingSync);
cudaEventCreateWithFlags(&gaussDoneEvent, cudaEventDisableTiming);
cudaEventCreateWithFlags(&gaussDoneEvent2, cudaEventDisableTiming|cudaEventBlockingSync);

CPU use is up a little on the Arecibo tasks but about the same on the GUPPIs.
This will work, IF it keeps working.
ID: 1811760 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1811761 - Posted: 23 Aug 2016, 7:09:50 UTC - in response to Message 1811733.  
Last modified: 23 Aug 2016, 7:12:38 UTC

Jason, could you email me the problematic WU and the correct result file to compare to so I can start debugging?
Both cases the extra pulse and autocorr 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: 1811761 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 . . . 36 · Next

Message boards : Number crunching : Monitoring inconclusive GBT validations and harvesting data for testing


 
©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.