Sieve in Production; Large and Solo deprecated
log in

Advanced search

Message boards : News : Sieve in Production; Large and Solo deprecated

1 · 2 · 3 · 4 . . . 5 · Next
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: 2
Message 21330 - Posted: 23 Sep 2015, 18:48:17 UTC

The collatz_sieve applications have been changed from beta to production status so all volunteers will now be getting them. I also increased the sieve workunit size by 4 times. I expect it will take BOINC a while to re-adjust it's runtime estimates as they are based on floating point calculations which aren't used in Collatz except to show the percent complete.

I deprecated the solo_collatz and large_collatz applications and disabled the associated work generators. Any that are in progress will still be validated, you just won't get any new solo or large WUs since you will receive sieve WUs instead.

If the sieve application fails immediately after it starts, you probably need to install the latest Microsoft Visual C++ runtime and, since BOINC will send your computer both 32 and 64 bit applications, you will need to install both the 32 and 64 but C++ runtime (see main page for a link or just Google it).

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: 2
Message 21331 - Posted: 23 Sep 2015, 20:31:33 UTC

BOINC uses shared memory to store the IDs of the workunits that have been generated but not yet sent. Given the number of platforms times the number of plan classes times the number of applications, the result is that just a few of each WU available for a given platform, plan class, and application. While I can control the weight of each application, I can't control the way BOINC allocates per plan class or platform. Fewer apps, platforms, and plan classes means more WUs available per app, platform, and plan class.

That's why the solo and large work generators are now shut down. The mini collatz will soon follow. Once they are deprecated, I can add a larger sieve app to keep fast GPUs busy for a hour or two rather than just a few minutes.

The Android ARM devices are much slower than GPUs (e.g. a single GPU does the work of hundreds of Android ARM devices) and I think they will need to use a much smaller sieve size which means they will run even slower than they do now compared to the GPUs. So, I'm torn between dropping the micro app completely vs adding a micro sieve app for Android devices.

Gator 1-3
Send message
Joined: 15 Sep 15
Posts: 4
Credit: 1,025,209,621
RAC: 365,859
Message 21350 - Posted: 28 Sep 2015, 19:30:17 UTC

I don't know if this is true of all 32-bit machines, but mine will not allow me to update the C++ for 64-bit. I can update 32-bit with no problem, but the 64-bit gets an error message and closes when I try to run it on my 32-bit systems.

On the 32-bit systems, when I open it up to downloading sieve tasks, all of them fail as soon as they start. I'm assuming that's because the tasks are all 64-bit tasks.

2 questions: is there a way to get the 64-bit C++ to update on 32-bit systems? If the answer to that question is no, then is there a way to get only 32-bit tasks for my 32-bit machines, or am I just going to have to deal with having a ton of tasks that error just to get the occasional 32-bit task?

Thanks

Profile BarryAZ
Send message
Joined: 21 Aug 09
Posts: 251
Credit: 13,170,943,836
RAC: 23,587,350
Message 21356 - Posted: 29 Sep 2015, 19:07:17 UTC

Hmm -- it appears with this change that my eight year run with Collatz will be at an end --- might be a function of older hardware (running ATI 4850's for Collatz).

One workstation is an old reliable XP 32. The other a Windows 7 64.
____________

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: 2
Message 21358 - Posted: 30 Sep 2015, 2:27:29 UTC - in response to Message 21356.

Hmm -- it appears with this change that my eight year run with Collatz will be at an end --- might be a function of older hardware (running ATI 4850's for Collatz).

One workstation is an old reliable XP 32. The other a Windows 7 64.


An HD 4850 should be able to run the OpenCL app. The original AMD OpenCL 1.0 release even supported HD 3870 GPUs. Collatz doesn't use any features from newer OpenCL versions, so it should run OK. Does BOINC recognize it runs OpenCL? Does it request OpenCL tasks?

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: 2
Message 21359 - Posted: 30 Sep 2015, 3:39:51 UTC - in response to Message 21350.

I don't know if this is true of all 32-bit machines, but mine will not allow me to update the C++ for 64-bit. I can update 32-bit with no problem, but the 64-bit gets an error message and closes when I try to run it on my 32-bit systems.

On the 32-bit systems, when I open it up to downloading sieve tasks, all of them fail as soon as they start. I'm assuming that's because the tasks are all 64-bit tasks.

2 questions: is there a way to get the 64-bit C++ to update on 32-bit systems? If the answer to that question is no, then is there a way to get only 32-bit tasks for my 32-bit machines, or am I just going to have to deal with having a ton of tasks that error just to get the occasional 32-bit task?

Thanks


If it is a 32-bit machine, it will only get 32-bit apps. In the app log you will see "Collatz Conjecture Sieve 1.21 Windows i686 for OpenCL" and the i686 means 32-bit. It would say "x86_64" if it was 64-bit.

That it reports error -11 when it compiles the kernel (a.k.a. clBuildProgram) would imply that the driver can't compile the kernel but it isn't giving an error. I'll see if there are any suggested fixes for that as having no build log makes it a little tough to figure out what's going wrong.

Profile BarryAZ
Send message
Joined: 21 Aug 09
Posts: 251
Credit: 13,170,943,836
RAC: 23,587,350
Message 21363 - Posted: 30 Sep 2015, 5:32:31 UTC - in response to Message 21359.

It can get the tasks -- they just immediate go computation error on the XP system

On the other systems (which are Win 7 64 bit with an HD 7750 -- that gets no tasks at all), and an HD 6750 -- that one does get tasks I will see if it handles them).

On those Windows 7 64 bit systems, I have the latest BOINC and the latest AMD drivers deployed.

On the Windows XP system I did try resetting Collatz, I also did the delete project and rejoin -- same thing computation error city.
____________

Profile Conan
Avatar
Send message
Joined: 17 Jun 09
Posts: 40
Credit: 110,613,769
RAC: 0
Message 21364 - Posted: 30 Sep 2015, 11:36:36 UTC
Last modified: 30 Sep 2015, 11:42:18 UTC

The Sieve tasks are running fine on my 32 Bit XP computer using the CPU (I am not running any GPUs currently).

The 32 Bit application also only takes a third of the time of my 64 Bit CPU Linux applications, not sure why that is.
Is the Linux 64 Bit CPU application just not optimised like the Windows CPU application?

Conan
____________

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: 2
Message 21367 - Posted: 30 Sep 2015, 15:24:08 UTC - in response to Message 21364.

The Sieve tasks are running fine on my 32 Bit XP computer using the CPU (I am not running any GPUs currently).

The 32 Bit application also only takes a third of the time of my 64 Bit CPU Linux applications, not sure why that is.
Is the Linux 64 Bit CPU application just not optimised like the Windows CPU application?

Conan


Thanks for reporting it. It really should be about the same performance as the Windows version and the 64-bit app should be faster since using 64 bit integers requires half the calculations as when using 32-bit integers. So either it is executing the 32-bit code instead of the 64-bit code or optimization isn't enabled (-O3).

Profile BarryAZ
Send message
Joined: 21 Aug 09
Posts: 251
Credit: 13,170,943,836
RAC: 23,587,350
Message 21368 - Posted: 30 Sep 2015, 16:01:46 UTC - in response to Message 21367.

The Windows 7 64 bit system with an HD 6750 is getting and processing the GPU units -- that system seems to be working fine.

I am not sure what's amiss with the Windows 7 64 bit system with an HD 7850 -- it simply isn't getting units -- it requests them and gets none.
____________

Profile BarryAZ
Send message
Joined: 21 Aug 09
Posts: 251
Credit: 13,170,943,836
RAC: 23,587,350
Message 21369 - Posted: 30 Sep 2015, 16:07:41 UTC - in response to Message 21368.

OK -- user error regarding the HD 7750 -- my default settings in Collatz (as opposed to the subgroups) was not configured to receive the sieve units -- that's fixed.

Gator 1-3
Send message
Joined: 15 Sep 15
Posts: 4
Credit: 1,025,209,621
RAC: 365,859
Message 21374 - Posted: 30 Sep 2015, 19:22:00 UTC

I had to update a program I use on my 32 bit system, which required a reboot. I don't know if rebooting is what did the trick, but I can now process a sieve workunit without getting an error message. It may have been that updating the C++ runtime needed to reboot to take effect and I missed it.

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: 2
Message 21375 - Posted: 30 Sep 2015, 19:23:25 UTC - in response to Message 21364.

The Sieve tasks are running fine on my 32 Bit XP computer using the CPU (I am not running any GPUs currently).

The 32 Bit application also only takes a third of the time of my 64 Bit CPU Linux applications, not sure why that is.
Is the Linux 64 Bit CPU application just not optimised like the Windows CPU application?

Conan


The 1.20 version for Linux x86_64 did not have the optimization flag set in the makefile. Version 1.21 should be considerably faster.

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: 2
Message 21376 - Posted: 30 Sep 2015, 19:25:27 UTC - in response to Message 21374.

I had to update a program I use on my 32 bit system, which required a reboot. I don't know if rebooting is what did the trick, but I can now process a sieve workunit without getting an error message. It may have been that updating the C++ runtime needed to reboot to take effect and I missed it.


That may be possible. If some other application was using the previous C++ runtime the new DLL couldn't be loaded until the other app was closed.

waffleironhead
Send message
Joined: 17 Jan 10
Posts: 1
Credit: 38,241,339
RAC: 6,158
Message 21382 - Posted: 2 Oct 2015, 2:22:40 UTC

So with this release, no more CAL/Brook+ anymore? It getting colder here up north so I figured I'd attach my AMD hd4200 igp.

JcJet
Send message
Joined: 9 Dec 09
Posts: 1
Credit: 18,344,928
RAC: 0
Message 21390 - Posted: 2 Oct 2015, 16:34:35 UTC

Installed both runtimes from thme main page ling, and it's still fails after start. Oh well, will check back a few months later if you actually finish it...

Szymon Jaranowski
Send message
Joined: 5 Feb 14
Posts: 4
Credit: 12,645,423
RAC: 0
Message 21391 - Posted: 2 Oct 2015, 18:42:04 UTC

Is there any chance of seeing new production applications on Brook+/CAL (ati14)? I'm a long-term user of Radeon HD 3650.

klepel
Send message
Joined: 2 Mar 12
Posts: 1
Credit: 24,707,188
RAC: 0
Message 21393 - Posted: 3 Oct 2015, 1:02:34 UTC

Slicker,
Me too! I started the whole Collatz Conjecture thing for the CAL/Brook+ AMD hd4200 igp application - I PMed you to start it years ago… And I always used it, as a fun thing to run besides GPUGRID on the power full GPs, (Sorry to say it) I will only run Collatz Conjecture as a second back-up application.

So please - please don’t let it die!

As well I need a short version of the Sieve Program (Same length as Mini) as I used the Mini application to run it on my Laptop during charging the battery overnight. With you actual Sieve length I will never finish on time and I am not willing to run my Laptop on a 24 hours / 7 days.

As well as with the Micro application: I have not used it on my Smartphone, but I still think, this application on Android and Raspberry PI is worthwhile to maintain, as it is available for another group of users: students, pupils, which might not run BOINC on a computer (farm) but might run it on their smartphone or Raspberry PI as a new little program to test it and addicted them to BOINC. I ran it on a USB Memo-stick with Puppy Linux…

And one last thought, you might as well harvest all the old computer parts which are not of much use on other projects like GPUGRID, but still give descend points on yours. I personally tried to install LINUX Mint on an old PC with a GTX 9800, just to use it for Einstein and for Collatz, and after months of trying I succeeded! I would not have bought a new computer or motherboard but I am willing to use these parts on a project like Collatz.

Hope we are changing your mind!

Gordon Lack
Send message
Joined: 14 Apr 12
Posts: 12
Credit: 5,021,830
RAC: 1,379
Message 21395 - Posted: 3 Oct 2015, 13:22:29 UTC - in response to Message 21330.

I deprecated the solo_collatz and large_collatz applications
So I enabled sieve on my Linux system with an AMD 6670 GPU (which is a single-precision GPU).

If the sieve application fails immediately after it starts...
It did, but this being Linux it's not a MS Vidual C++ runtime issue...

This is the error log:


<core_client_version>7.2.42</core_client_version>
<![CDATA[
<message>
process exited with code 1 (0x1, -255)
</message>
<stderr_txt>
Collatz Conjecture Sieve 1.21 Linux 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
LLVM ERROR: Cannot select: 0x18db810: i32 = setcc 0x1697ab0, 0x16960a0, 0x18db710 [ORD=14] [ID=50]
0x1697ab0: i64 = add 0x16960a0, 0x1696da0 [ORD=10] [ID=46]
0x16960a0: i64 = AMDILISD::VEXTRACT 0x1390ae0, 0x13901e0 [ORD=9] [ID=36]
0x1390ae0: v2i64,ch = CopyFromReg 0xf39610, 0x13909e0 [ORD=9] [ID=30]
0x13909e0: v2i64 = Register %vreg38 [ORD=9] [ID=2]
0x13901e0: i32 = TargetConstant<1> [ORD=2] [ID=21]
0x1696da0: i64 = add 0x1696aa0, 0x1696ca0 [ORD=8] [ID=45]
0x1696aa0: i64,ch = load 0xf39610, 0x16965a0, 0x16966a0<LD4[%arrayidx], zext from i32> [ORD=5] [ID=44]
0x16965a0: i32 = add 0x13904e0, 0x16964a0 [ORD=3] [ID=43]
0x13904e0: i32,ch = CopyFromReg 0xf39610, 0x13903e0 [ORD=3] [ID=29]
0x13903e0: i32 = Register %vreg35 [ORD=3] [ID=1]
0x16964a0: i32 = shl 0x18db210, 0x16963a0 [ORD=3] [ID=42]
0x18db210: i32 = AMDILISD::VEXTRACT 0x1390ee0, 0x13901e0 [ORD=2] [ID=38]
0x1390ee0: v4i32,ch = llvm.AMDIL.get.global.id 0xf39610, 0x1390de0 [ORD=1] [ID=32]
0x1390de0: i32 = TargetConstant<2580> [ORD=1] [ID=4]
0x13901e0: i32 = TargetConstant<1> [ORD=2] [ID=21]
0x16963a0: i32 = Constant<2> [ORD=3] [ID=7]
0x16966a0: i32 = undef [ORD=4] [ID=8]
0x1696ca0: i64 = shl 0x16969a0, 0x1696ba0 [ORD=7] [ID=40]
0x16969a0: i64 = zero_extend 0x1390ce0 [ORD=6] [ID=37]
0x1390ce0: i32,ch = CopyFromReg 0xf39610, 0x1390be0 [ORD=6] [ID=31]
0x1390be0: i32 = Register %vreg39 [ORD=6] [ID=3]
0x1696ba0: i32 = Constant<26> [ORD=7] [ID=9]
0x16960a0: i64 = AMDILISD::VEXTRACT 0x1390ae0, 0x13901e0 [ORD=9] [ID=36]
0x1390ae0: v2i64,ch = CopyFromReg 0xf39610, 0x13909e0 [ORD=9] [ID=30]
0x13909e0: v2i64 = Register %vreg38 [ORD=9] [ID=2]
0x13901e0: i32 = TargetConstant<1> [ORD=2] [ID=21]

</stderr_txt>
]]>
Any ideas? the solo_collatz ran without issue.

Profile IamBill46
Send message
Joined: 1 Jul 14
Posts: 1
Credit: 8,454,927
RAC: 0
Message 21396 - Posted: 3 Oct 2015, 15:12:56 UTC

I am getting these errors (running Windows 10, 64-bit, desktop:

PROJECT PROGRESS Status
Collatz Conjecture 100.000% Computation Error (0.873 CPUs + 1 NVIDIA GPU)

Application
Collatz Sieve 1.21 (opencl_nvidia_gpu)


Any app running 1.20 is running OK.

Iambill46


[/img]

1 · 2 · 3 · 4 . . . 5 · Next
Post to thread

Message boards : News : Sieve in Production; Large and Solo deprecated


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.