Message boards :
Number crunching :
Multi core greater than 80 core
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 · Next
Author | Message |
---|---|
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
Same deal with that version: Are you eventually logged in using some sort of remote desktop? The client and apps must have access to the same session the graphics is running in. Peter |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
"use all coprocessors" will work only after the line "No usable GPUs found" disappears! Did you try GPU-Z? Post the contents of cc_config.xml  - ALF - "Find out what you don't do well ..... then don't do it!" :)  |
-BeNt- Send message Joined: 17 Oct 99 Posts: 1234 Credit: 10,116,112 RAC: 0 |
Same deal with that version: The only thing that causes that is RDP in windows. If you are logging in via a remote client use one of the VNC's. I prefer TightVNC runs as fast as RDP in windows. I know some use GoToMyPC and things like that. Traveling through space at ~67,000mph! |
MarkJ Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5 |
I raised the issue of 160 cores on the BOINC Alpha mailing list. It appears that due to the way Windows handles things when it has more than 64 cpus, BOINC doesn't quite work. To quote David's email on the list: It turns out Windows' support for >64 cores is funky, as usual. A host's cores are divided into "processor groups" of at most 64 each (presumably because of 64-bit bitmaps for affinity etc.). Calls like GetSystemInfo() return info only about the calling process's processor group. There was a follow up email suggesting a work around for the GetSystemInfo(), well thats what I interpreted it to be. No word as to what version of the BOINC client might support this, but at least they know what needs to be done. BOINC blog |
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
-BeNt- wrote: Pepo wrote:Bry B wrote:3/10/2011 6:34:46 PM | | GPUs have become unusable; disabling tasks RDP is not the only know reason for GPU to become unavailable for BOINC. The same happens when multiple users log in on a desktop, and the one running BOINC has her/his session locked. I know that mainframes and servers are mostly accessed and used by other means than just a direct cables to graphics card and mouse/keyboard connector. A plain physically connected KVM switch with remote connection over network etc. would possibly not be a problem, but any direct logical connection could possibly behave as some sort of RDP and render the GPUs unusable for BOINC. MarkJ wrote: I raised the issue of 160 cores on the BOINC Alpha mailing list. It appears that due to the way Windows handles things when it has more than 64 cpus, BOINC doesn't quite work. The work-around addressed just obtaining the total number of processors. The question remains, whether the client process will be able to launch task child processes on other processor groups, or will be restricted to its initial processor group, without doing some addidional work with (sort-of) group affinities. Peter |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I raised the issue of 160 cores on the BOINC Alpha mailing list. It appears that due to the way Windows handles things when it has more than 64 cpus, BOINC doesn't quite work. Have they looked into GetNativeSystemInfo() to see if it has the same limitation they are currently seeing with GetSystemInfo()? SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
Have they looked into GetNativeSystemInfo() to see if it has the same limitation they are currently seeing with GetSystemInfo()? Should not be necessary for a native 64-bit client: MSDN wrote: GetNativeSystemInfo Function And Bryan has installed "client version 6.12.18 for windows_x86_64". But I'm shortly expecting 6.12.19. Bryan will tell... Peter |
Bry B Send message Joined: 3 Apr 99 Posts: 53 Credit: 832,165 RAC: 0 |
yes I am RDPing into the system. I will try to log on locally. Another problem is that the app keeps crashing. I have the debugger up but dont have symbols for your app. If anyone wants to look at it and knows how to use KD I can set it up. This could be related to my memory risers which I will be replacing today. |
-BeNt- Send message Joined: 17 Oct 99 Posts: 1234 Credit: 10,116,112 RAC: 0 |
yes I am RDPing into the system. I will try to log on locally. Another problem is that the app keeps crashing. I have the debugger up but dont have symbols for your app. If anyone wants to look at it and knows how to use KD I can set it up. This could be related to my memory risers which I will be replacing today. Yeah figured that was probably what was going on, that's the most common cause noted here causing for the gpu's to become unusable. Traveling through space at ~67,000mph! |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
|
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
Can we dumb this down a little and just give a one-liner on why those of use with 40 physical cores (and 80 threads) may only be seeing 20 in our account's "Computer Information" pages Example of 40 core machine: http://setiathome.berkeley.edu/show_host_detail.php?hostid=5848928 Thanks :) |
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
Can we dumb this down a little and just give a one-liner on why those of use with 40 physical cores (and 80 threads) may only be seeing 20 in our account's "Computer Information" pages Could you please give the 6.12.26 a try? (But... sorry, I've just noticed, that the changeset [trac]changeset:23215[/trac] (which was expected to solve this issue) was not yet ported to the 6.12 line)-: Peter |
ML1 Send message Joined: 25 Nov 01 Posts: 20265 Credit: 7,508,002 RAC: 20 |
Can we dumb this down a little and just give a one-liner on why those of use with 40 physical cores (and 80 threads) may only be seeing 20 in our account's "Computer Information" pages It is a "quirk"/"feature"/"limit" of how the Windows kernel has been written to group together processing threads to help scheduling. From earlier in this thread, Dr A is working on or already has put in a fix into Boinc to work around the feature for Windows. If you were to try for example Linux[*], you would not see the problem (unless that is you had more than 1024[**] processors!). The Linux schedulers do not have such limits to worry about. Happy fast parallel crunchin', Martin [*] And before the Windows fan-boys jump in to supposedly defend their 'very best OS that ever can exist': Yes... If that machine is being used in a corporate Windows-only (locked in) environment, then Linux is 'only good' as a test case. Unless that is, there is going to be a migration over to Linux in any case... [**] If so, you'd already be using Linux tweaked for whatever number of processors for whatever special system you were running. See new freedom: Mageia Linux Take a look for yourself: Linux Format The Future is what We all make IT (GPLv3) |
hbomber Send message Joined: 2 May 01 Posts: 437 Credit: 50,852,854 RAC: 0 |
Its matter of choice will Windows group running threads on one processor group or not. There are ways for proper enumeration of NUMA nodes and logical processor count per node. BOINC is not group aware application as it seems. It's not Windows "problem". |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Can we dumb this down a little and just give a one-liner on why those of use with 40 physical cores (and 80 threads) may only be seeing 20 in our account's "Computer Information" pages The work around is in this post Claggy |
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
The work around is in this post Yes, this works reliably. But the cleanest way would still be to finally apply both changesets [trac]changeset:23214[/trac]+[trac]changeset:23215[/trac] onto the 6.12 line ;-( Peter |
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
And where exactly does this magical cc_config.xml file live? I cannot find a single copy of it anywhere on my system. Or is it a file that needs to be created? And if so, where does this file need to be placed? Edit: For what it's worth...I'm not a developer. I'm a validation engineer in a lab full of Dell R900 and R910 servers. |
Pepo Send message Joined: 5 Aug 99 Posts: 308 Credit: 418,019 RAC: 0 |
And where exactly does this magical cc_config.xml file live? I cannot find a single copy of it anywhere on my system. Or is it a file that needs to be created? And if so, where does this file need to be placed? If you open your BOINC Messages log, at the very beginning (among first few lines) you will see something like 27.04.2011 17:37:34 | | Data directory: C:\ProgramData\BOINC This is the folder where your cc_config.xml file lives its magical life. And yes, it is being manually created and maintained (until maybe the 6.13 line will start to understand the get_cc_config + set_cc_config GUI RPCs). Peter |
gcpeters Send message Joined: 20 May 99 Posts: 67 Credit: 109,352,237 RAC: 1 |
I created the xml file and performed an update from the BOINC manager. My account list of computers still shows this system only having 20 cores. Thoughts? |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
I created the xml file and performed an update from the BOINC manager. My account list of computers still shows this system only having 20 cores. Thoughts? Do you get the following when you do the read config file? (or restart Boinc): 05/05/2011 19:15:00 Number of usable CPUs has changed from 2 to 4. Running benchmarks. 05/05/2011 19:15:01 Running CPU benchmarks 05/05/2011 19:15:01 Suspending computation - running CPU benchmarks 05/05/2011 19:15:01 SETI@home Beta Test [cpu_sched] Preempting 07mr11ah.5998.9474.206158430213.14.45_1 (left in memory) 05/05/2011 19:15:01 SETI@home [cpu_sched] Preempting ap_03dc10ab_B4_P0_00237_20110420_24663.wu_1 (left in memory) 05/05/2011 19:15:01 SETI@home Beta Test [cpu_sched] Preempting 07mr11ah.5998.9474.206158430213.14.219_0 (left in memory) 05/05/2011 19:15:01 SETI@home [cpu_sched] Preempting 18no10ac.10985.18472.6.10.41_0 (left in memory) 05/05/2011 19:15:33 Benchmark results: 05/05/2011 19:15:33 Number of CPUs: 4 05/05/2011 19:15:33 4021 floating point MIPS (Whetstone) per CPU 05/05/2011 19:15:33 11939 integer MIPS (Dhrystone) per CPU 05/05/2011 19:15:33 [dcf] scaling all duration correction factors by 0.973263 05/05/2011 19:15:33 SETI@home Finished download of 11mr07ac.26889.154854.6.10.207 05/05/2011 19:15:33 Resuming computation and does Boinc start 20 tasks? (Info on configuring Boinc is on the Client configuration page of the Boinc Wiki) Claggy |
©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.