Posts by Raistmer

1) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2011295)
Posted 8 Sep 2019 by Profile Raistmer
Post:
Well, at least on OpenCL runtime level this shouldn't matter.
It operates logical entities like CU, queue and work-item w/o knowledge of their implementation in hardware. Driver does though. So it seems AMD driver doesn't understand AMD hardware well enough.
2) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010780)
Posted 5 Sep 2019 by Profile Raistmer
Post:
Well, it overflows almost immediately - zero chirp.
Maybe, very first FFT was bad one.
Spike search performed right after FFT. And on zero chirp even de-chirping kernel shouldnotaffect result.
Plain FFT and comparison of bin's power with threshold.
3) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010741)
Posted 5 Sep 2019 by Profile Raistmer
Post:
I'd suggest that somebody with a card and some programming experience grabs https://github.com/Oblomov/clinfo (Windows ready-built at foot of page: linux needs - I think - building from sources) and posts the output from that. It will carry far more weight with AMD.

+++
4) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010724)
Posted 5 Sep 2019 by Profile Raistmer
Post:
https://community.amd.com/message/2928820
5) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010720)
Posted 5 Sep 2019 by Profile Raistmer
Post:
And did someone thread about it on AMD OpenCL forums?
Anyone with ability to do offline testing and possession of such "broken" hardware+software?

Phoronix did testing and reviews of the RX 5700XT and could not get the card and drivers to pass the OpenCL parts of their standardized test suite.

Thanks. Shame on AMD . Natural dunces :/
6) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010627)
Posted 4 Sep 2019 by Profile Raistmer
Post:
This is becoming much more common, similar to the problem with the APs,

https://setiathome.berkeley.edu/workunit.php?wuid=3597951375
   Task    Computer            Sent                 Time reported                Status             Run time  CPU time  Credit                         Application
7934624541 8534188  8 Aug 2019,  5:33:40 UTC  8 Aug 2019,  5:50:25 UTC  Completed and validated        17.34    13.97    1.87   SETI@home v8 v8.22 ([b]opencl_ati5_SoG_nocal[/b]) windows_intelx86
7936540283 6942127  8 Aug 2019, 21:19:24 UTC  8 Aug 2019, 21:55:40 UTC  Completed and validated        23.18    20.08    1.87   SETI@home v8 v8.22 ([b]opencl_ati_nocal[/b]) windows_intelx86
All the AMD GPUs have Hundreds of Invalids elsewhere. I suggest action sooner rather than later as it will undoubtedly become worse with more RX 5700 XT GPUs arriving. Maybe just One AMD GPU per WorkUnit?




For future reference could anyone posting such comparisons also to grab stderr outputs too while they are available, please.
7) Message boards : Number crunching : Flakey AMD/ATI GPUs, including RX 5700 XT, Cross Validating, polluting the Database (Message 2010625)
Posted 4 Sep 2019 by Profile Raistmer
Post:
And did someone thread about it on AMD OpenCL forums?
Anyone with ability to do offline testing and possession of such "broken" hardware+software?
8) Message boards : Number crunching : SETI@home v8.22 Windows GPU applications support thread (Message 2009487)
Posted 27 Aug 2019 by Profile Raistmer
Post:
VHAR is the kind of task where SoG plays at full power. No PulseFind. So almost whole OpenCL calls scheduled to run to GPU immediately - true parallel execution. If driver fails to process such long sequence it restarted.
That's why VHAR might be more fragile for any driver changes.
9) Message boards : Number crunching : whole serie of data blocks failing with SoG (Message 2009486)
Posted 27 Aug 2019 by Profile Raistmer
Post:
Such symptoms usually mean video driver restart. App doesn't get informed about such event, last OpenCL runtime call just never returns. Hence abortion by deadline reaching.
To change driver is good advice provided app worked OK with another driver version. And to narrow area where to find issue I would recommend to add (temporarily) -v 2 option.
Also there was special debug build that reports each OpenCL call into stderr - that would be ideal to see where exactly problem occurs.
From OS side-check system log to see if driver restart events had place.
10) Message boards : Number crunching : Windows 10 - Yea or Nay? (2) (Message 2008804)
Posted 22 Aug 2019 by Profile Raistmer
Post:
I would like to share some hack I developed to stop my netbook from being unusable by its OS with few days period.
It has Win10 (home) and only 32GB "SSD"(that could not be real one) drive in it. So OS updates constantly eat it up to the point of OS instability due to lack of room on system drive.
I really tired to block updates via 3rd party stuff and tricks from internet so I:
1) found place where updates downloaded C:\Windows\SoftwareDistribution\
2) run as administrator
3) edited NTFS security descriptor for that dir to DENY System account for accessing to it (to be sure also took ownership to own acc).
4) deleted all it has to that moment.

And few months already I enjoy my netbook working FOR me,not AGAINST me.
Small win in perpetual M$ battle. In quite radical way but I saved from OS restarts when it wants with losing my work and from constant lack of room on system drive. Maybe some find it useful too,especially for devices with small system partition.
11) Message boards : Number crunching : Nvidia Cuda Errors (Message 2007142)
Posted 13 Aug 2019 by Profile Raistmer
Post:
Try to use CUDA 3.2 app instead of current one.
It has lower memory requirements and most probably will provide better performance also.
CUDA42 was slowest for almost all NV-based systems I used.
12) Message boards : Number crunching : SETI & the system page file? (Message 2002332)
Posted 13 Jul 2019 by Profile Raistmer
Post:
Well, interesting application for page file... but 2 questions arise:
1) Page file can contain only moveable pages, not pinned ones. So, how such defrag could ever work when pinned memory becomes fragmented? (actually, can't at all).
2) Not pinned memory accessible as virtual pages. That is physical pages are mapped into processes virtual address spaces. And actually nothing prevents to move physical pages around still maintaining same maping to address spaces of processes, doing "in memory" defrag. Why whould Windows need page file then?...
In short: pinned memory can't be swapped out at all and moveable can be relocated between physical pages in memory.
13) Message boards : News : New SETI Perspectives: "Seeing the Unseeable: The Black Hole Image" (Message 2001796)
Posted 9 Jul 2019 by Profile Raistmer
Post:
Да - осилим. Приветствую - недавно присоединился с 1 июля к проекту сети - как успехи в обработке? Я из Сибири. С Братска.

Приветствую в проекте :)
14) Message boards : Number crunching : SETI & the system page file? (Message 2000520)
Posted 1 Jul 2019 by Profile Raistmer
Post:
Having swap file "just for safe" means trust in OS algorithms for page evictions.
I would not too trust Windows's ones cause saw swaps even when enough physical memory was available. It has too many different user behavior prediction mechanisms to load RAM with useless data and swap useful one to disk.
In case of BOINC + compute intensive app (as SETI) any real need in swap-file will mean enormous performance degradation. App's data accessed constantly.
Single use of swap here is to keep suspended tasks "in memory". But even here one should compare time for re-initialization from state file (state file small so much less drive accesses when on swap but memory structures creation and once-per-task computations should be repeated) and swap from pagefile.
So, for compute-only hosts one should avoid pagefile at all, use large memory pages (to reduce TLB pollution) and use pinned memory (to simplify address translation). Not all of this easely user-controlled, but...
15) Message boards : Number crunching : Crunching time with respect to Angle Range (Message 1998662)
Posted 18 Jun 2019 by Profile Raistmer
Post:

Instead, I think it might be some other factor in the search specification which dominates for GPUs. But I have no idea what that might be.

Number of "just under threshold" signals.
In Pulses and Triplets (at least) they will cause either serialization (perhaps for CUDA special) or re-calculation by CPU(for SoG).

EDIT: to prove or disprove this ones running OpenCL builds could look into statistics part of stderr:


class Gaussian_transfer_not_needed: total=0, N=0, <>=0, min=0 max=0
class Gaussian_transfer_needed: total=0, N=0, <>=0, min=0 max=0


class Gaussian_skip1_no_peak: total=0, N=0, <>=0, min=0 max=0
class Gaussian_skip2_bad_group_peak: total=0, N=0, <>=0, min=0 max=0
class Gaussian_skip3_too_weak_peak: total=0, N=0, <>=0, min=0 max=0
class Gaussian_skip4_too_big_ChiSq: total=0, N=0, <>=0, min=0 max=0
class Gaussian_skip6_low_power: total=0, N=0, <>=0, min=0 max=0


class Gaussian_new_best: total=0, N=0, <>=0, min=0 max=0
class Gaussian_report: total=0, N=0, <>=0, min=0 max=0
class Gaussian_miss: total=0, N=0, <>=0, min=0 max=0


class PC_triplet_find_hit: total=41744, N=41744, <>=1, min=1 max=1
class PC_triplet_find_miss: total=832, N=832, <>=1, min=1 max=1


class PC_pulse_find_hit: total=30209, N=30209, <>=1, min=1 max=1
class PC_pulse_find_miss: total=12, N=12, <>=1, min=1 max=1
class PC_pulse_find_early_miss: total=7, N=7, <>=1, min=1 max=1
class PC_pulse_find_2CPU: total=0, N=0, <>=0, min=0 max=0


class PoT_transfer_not_needed: total=41737, N=41737, <>=1, min=1 max=1
class PoT_transfer_needed: total=840, N=840, <>=1, min=1 max=1

Names mostly self-explanatory ( I hope)

For example: class PC_triplet_find_miss: total=832, N=832, <>=1, min=1 max=1
So, 832 times triplets were relooked by CPU. Obviously task can't have so many Triplets in result so they were signal candidates,not real Triplets.
Nevertheless their number slowed down progress of that particular task.
16) Message boards : News : 20 years and counting! (Message 1994563)
Posted 21 May 2019 by Profile Raistmer
Post:
SETI@home member since 16 May 1999. I remember hearing about the project in the news and immediately signing up.

I agree, it seems like a long time.... : )

Is there a way to know not just the date but also how far down the list of who signed on the earliest?


Started under another account those days with mail on not existing by now server... but still holding first certificates printed on matrix printer :)
17) Message boards : News : 20 years and counting! (Message 1994562)
Posted 21 May 2019 by Profile Raistmer
Post:
Happy anniversary to all !!!
18) Message boards : Number crunching : Downloading tasks for offline re-check (Message 1991785)
Posted 28 Apr 2019 by Profile Raistmer
Post:

Edit - the big problem comes in when a user has both a new generation card and an older generation card in the same computer, and manually sets BOINC to use all GPUs. The plan class mechanism can't overrule the user's choices, to keep the restricted application off the lesser card.

BOINC client could obey plan class restrictions too. Actually it knows all needed info. It's client who reports precise host info to server to make choice what to send and what not to send.
In return client recives plan class along with its restrictions. Why not doing correct scheduling then?....

I think app itself should have fail-safe measures like those I embed in OpenCL apps. App checks GPU compatibility. And then should just end with error code instead of providing wrong results.
It will be problem on mixed GPU hosts still, but at least will not resut in any problems with reported results validity.
19) Message boards : Number crunching : Downloading tasks for offline re-check (Message 1991724)
Posted 27 Apr 2019 by Profile Raistmer
Post:
Hmmmm, the amount of Memory Booster needed to fix that post would probably pose a health risk. Perhaps you should ask Raistmer where he was getting all those Mac Apps he passed on to Eric.

Well, what reminder actually I need is do we had any of Petri's builds on beta already or still not?
For OpenCL issues on Mac there could be good solution to abandon OpenCL in favor of Petri's CUDA special (and this requires passing through beta). At least for NV.
20) Message boards : Number crunching : High performance Linux clients at SETI (Message 1991425)
Posted 25 Apr 2019 by Profile Raistmer
Post:
I played around with values and could not detect any improvement from stock. Petri also commented that stock values are best. The -unroll 1 has to do with the improved pulse search algorithm.

Thanks. So I'll just collect statistics running 3 CPU + GPU with default settings.


Next 20


 
©2019 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.