Collatz Sieve 1.20 released for Windows, Linux, and OS X
log in

Advanced search

Message boards : News : Collatz Sieve 1.20 released for Windows, Linux, and OS X

Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 21191 - Posted: 4 Sep 2015, 20:48:41 UTC

Version 1.20 of the Collatz Sieve has been released for testing. (You have to enable beta/test apps in your preferences in order to get collatz_sieve workunits.) The code has been cleaned up since 1.10 and a couple bugs fixed. Both 32 and 64 bit versions for Linux and Windows are now available. OS X is limited to 64 bit only.

The Linux applications were compiled on Oracle's version of RHEL 6.5 which is from the CUDA 3.0 era so I'm hoping the binaries will work on other newer Linux distributions. As usual, if it doesn't, let me know.

The OS X applications were compiled on Mavericks so I have no idea whether older Macs will be able to run it or now. Supposedly, it should work for 10.5 or later but I'm not holding my breath on that one. I'm hoping the kernel will build properly on the AMD machines as I've only tested on Mavericks with an nVidia GPU.

Profile [AF>Le_Pommier>MacBidouille.com]Prof
Send message
Joined: 27 Dec 11
Posts: 22
Credit: 1,446,110,953
RAC: 234,229
Message 21195 - Posted: 4 Sep 2015, 22:06:12 UTC - in response to Message 21191.
Last modified: 4 Sep 2015, 22:09:03 UTC

Hi!
I'Ve try that Immediately..... Errors in 2 sec!!!
If you need more infos, tell me.

copy of the Stderr :

Stderr output

<core_client_version>7.4.42</core_client_version>
<![CDATA[
<message>
process exited with code 245 (0xf5, -11)
</message>
<stderr_txt>
Collatz Conjecture Sieve 1.20 OS X x86_64 for OpenCL
Written by Slicker (Jon Sonntag) of team SETI.USA
Based on the AMD Brook+ kernels by Gipsel of team Planet 3DNow!
Sieve code and OpenCL optimization provided by Sosiris of team BOINC@Taiwan
BUILD LOG
Error returned by cvms_element_build_from_source
clBuildProgram() failed with error (-11)
Error: (-11)Program build failure at 1163 of SetupOpenCL

Error -11. Processing Aborted.
23:58:43 (95098): called boinc_finish

</stderr_txt>
]]>

Rymorea
Send message
Joined: 14 Oct 14
Posts: 100
Credit: 200,411,819
RAC: 3
Message 21196 - Posted: 4 Sep 2015, 23:25:22 UTC
Last modified: 4 Sep 2015, 23:30:06 UTC

I start new ver 1.2 wus come 1by1 every 2 min interval.
It shows 38hour but done in 2 min.
I try 5 and all Validate error. I stop to try new version till you find the problem.

http://boinc.thesonntags.com/collatz/result.php?resultid=21712987
http://boinc.thesonntags.com/collatz/result.php?resultid=21712972
http://boinc.thesonntags.com/collatz/result.php?resultid=21712926
http://boinc.thesonntags.com/collatz/result.php?resultid=21711895
http://boinc.thesonntags.com/collatz/result.php?resultid=21711190
____________
Seti@home Classic account User ID 955 member since 8 Sep 1999 classic CPU time 539,770 hours

Profile dkester788
Send message
Joined: 11 Feb 14
Posts: 4
Credit: 8,610,185
RAC: 0
Message 21197 - Posted: 5 Sep 2015, 3:13:49 UTC

Looks like the same thing with all my Sieve 1.20 WUs. I've tried 25+ and all came back "Validate Error". If I can provide additional information please let me know.

Thanks!

Profile dkester788
Send message
Joined: 11 Feb 14
Posts: 4
Credit: 8,610,185
RAC: 0
Message 21198 - Posted: 5 Sep 2015, 3:16:46 UTC - in response to Message 21197.
Last modified: 5 Sep 2015, 3:25:02 UTC

http://boinc.thesonntags.com/collatz/result.php?resultid=21718018
http://boinc.thesonntags.com/collatz/result.php?resultid=21717941
http://boinc.thesonntags.com/collatz/result.php?resultid=21717906
http://boinc.thesonntags.com/collatz/result.php?resultid=21717884
http://boinc.thesonntags.com/collatz/result.php?resultid=21717813
http://boinc.thesonntags.com/collatz/result.php?resultid=21717052
http://boinc.thesonntags.com/collatz/result.php?resultid=21654393

A small sample of the "Validate Failures - the last one is a "Valid" on 1.10.

Profile Thomas Krajewsky
Avatar
Send message
Joined: 2 Feb 10
Posts: 13
Credit: 379,296,674
RAC: 0
Message 21199 - Posted: 5 Sep 2015, 7:53:43 UTC

invalid
http://boinc.thesonntags.com/collatz/result.php?resultid=21722593
http://boinc.thesonntags.com/collatz/result.php?resultid=21721088

error (no stderr output, state: 998 (0x3e6) Unknown error number)
http://boinc.thesonntags.com/collatz/result.php?resultid=21722905
http://boinc.thesonntags.com/collatz/result.php?resultid=21723123
http://boinc.thesonntags.com/collatz/result.php?resultid=21722162

error
http://boinc.thesonntags.com/collatz/result.php?resultid=21723240
stderr output:
<core_client_version>7.6.9</core_client_version>
<![CDATA[
<message>
Zugriff verweigert (-> german for 'access denied')
(0x5) - exit code 5 (0x5)
</message>
<stderr_txt>
Collatz Conjecture Sieve 1.20 Windows x86_64 for OpenCL
Written by Slicker (Jon Sonntag) of team SETI.USA
Based on the AMD Brook+ kernels by Gipsel of team Planet 3DNow!
Sieve code and OpenCL optimization provided by Sosiris of team BOINC@Taiwan
Unrecognized command: items_per_kernel=20
Unrecognized command: reducecpu=0
Collatz Config Settings:
verbose 1 (yes)
kernels/reduction 48
threads 2^7 (128)
lut_size 13 (65536 bytes)
sieve_size 2^30 (51085096 bytes)
sleep 1
cache_sieve 1 (yes)
reducecpu 0 (no)
Platform NVIDIA
Device 0000007A9194BA30
Max Dimensions 3
Max Work Items 1024 1024 64
Max Work Groups 1024
Max Kernel Threads 1024
Device Vendor NVIDIA Corporation
Name GeForce 820M
Driver Version 355.82
OpenCL Version OpenCL 1.1 CUDA
actual threads 128
Error: (-5)Out of resources at 1641 of RunSteps
Error: GPU steps do not match CPU steps. Workunit processing aborted.
09:30:55 (1452): called boinc_finish

</stderr_txt>
]]>
The .config for this wu was the same as for 1.10, which were running well.

Rymorea
Send message
Joined: 14 Oct 14
Posts: 100
Credit: 200,411,819
RAC: 3
Message 21200 - Posted: 5 Sep 2015, 9:57:49 UTC

I have upgrade the boinc ver 7.4.42 to 7.6.9 and want to try again.

all 5 Validate error.

http://boinc.thesonntags.com/collatz/result.php?resultid=21725053
http://boinc.thesonntags.com/collatz/result.php?resultid=21725015
http://boinc.thesonntags.com/collatz/result.php?resultid=21724944
http://boinc.thesonntags.com/collatz/result.php?resultid=21724164
http://boinc.thesonntags.com/collatz/result.php?resultid=21724127

If you need additional log you can inform.

[AF>Amis des Lapins] Jean-Luc
Send message
Joined: 11 Jun 10
Posts: 6
Credit: 2,565,041,701
RAC: 34,699
Message 21201 - Posted: 5 Sep 2015, 11:02:45 UTC - in response to Message 21200.
Last modified: 5 Sep 2015, 11:03:12 UTC

For me :
Invalid :
http://boinc.thesonntags.com/collatz/result.php?resultid=21725177
http://boinc.thesonntags.com/collatz/result.php?resultid=21725035
On my R9 290X

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 21205 - Posted: 5 Sep 2015, 16:25:33 UTC - in response to Message 21199.

invalid
http://boinc.thesonntags.com/collatz/result.php?resultid=21722593
http://boinc.thesonntags.com/collatz/result.php?resultid=21721088

error (no stderr output, state: 998 (0x3e6) Unknown error number)
http://boinc.thesonntags.com/collatz/result.php?resultid=21722905
http://boinc.thesonntags.com/collatz/result.php?resultid=21723123
http://boinc.thesonntags.com/collatz/result.php?resultid=21722162

error
http://boinc.thesonntags.com/collatz/result.php?resultid=21723240
stderr output:
<core_client_version>7.6.9</core_client_version>
<![CDATA[
<message>
Zugriff verweigert (-> german for 'access denied')
(0x5) - exit code 5 (0x5)
</message>
<stderr_txt>
Collatz Conjecture Sieve 1.20 Windows x86_64 for OpenCL
Written by Slicker (Jon Sonntag) of team SETI.USA
Based on the AMD Brook+ kernels by Gipsel of team Planet 3DNow!
Sieve code and OpenCL optimization provided by Sosiris of team BOINC@Taiwan
Unrecognized command: items_per_kernel=20
Unrecognized command: reducecpu=0
Collatz Config Settings:
verbose 1 (yes)
kernels/reduction 48
threads 2^7 (128)
lut_size 13 (65536 bytes)
sieve_size 2^30 (51085096 bytes)
sleep 1
cache_sieve 1 (yes)
reducecpu 0 (no)
Platform NVIDIA
Device 0000007A9194BA30
Max Dimensions 3
Max Work Items 1024 1024 64
Max Work Groups 1024
Max Kernel Threads 1024
Device Vendor NVIDIA Corporation
Name GeForce 820M
Driver Version 355.82
OpenCL Version OpenCL 1.1 CUDA
actual threads 128
Error: (-5)Out of resources at 1641 of RunSteps
Error: GPU steps do not match CPU steps. Workunit processing aborted.
09:30:55 (1452): called boinc_finish

</stderr_txt>
]]>
The .config for this wu was the same as for 1.10, which were running well.


Lower the sieve size to 28. The 1.10 subdivided the sieve into 4 parts whereas the 1.20 does not so it now runs 4 times as many items per kernel. Try using 28 for the sieve to get the same effect. On the flip side, because it now runs one kernel per sieve instead of 4, you should be able to increase the kernels per reduction.

Then again, you'll have to wait for v1.21 which fixes the result/validator issue that is happening now.

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,752,306
RAC: 339
Message 21206 - Posted: 5 Sep 2015, 16:31:12 UTC
Last modified: 5 Sep 2015, 16:35:11 UTC

Validate errors on my macs with intel and with nvidia. I have no config files set.

http://boinc.thesonntags.com/collatz/result.php?resultid=21732299

These tasks are failing to validate across all typed of OSs and GPUs too.

http://boinc.thesonntags.com/collatz/workunit.php?wuid=18461714
____________
Dublin, California
Team: SETI.USA

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 21208 - Posted: 5 Sep 2015, 16:40:53 UTC - in response to Message 21195.

BUILD LOG
Error returned by cvms_element_build_from_source
clBuildProgram() failed with error (-11)
Error: (-11)Program build failure at 1163 of SetupOpenCL

Error -11. Processing Aborted.
23:58:43 (95098): called boinc_finish


Notice the blank line? That's where the actual line number and error in the kernel are supposed to print. It looks like Apples/AMD's compiler still has the bug where it just crashes rather than return the error that occurred. I guess that's what you get when a company is more focused on wrist watches rather than operating systems.

There are ZERO differences between the OpenCL kernels used on the Windows, Linux and OS X applications. Given that it compiles fine on the other platforms and when using the nVidia driver on OS X, it seems to me that the issue is clearly with the AMD/OS X driver which means there is nothing I can do other than complain to AMD and Apple that the problem still isn't fixed.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 21209 - Posted: 5 Sep 2015, 16:45:09 UTC - in response to Message 21206.

Validate errors on my macs with intel and with nvidia. I have no config files set.

http://boinc.thesonntags.com/collatz/result.php?resultid=21732299

These tasks are failing to validate across all typed of OSs and GPUs too.

http://boinc.thesonntags.com/collatz/workunit.php?wuid=18461714


But they are finishing and reporting results. That's much better than not being able to compile the app. The issue is either in the output of the app or on the server. Since the output is supposed to be the same and since the server code hasn't changed, my guess is that the code cleanup I did so all platforms could share the exact same code as much as possible caused a problem in the result output.

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 21210 - Posted: 5 Sep 2015, 16:52:08 UTC - in response to Message 21196.

I start new ver 1.2 wus come 1by1 every 2 min interval.
It shows 38hour but done in 2 min.
I try 5 and all Validate error. I stop to try new version till you find the problem.

http://boinc.thesonntags.com/collatz/result.php?resultid=21712987
http://boinc.thesonntags.com/collatz/result.php?resultid=21712972
http://boinc.thesonntags.com/collatz/result.php?resultid=21712926
http://boinc.thesonntags.com/collatz/result.php?resultid=21711895
http://boinc.thesonntags.com/collatz/result.php?resultid=21711190


As far as estimates go, BOINC sucks. It tries to be smart and ends up being really dumb. The original sieve WUs were very large. BOINC still hasn't figured out that when the WU size changes, it's estimates should also change. No, it uses the past performance to judge future work and slowly adjusts over time. Secondly, the integer operations per second of the CPU are a really poor way to measure GPU performance so BOINC uses estimated GPU flops using a fancy algorithm that worked for a specific type of GPU for a given model line. As hardware changes, the estimate gets further and further off. Again, K.I.S.S. works really well. The fancier the algorithms, the worse it has gotten. But, it works OK for CPU apps that have variable WU sizes and use floating point computations, so that's what everyone else has to use.

Tip: Always ignore the BOINC time estimate. If you want to believe it, I have a bridge to sell you....

There is a bug in the app output that I need to find. According to the validator, the max steps and total steps are showing up as 0 which is why validation is failing. The results shown in the stderr output ARE valid though. It's just that the result file returned to the server is screwed up.


Post to thread

Message boards : News : Collatz Sieve 1.20 released for Windows, Linux, and OS X


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.