Lunatics Windows Installer v0.38 release notes

Message boards : Number crunching : Lunatics Windows Installer v0.38 release notes
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 11 · Next

AuthorMessage
Profile Miep
Volunteer moderator
Avatar

Send message
Joined: 23 Jul 99
Posts: 2412
Credit: 351,996
RAC: 0
Message 1117796 - Posted: 16 Jun 2011, 12:01:06 UTC

Lunatics Installer Dudes and Dudette are pleased to announce the availibility of the v0.38 installer in the Lunatics main download area
Please take care you use the right installer (32/64 bit) for your system.

We made the effort of writing a README. Available both separately and as part of the install.

Should you have any questions or problems - ask! Post in this thread or open a new one.

New in this installer:
Automatic BOINC client (but not manager!) shutdown and restart.
ATI applications: AP r521, MB r177 [AP r453 hybrid (Brook) for older/smaller cards]
CUDA MB application: x38g_cuda32

ONLY applications for GPU have been added/changed - if you are currently running optimised apps but do NOT have a GPU (either ATI or NVidia) that can crunch you do NOT need this installer

NB We use default values throughout the app_info.xml. If you have previously installed optimised applications and manually edited app_info.xml (e.g. counts for Fermi, cmdline for ATI) you WILL lose those edits on running the installer and have to redo them.


Known issues - general
The installer is supposed to shut down the BOINC client while leaving the Manager running - noticable from the red 'disconnected' dot in the tray. After finishing the BOINC client should restart and the red dot vanish when the Manager reconnects. This make take a few minutes. On some systems however this fails to work. Please ensure BOINC has restarted.
Advanced users only: If you know you are going to edit app_info.xml, it may be easier to shutdown and restart BOINC manually, so you can get the changes into app_info before the restart.

Known issues - ATI apps
Minimum required driver: Something that is OpenCL capable. You can check with GPU-Z (see readme). For your convenience a list:
Catalyst 10.7b with SDK
Catalyst 10.10 APP without SDK
Catalyst 11.3 or higher - if you need to upgrade this should be the easiest, safest option. 11.2 can also be used.

NB using this installer will give you arbitrary default cmdline parameter entries. If you are already running ATI applications: Make a note of what numbers you are using to be able to restore to those values PRIOR to running the installer. This also applies to setting the [count] variable according to your -instances per device variable for multithreading.

The ATI MB application will not work on ATI cards with workgroup size 128 (e.g. HD43xx). To check this run CLinfo, downloadable (at present) from http://boinc.berkeley.edu/dl/clinfo.zip
The Hybrid AP application uses about 80% of a CPU core and does only about 20% of the calculations on the GPU. Parameters have been choosen to reflect this usage. However present BOINC can not reserve '0.8' of a CPU core and therefore reserves none. This means your CPUs will be over committed and runtimes and DCF will suffer. You do increase the overall output though.

Known issues - CUDA app
Requires minimum CUDA 3.2 capable NVidia driver: 263.06 (260.99 on notebooks)

BOINC API has been fixed, newer drivers (270+) can be used.
VRAM requirement has increased from x32f. It should be able to JUST run on a 256 MiB card, provided you free up as much VRAM as possible.
CPU fallback is working again. IF you experience extremely long runtimes or -177 errors, check the stderr for a line reading:
setiathome_CUDA: CUDA runtime ERROR in device memory allocation (Step 1 of 3). Falling back to HOST CPU processing...
or something similar and correct your memory issues if found.
Newer (270+) drivers use more kernel RAM - you may run into memory problems, especially on 256 MiB cards or when multithreading on Fermis or with a lot of cards in one host.
NB with a 256 MiB card you may be forced to stay on 266 drivers until a newer app with reduced memory requirements is available.
Fermi card owners need to decrease the [count] entries in app_info.xml again, should they wish to multithread: Find all instances of [count]1[/count] in your app_info.xml and decrease to 0.5/0.3 to run 2/3 tasks at the same time.
Task validation is geared toward V7 - you may experience more inconclusives in the transition period.

General Notes:
With Windows Vista and Windows 7, do not install BOINC in 'Service' or 'Protected Application Execution' mode for GPU crunching, and do not connect to a Vista/7 computer with Windows Remote Desktop while it is running a GPU application.
Switching users can also cause the GPU app to fail [BOINC versions before 6.12.28]

Tuning cmd line parameters:
For AP: If you get lags or worse driver restarts, decrease unroll. If you run stable you can try to increase it. ffa_block should be a multiple of ffa_block_fetch.
Start low e.g. -ffa_block 4096 -ffa_block_fetch 2048 and try increasing if it runs fine. Sticking to the power of 2 line at first can help.
The will be a system specific best combination - too low slows you down, too high slows you down as well.
Remember AP runtimes depend on blanking % so only compare runtimes on similar blanking %.
For MB: other way round - if you experience lag or driver restarts increase -period_iteration_num

Thanks go to Jason and Raistmer for the applications and to Richard for getting the BOINC stop/restart to work and actually building the installer. You may shout at me, if the app_info is wrong.
I would also like to take this opportunity to thank our active alpha testers (in no particular order and apologies to anybody I've forgotten) Claggy, Ghost, arkayn, Mike, SciManSteve and Vyper, perryjay for reliable beta testing and advertisement as well as Fredericx51 for valuable insight into what it looks like when something went wrong.
Carola
-------
I'm multilingual - I can misunderstand people in several languages!
ID: 1117796 · Report as offensive
Profile Helli_retiered
Volunteer tester
Avatar

Send message
Joined: 15 Dec 99
Posts: 707
Credit: 108,785,585
RAC: 0
Germany
Message 1117812 - Posted: 16 Jun 2011, 12:47:05 UTC

Many Information...

May i ask a question, neverless? Is there a reason why i should update my CUDA App x32f to x38g_cuda32?


Helli
ID: 1117812 · Report as offensive
Profile Miep
Volunteer moderator
Avatar

Send message
Joined: 23 Jul 99
Posts: 2412
Credit: 351,996
RAC: 0
Message 1117815 - Posted: 16 Jun 2011, 13:05:56 UTC - in response to Message 1117812.  
Last modified: 16 Jun 2011, 13:09:18 UTC

May i ask a question, neverless? Is there a reason why i should update my CUDA App x32f to x38g_cuda32?


This thread is specifically for asking questions and reporting problems ;)

a) very likely faster - depends on your system of course. I've got some 10-15%.

b) CUDA 4 compliant BOINC API exit code - won't clash with 270+ NVidia drivers.
c) It's better with VLAR (not that you should be seeing any on the GPU atm)
d) working CPU fallback - mainly something for small slow cards
e) additional benefits I'm not aware of, that Jason might point out.
Carola
-------
I'm multilingual - I can misunderstand people in several languages!
ID: 1117815 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1117816 - Posted: 16 Jun 2011, 13:08:06 UTC - in response to Message 1117812.  

Many Information...

May i ask a question, neverless? Is there a reason why i should update my CUDA App x32f to x38g_cuda32?


Helli

It's quick a bit faster, ;-)

Claggy
ID: 1117816 · Report as offensive
Profile Miep
Volunteer moderator
Avatar

Send message
Joined: 23 Jul 99
Posts: 2412
Credit: 351,996
RAC: 0
Message 1117817 - Posted: 16 Jun 2011, 13:10:51 UTC - in response to Message 1117816.  

Many Information...

May i ask a question, neverless? Is there a reason why i should update my CUDA App x32f to x38g_cuda32?


Helli

It's quick a bit faster, ;-)

Claggy


And that is, as Jason keeps repeating, with only a negligible amount of actual optimisation put into the code.
Carola
-------
I'm multilingual - I can misunderstand people in several languages!
ID: 1117817 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1117823 - Posted: 16 Jun 2011, 13:22:13 UTC - in response to Message 1117818.  

Is it possible to download AP r521 separately somewhere. The version on Lunatics site is r456.

I don't want to risk messing up a working system with the installer, when I can so easily manually install the app.

Download the Installer, then extract the apps from the Installer with 7-Zip,

Claggy
ID: 1117823 · Report as offensive
Profile Miep
Volunteer moderator
Avatar

Send message
Joined: 23 Jul 99
Posts: 2412
Credit: 351,996
RAC: 0
Message 1117827 - Posted: 16 Jun 2011, 13:30:58 UTC - in response to Message 1117823.  

Is it possible to download AP r521 separately somewhere. The version on Lunatics site is r456.

I don't want to risk messing up a working system with the installer, when I can so easily manually install the app.

Download the Installer, then extract the apps from the Installer with 7-Zip,

Claggy


I was wondering if you could do that...

Anyway I've stuck it here
Carola
-------
I'm multilingual - I can misunderstand people in several languages!
ID: 1117827 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117831 - Posted: 16 Jun 2011, 13:42:59 UTC - in response to Message 1117829.  
Last modified: 16 Jun 2011, 13:48:06 UTC

Is it possible to download AP r521 separately somewhere. The version on Lunatics site is r456.

I don't want to risk messing up a working system with the installer, when I can so easily manually install the app.

Download the Installer, then extract the apps from the Installer with 7-Zip,

Claggy


Thanks Claggy, that was easy. I even had 7-Zip already installed :-)

Now to the manual labor of playing with my app_info.

You will find the basic framework for the app_info segment in the file APATI.aistub, also available for extraction from the installer. Remember that you will also need the matching Astropulse_Kernels_r521.cl file, and to update that <file_ref> alongside the main program update.

Edit - you don't, strictly speaking, need the .pdb file that Carola has put in the stand-alone download, although that will be helpful in passing diagnostic information back to Raistmer in the unlikely event that the application crashes.
ID: 1117831 · Report as offensive
Profile Helli_retiered
Volunteer tester
Avatar

Send message
Joined: 15 Dec 99
Posts: 707
Credit: 108,785,585
RAC: 0
Germany
Message 1117832 - Posted: 16 Jun 2011, 13:49:41 UTC

Thanks Claggy and Miep for your fast reply.

That means for me it's not really necessary to switch now. Actually i have no problems with my both
Cuda Rigs, neither with downclocking or anything else. Because my Graphic Cards are only Co-Processor
Cards here ;-) , there is no need for me to upgrade my Graphic Card Driver.

Helli
ID: 1117832 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1117833 - Posted: 16 Jun 2011, 13:51:00 UTC

I had a small issue of BOINC nuking all the CUDA tasks after running the v0.38 installer. However, I think it was probably my own doing and since I backed everything up I didn't actually loose anything.

With the installer I notice the plan class for for the versions are like this which is probably correct.

<version_num>610</version_num>
<plan_class>cuda_fermi</plan_class>
<version_num>609</version_num>
<plan_class>cuda23</plan_class>
<version_num>608</version_num>
<plan_class>cuda</plan_class>

As I had manually added 609 and 610 sections to my old file it looked like this.

<version_num>610</version_num>
<plan_class>cuda</plan_class>
<version_num>609</version_num>
<plan_class>cuda</plan_class>
<version_num>608</version_num>
<plan_class>cuda</plan_class>

Once I saw the difference I changed them all to <plan_class>cuda</plan_class> as I had previously.

Should I let the CUDA tasks run dry and then correct the <plan_class> information, or does it not really matter that much?
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1117833 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117839 - Posted: 16 Jun 2011, 14:04:10 UTC - in response to Message 1117833.  

I had a small issue of BOINC nuking all the CUDA tasks after running the v0.38 installer. However, I think it was probably my own doing and since I backed everything up I didn't actually loose anything.

With the installer I notice the plan class for for the versions are like this which is probably correct.

<version_num>610</version_num>
<plan_class>cuda_fermi</plan_class>
<version_num>609</version_num>
<plan_class>cuda23</plan_class>
<version_num>608</version_num>
<plan_class>cuda</plan_class>

As I had manually added 609 and 610 sections to my old file it looked like this.

<version_num>610</version_num>
<plan_class>cuda</plan_class>
<version_num>609</version_num>
<plan_class>cuda</plan_class>
<version_num>608</version_num>
<plan_class>cuda</plan_class>

Once I saw the difference I changed them all to <plan_class>cuda</plan_class> as I had previously.

Should I let the CUDA tasks run dry and then correct the <plan_class> information, or does it not really matter that much?

It doesn't really matter.

The plan_class elements in the installer have been chosen to match the versions that would be issued with stock (non-optimised) applications. We would always advise people to follow that convention when manually adding elements to app_info.xml

In your situation, here's what I'd do:

Stick with the backup and your current cache of workunits.

Look at the version numbers shown in BOINC Manager for the tasks you're actually running/waiting to run. I expect they will all show 6.10 by now.

Double-up your existing app_version sections, so they read

<version_num>610</version_num>
<plan_class>cuda_fermi</plan_class>
<version_num>610</version_num>
<plan_class>cuda</plan_class>

(in that order)

Do the same for 609 and 608 if you have any tasks for them, but I doubt if it'll be necessary.

Wait a few days. As tasks labelled "6.10 (cuda)" are processed and reported, they should be replaced by tasks labelled "6.10 (cuda_fermi)". Once that process is complete, and all tasks reported, it should be safe to re-run the installer.

That way, you avoid having to make any manual alterations to client_state.xml, which is always a risky process.
ID: 1117839 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117840 - Posted: 16 Jun 2011, 14:08:01 UTC - in response to Message 1117834.  

Thank you Richard, and all other specialist, for being so helpful and so quick in answering questions, and giving support.

Yup, Carola has had to go out on other duties, so I've taken the helm for now. We'll try and ensure that one or other of us is available for tech support in the early stages of the release, though we greatly appreciate the input from others too - it will be nice to get some sleep in due course :-)
ID: 1117840 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117868 - Posted: 16 Jun 2011, 15:12:49 UTC - in response to Message 1117859.  

Houston i have a problem.

i installed lunatics v0.38 and then installed nvidia 275.33 drivers.
Started reporting back completed WU's after 13:08:00 UTC using the new configuration.
For those WU's where others had reported in before me i noticed a preponderance
of 'Completed, validation inconclusive' Status entries.
All were a disagreement in the 'Spike Count'.
i did suspend all of the WU's that would take longer and started processing
some of the short WU's to get more data points.
More of the short WU's seem to be validating, but not all.
GPU temp at 65C. Will drop the OC on the GPU down to 830 from 900, and
if that doesn't help may revert to earlier nvidia drivers.
Any suggestions?

Yes, there is a known but temporary issue to do with inconclusive validations.

As Jason_gee (who wrote the x38g CUDA application inclided in the installer) put it:

About inconclusives with x38 series: Certain kinds of inconclusives against particular kinds of wingman results are expected:

- Some, usually large, number of low power spikes around threshold. (usually against legacy CPU stock or AKv8 )
- lots of triplets closely spaced in the task (Usually against stock GPU, possibly with -12 opt GPU didn't get)
- a tiny variation in 1 or more other signals (against earlier, usually GPU, results)
- disagreeing -9 overflows, between CPU & GPU ( inherited from nVidia code, may or may not be addressed later)
- dodgy wingmen results (of course, e.g. wingman running V12 on a Fermi, or other problems)

The main reasons for this are improvement to accuracy of spikes & the chirp, as well as the better triplet handling we already inherited from Joe's improvement to reduce -12's

The purpose of doing so is quite deliberate, as we introduce code to both GPU & CPU codebases, stock and optimised, that agree better between CPU & GPU going into V7. That means that while there will be initial expected extra inconclusives, going into V7, as more applications adopt the improvements, CPU & GPU apps will agree very tightly making inconclusives somewhat rarer.

So it's a transitional period of sorts, designed to bring everything into better agreement going into V7.

If you see inconclusives that you are not sure fit these expected patterns, then I'd like to know about those for further investigation. As the expected types become understood though, over the testing period the validation character seems already pretty clear.

We're pretty certain that the vast majority of x38g results will validate in the end, and in the longer term (when the improved accuracy is transferred to the stock and other optimised applications, the first-time validation rate will go back to, and even above, what it previously was.

The x38g application is designed to work better with the newer drivers, and to eliminate the down-clocking problem that some users have reported. The 275-series drivers should show a speed improvement as well.

You are right to keep an eye on your temperatures and validation rates, especially when overclocking, with the new apps - the extra processing speed will put a greater strain on them than before. But provided the tasks validate in the end - as they should do, unless you draw a quorum partner with problems of their own - you should have no additional problems.
ID: 1117868 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117872 - Posted: 16 Jun 2011, 15:18:10 UTC - in response to Message 1117860.  

And more questions: To get AP tasks for ATI, is it a must to enable "Send work from other apps" on web preferences page, even if Astropulse v5.05 already is enabled?

Questions, questions :-)

EDIT: added: I rescheduled back to the CPU from the ATI GPU, lots of tasks, just to get Boinc asking for GPU tasks again, which it does, but doesn't get any. It doesn't get any MB GPU tasks, because I disabled MB on the preferences page, and only have AP 5.05 enabled, and not "Send work from other apps". Maybe I need to have "Send work from other apps" enabled too, to make it send tasks for AP ATI, but then I get MB ATI too I guess....

Oh well

AP ATI and MB ATI require different applications and different app_info sections. Whether you are installing the apps by hand (as you are doing), or running the installer itself, you have the choice of selecting none, either or both of these applications.

If you don't install the MB ATI app, BOINC won't ask for that kind of work. As to AP ATI - that depends on the availability of AstroPulse work in general, and as testers have found and reported here, that can be in short supply some of the time.
ID: 1117872 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34351
Credit: 79,922,639
RAC: 80
Germany
Message 1117873 - Posted: 16 Jun 2011, 15:19:28 UTC - in response to Message 1117860.  

And more questions: To get AP tasks for ATI, is it a must to enable "Send work from other apps" on web preferences page, even if Astropulse v5.05 already is enabled?

Questions, questions :-)

EDIT: added: I rescheduled back to the CPU from the ATI GPU, lots of tasks, just to get Boinc asking for GPU tasks again, which it does, but doesn't get any. It doesn't get any MB GPU tasks, because I disabled MB on the preferences page, and only have AP 5.05 enabled, and not "Send work from other apps". Maybe I need to have "Send work from other apps" enabled too, to make it send tasks for AP ATI, but then I get MB ATI too I guess....

Oh well


No its not needed.
I leave it unchecked myself.

Note its hard to get APs atm.




With each crime and every kindness we birth our future.
ID: 1117873 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1117874 - Posted: 16 Jun 2011, 15:22:47 UTC

Great job guys.
So far the first few tasks have been approx. 100 seconds faster on VHARs with driver 267.59; I'll let things run for a few days then maybe update the drivers.

http://setiathome.berkeley.edu/result.php?resultid=1949057328
http://setiathome.berkeley.edu/result.php?resultid=1949065826
</Tazz>
ID: 1117874 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117876 - Posted: 16 Jun 2011, 15:24:41 UTC - in response to Message 1117873.  

Note its hard to get APs atm.

8,567 ready to send ;-)

They should start reaching the head of the queue soon.
ID: 1117876 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1117884 - Posted: 16 Jun 2011, 15:34:02 UTC - in response to Message 1117860.  

YOu'll get them eventually. I think the shorty storm has hindered the production of AP WU's


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1117884 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14674
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1117887 - Posted: 16 Jun 2011, 15:39:30 UTC - in response to Message 1117879.  

Then again, when Boinc asks for GPU tasks, it doesn't tell what kind of GPU tasks it wants, if it is as in my case since I have installed both ATI GPU apps, ATI MB it asks for, or ATI AP.

BOINC will actually just ask for "work for ATI GPU" - it isn't fussy. It's up to the project to work out which type of work is available, MB or AP, and assign it to the GPU in response to the request.
ID: 1117887 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1117890 - Posted: 16 Jun 2011, 15:42:59 UTC - in response to Message 1117888.  

You sound like a nervous father to be


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1117890 · Report as offensive
1 · 2 · 3 · 4 . . . 11 · Next

Message boards : Number crunching : Lunatics Windows Installer v0.38 release notes


 
©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.