Multi core greater than 80 core

Message boards : Number crunching : Multi core greater than 80 core
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 8 · Next

AuthorMessage
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085519 - Posted: 10 Mar 2011, 1:50:40 UTC - in response to Message 1085509.  


If BOINC is using GetSystemInfo() function to find the number of logical processors in the computer
I think BOINC will see no more than 64 CPUs.

I hope the workaround will be to set in cc_config.xml
   <options>
      <ncpus>160</ncpus>
   </options>


(this was proposed already in the second post in this thread)

I'm not sure will the 160 running tasks be properly assigned to different logical CPUs or will run in just a group of 64 CPUs.



That just tells BOINC to run that number of instances of the science app IIRC. Windows SHOULD take care of assigning resources correctly.

EDIT: ncpus defiantly works
3/9/2011 8:24:57 PM Processor: 256 GenuineIntel Intel(R) Core(TM) i7 CPU 860 @ 2.80GHz [Family 6 Model 30 Stepping 5]
Also if you do not have 256 cpu's make sure you suspend running work before setting that & running BOINC. As trying to run 16 threads per cpu on a wee i7 with only 4GB of ram isn't a good thing. lol
3/9/2011 8:29:23 PM Number of usable CPUs has changed from 256 to 8. Running benchmarks. Phew!



I did just fire up with Superdome 2. Which is 64 socket and 256LP IA-64. I am running about 12% resources and its not scaling so far like the smaller 4 socket.



ID: 1085519 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085520 - Posted: 10 Mar 2011, 1:51:32 UTC - in response to Message 1085519.  

it thinks its a 32 CPU. LOL
ID: 1085520 · 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 1085522 - Posted: 10 Mar 2011, 1:57:46 UTC - in response to Message 1085519.  
Last modified: 10 Mar 2011, 1:58:27 UTC


If BOINC is using GetSystemInfo() function to find the number of logical processors in the computer
I think BOINC will see no more than 64 CPUs.

I hope the workaround will be to set in cc_config.xml
   <options>
      <ncpus>160</ncpus>
   </options>


(this was proposed already in the second post in this thread)

I'm not sure will the 160 running tasks be properly assigned to different logical CPUs or will run in just a group of 64 CPUs.



That just tells BOINC to run that number of instances of the science app IIRC. Windows SHOULD take care of assigning resources correctly.

EDIT: ncpus defiantly works
3/9/2011 8:24:57 PM Processor: 256 GenuineIntel Intel(R) Core(TM) i7 CPU 860 @ 2.80GHz [Family 6 Model 30 Stepping 5]
Also if you do not have 256 cpu's make sure you suspend running work before setting that & running BOINC. As trying to run 16 threads per cpu on a wee i7 with only 4GB of ram isn't a good thing. lol
3/9/2011 8:29:23 PM Number of usable CPUs has changed from 256 to 8. Running benchmarks. Phew!



I did just fire up with Superdome 2. Which is 64 socket and 256LP IA-64. I am running about 12% resources and its not scaling so far like the smaller 4 socket.


[ img ]http://i251.photobucket.com/albums/gg287/bryananddaun/superdome2.jpg[ /img ]


Can you try <ncpus>128</ncpus> in a cc_config.xml and see if it uses 128 of the processors? I have a feeling that there might be limits set per user on the servers you are using.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1085522 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085523 - Posted: 10 Mar 2011, 1:59:46 UTC - in response to Message 1085491.  


If BOINC is using GetSystemInfo() function to find the number of logical processors in the computer
I think BOINC will see no more than 64 CPUs.

I hope the workaround will be to set in cc_config.xml
   <options>
      <ncpus>160</ncpus>
   </options>


(this was proposed already in the second post in this thread)

I'm not sure will the 160 running tasks be properly assigned to different logical CPUs or will run in just a group of 64 CPUs.




Thanks, I set this on the 256 way. lets see what happens.

3/9/2011 5:58:36 PM SETI@home suspended by user
3/9/2011 5:58:41 PM Re-reading cc_config.xml
3/9/2011 5:58:41 PM Re-read config file
3/9/2011 5:58:41 PM Config: use at most 256 CPUs
3/9/2011 5:58:41 PM log flags: file_xfer, sched_ops, task, sched_op_debug
3/9/2011 5:58:41 PM Number of usable CPUs has changed from 32 to 256. Running benchmarks.
3/9/2011 5:58:41 PM Running CPU benchmarks
3/9/2011 5:58:42 PM Suspending computation - running CPU benchmarks
3/9/2011 5:58:55 PM Re-reading cc_config.xml
3/9/2011 5:58:55 PM Re-read config file
3/9/2011 5:58:55 PM Config: use at most 256 CPUs
3/9/2011 5:58:55 PM log flags: file_xfer, sched_ops, task, sched_op_debug

ID: 1085523 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085525 - Posted: 10 Mar 2011, 2:03:23 UTC - in response to Message 1085523.  

[quote]
If BOINC is using GetSystemInfo() function to find the number of logical processors in the computer
I think BOINC will see no more than 64 CPUs.

I hope the workaround will be to set in cc_config.xml
   <options>
      <ncpus>160</ncpus>
   </options>


(this was proposed already in the second post in this thread)

I'm not sure will the 160 running tasks be properly assigned to different logical CPUs or will run in just a group of 64 CPUs.




That helped, went from 12 to 26 on the 256 way. Ill try it on the X64 160 now.
ID: 1085525 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085528 - Posted: 10 Mar 2011, 2:11:43 UTC - in response to Message 1085525.  

[quote]
If BOINC is using GetSystemInfo() function to find the number of logical processors in the computer
I think BOINC will see no more than 64 CPUs.

I hope the workaround will be to set in cc_config.xml
   <options>
      <ncpus>160</ncpus>
   </options>


(this was proposed already in the second post in this thread)

I'm not sure will the 160 running tasks be properly assigned to different logical CPUs or will run in just a group of 64 CPUs.




That helped, went from 12 to 26 on the 256 way. Ill try it on the X64 160 now.




Actually that really helped!! Thanks.. We are aproaching 60% on the 256way IA-64 and 90%+ on the 160LP X64.


ID: 1085528 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085533 - Posted: 10 Mar 2011, 2:24:02 UTC - in response to Message 1085528.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.



ID: 1085533 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085534 - Posted: 10 Mar 2011, 2:26:55 UTC - in response to Message 1085533.  


ID: 1085534 · 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 1085536 - Posted: 10 Mar 2011, 2:27:54 UTC - in response to Message 1085533.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.



For the X64 160 LP box did you use <ncpus>160</ncpus> or a lower value?


SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1085536 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085550 - Posted: 10 Mar 2011, 2:59:37 UTC - in response to Message 1085536.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.



For the X64 160 LP box did you use <ncpus>160</ncpus> or a lower value?




I used the 160, its stable at 90-92% now
ID: 1085550 · Report as offensive
Profile soft^spirit
Avatar

Send message
Joined: 18 May 99
Posts: 6497
Credit: 34,134,168
RAC: 0
United States
Message 1085565 - Posted: 10 Mar 2011, 3:37:02 UTC - in response to Message 1085550.  

It will be really interesting to see how this experiment turns out Bryan.

We have many people running multiple core graphics cards to great success,
if you put 512 Shaders(graphics cores) together on 2-3 tasks they have some pretty amazing results. It will be quite interesting to see if your configuration gives them a run for the money.


Janice
ID: 1085565 · Report as offensive
Profile Lint trap

Send message
Joined: 30 May 03
Posts: 871
Credit: 28,092,319
RAC: 0
United States
Message 1085566 - Posted: 10 Mar 2011, 3:39:36 UTC
Last modified: 10 Mar 2011, 4:11:11 UTC

Something that hasn't been mentioned yet:

The guys here are talking about CPU scaling for a Dell server, which led to the power management settings:

http://serverfault.com/questions/196301/disable-cpu-scaling-in-windows-server-2008-r2

and on that page below the blue banner, they mention HP's default setting ignores the OS setting.

Martin


edited ... Why ignore the OS settings? The Dell server setting (see above) was also preventing the expected server performance.
ID: 1085566 · Report as offensive
-BeNt-
Avatar

Send message
Joined: 17 Oct 99
Posts: 1234
Credit: 10,116,112
RAC: 0
United States
Message 1085576 - Posted: 10 Mar 2011, 4:13:52 UTC

Interesting in deed. 64 threads on the cpu alone should prove to be a monster cruncher! Then throwing in some gpu's on top of that, it would be interesting to see how it stacks up against a quad gpu machine with dual processors for sure.
Traveling through space at ~67,000mph!
ID: 1085576 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1085579 - Posted: 10 Mar 2011, 4:19:50 UTC - in response to Message 1085576.  

Most likely to keep heat, energy costs and reliability in check.

Cheers.

ID: 1085579 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085592 - Posted: 10 Mar 2011, 4:48:29 UTC - in response to Message 1085565.  

It will be really interesting to see how this experiment turns out Bryan.

We have many people running multiple core graphics cards to great success,
if you put 512 Shaders(graphics cores) together on 2-3 tasks they have some pretty amazing results. It will be quite interesting to see if your configuration gives them a run for the money.



Thanks! I know on the IA-64 side I probably wont be able to find a performance GPU driver so I will probably only be able to add them to the X64 side. If all goes well Ill try it tomorrow.
ID: 1085592 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085593 - Posted: 10 Mar 2011, 4:58:44 UTC - in response to Message 1085566.  

Something that hasn't been mentioned yet:

The guys here are talking about CPU scaling for a Dell server, which led to the power management settings:

http://serverfault.com/questions/196301/disable-cpu-scaling-in-windows-server-2008-r2

and on that page below the blue banner, they mention HP's default setting ignores the OS setting.

Martin


edited ... Why ignore the OS settings? The Dell server setting (see above) was also preventing the expected server performance.


They are dealing with Nehalem CPU's For IA-64 this wont apply but for my X64 160P system it does. Since I work for the company in question (HP) I have all the proper settings for ACPI, Power, C-states, K groups, X2APIC etc.. Spent several months on these with our firmware and perforance guys. Great catch thanks for posting!!
ID: 1085593 · Report as offensive
Profile ML1
Volunteer moderator
Volunteer tester

Send message
Joined: 25 Nov 01
Posts: 20265
Credit: 7,508,002
RAC: 20
United Kingdom
Message 1085686 - Posted: 10 Mar 2011, 14:20:10 UTC - in response to Message 1085492.  

... 160 LP system not being able to scale properly. ...


You could check it out with Linux... That would make for a very interesting comparison! (Linux is designed to scale up well, especially for the scheduler.)


Linux..... barf... :) Sorry we are windows server users only. LOL


Not any more according to the latest news...


Good fun interesting testing there! Didn't know that Itanium systems were still being developed.

Shame there isn't a credit claimed RAC that was calculated on a host so that you could immediately see what rate of processing was being done, rather than waiting a month or so for the boinc server side RAC to glacially drift through its averaging... An instant RAC would make for much easier testing...


Good luck, and:

Happy fast crunchin'!,
Martin


See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)
ID: 1085686 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1085719 - Posted: 10 Mar 2011, 15:42:35 UTC - in response to Message 1085533.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.


"I am starting to understand": Congratulations! You are the first ;)

AFAIK no one have experience running BOINC/SETI on so big number of CPUs,
you are exploring uncharted territories.


P.S.
I see you are using SSE3 app but I think your CPU
Intel(R) Xeon(R) CPU E7- 2870 @ 2.40GHz
supports SSSE3
http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E7-2870%20AT80615007266AA.html

Rerun the Lunatics' Unified Installer and choose SSSE3 for faster/fastest app.


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1085719 · Report as offensive
Bry B

Send message
Joined: 3 Apr 99
Posts: 53
Credit: 832,165
RAC: 0
United States
Message 1085789 - Posted: 10 Mar 2011, 19:14:50 UTC - in response to Message 1085719.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.


"I am starting to understand": Congratulations! You are the first ;)

AFAIK no one have experience running BOINC/SETI on so big number of CPUs,
you are exploring uncharted territories.


P.S.
I see you are using SSE3 app but I think your CPU
Intel(R) Xeon(R) CPU E7- 2870 @ 2.40GHz
supports SSSE3
http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E7-2870%20AT80615007266AA.html

Rerun the Lunatics' Unified Installer and choose SSSE3 for faster/fastest app.




Ahh on the X64 side, good catch. How exactly do I rerun the Lunatics' Unified Installer? Thanks in advance
ID: 1085789 · 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 1085800 - Posted: 10 Mar 2011, 20:02:05 UTC - in response to Message 1085789.  

Superdome 2 is SCALING!! 99%

X64 160 LP box is also at 90%!!!! 5 to 10 percent more would be awesome if anyone has ideas.

this is getting easier now that I am starting to understand what you guys are talking about and understanding the tool.

Thanks everyone!

I think tomorrow I might throw some nasty GPU's in the X64 160 LP box and see what happens. My new memory risers will also me in which will help performance.


"I am starting to understand": Congratulations! You are the first ;)

AFAIK no one have experience running BOINC/SETI on so big number of CPUs,
you are exploring uncharted territories.


P.S.
I see you are using SSE3 app but I think your CPU
Intel(R) Xeon(R) CPU E7- 2870 @ 2.40GHz
supports SSSE3
http://www.cpu-world.com/CPUs/Xeon/Intel-Xeon%20E7-2870%20AT80615007266AA.html

Rerun the Lunatics' Unified Installer and choose SSSE3 for faster/fastest app.




Ahh on the X64 side, good catch. How exactly do I rerun the Lunatics' Unified Installer? Thanks in advance


Double click "Lunatics_Win64v0.37_(SSE3+)_AP505r409_AKv8bx64_Cudax32f.exe" or whichever version you have. :)
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1085800 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 . . . 8 · Next

Message boards : Number crunching : Multi core greater than 80 core


 
©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.