What was once old is new(er? ish?) again.

Message boards : Number crunching : What was once old is new(er? ish?) again.
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · Next

AuthorMessage
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 4869
Credit: 313,966,669
RAC: 677,024
United States
Message 1921460 - Posted: 26 Feb 2018, 23:11:16 UTC - in response to Message 1921456.  
Last modified: 26 Feb 2018, 23:14:58 UTC

After checking your CPU load while just running the GPU you maybe able to run 1 or maybe 2 CPU tasks just for the giggle factor Al. ;-)

Cheers.
Just pulled it up a little bit ago, and it seems to be bouncing between 30-40%, usually 30-35 seems typical, though one core seems to be between 60-90%, so I assume that the ~35% is an average of all cores? I need to put together an app_config file and get it using the whole core instead of the .04 it says it's using, and if it can handle it, I might for fun try running 2 tasks and see if it falls over or can grunt thru the load.


You could use this one with 1080Ti.

<app_config>
  <app_version>
    <app_name>setiathome_v8</app_name>
    <plan_class>opencl_nvidia_SoG</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>1</ngpus>
    <cmdline>-sbs 2048 -period_iterations_num 1 -tt 1500 -high_perf -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 256 -high_prec_timer</cmdline>
  </app_version>
  <app_version>
    <app_name>astropulse_v7</app_name>
    <plan_class>opencl_nvidia_100</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>1</ngpus>
    <cmdline>-unroll 28 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1</cmdline>
  </app_version>
<project_max_concurrent>4</project_max_concurrent>
</app_config>


[Edit] If the cpu_time greatly exceeds the run_time on cpu tasks, you could knock the project_max_concurrent down to 2 or 3. One gpu task with one cpu core supporting it and the another 1 or 2 cores doing cpu tasks.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1921460 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921461 - Posted: 26 Feb 2018, 23:15:42 UTC
Last modified: 26 Feb 2018, 23:19:40 UTC

Ok, going thru creating the app_config file now. Found the one that Marco posted in Marks thread, which has the line:

-sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 512 -oclfft_tune_ls 512 -oclfft_tune_bn 256 -oclfft_tune_cw 256 -pref_wg_num_per_cu 6 -period_iterations_num 1 -hp -high_prec_timer -high_perf -tt 150

but Marks has trimed off the front couple commands, and I left off the -v as he said it wasn't necessary, so this is now his:

-tt 1500 -period_iterations_num 1 -high_perf -high_prec_timer -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 256 -oclfft_tune_cw 256

I presume I just copy Marks over the top of everything starting with the sbs thru to the -high_perf that was in Marcos file?

ID: 1921461 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 4869
Credit: 313,966,669
RAC: 677,024
United States
Message 1921462 - Posted: 26 Feb 2018, 23:17:55 UTC - in response to Message 1921461.  

That'll work. Yes copy the command line tunings into the base app_config you created.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1921462 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921463 - Posted: 26 Feb 2018, 23:27:49 UTC

Ok, do I put Marks line into both the setiathome_v8 and astropulse_v7 sections of the file? I noticed that in Marco's file, those 2 sections were quite different.

Oh, one other thing with this machine, when I file-exit BOINC, it is still running in the background, it isn't down in the task bar, but when I pull up the running processes, it is there along with the task still running as well. I can't imagine that it's good to forcibly ending a task like that, but that is the only way I can get it to close. Is there some setting that I can choose that will make it shut down when I exit it?

ID: 1921463 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 4869
Credit: 313,966,669
RAC: 677,024
United States
Message 1921464 - Posted: 26 Feb 2018, 23:46:20 UTC - in response to Message 1921463.  

No, the tuning lines for MB and AP are completely different for the different apps.

Use the AP section out of my previous post.

In the 7.8.3 Manager under the File menu there is the selection for "Shut down connected client" When you do that the first time it will pop up a dialog asking whether you want to exit running tasks also. Once you acknowledge Yes to that dialog you can then just use the Exit Boinc selection under the File Menu and it will exit both the Manager and Client at the same time and leave no running tasks.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1921464 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921473 - Posted: 27 Feb 2018, 0:15:31 UTC
Last modified: 27 Feb 2018, 0:24:32 UTC

AI & Keith could you check, apparently the app_config is not working, maybe is the old .txt added by the notepath?

If he is running 1 WU at a time the time on the 580 secs range seems high, even for an Atom cpu I expect something in the range of 280-300 secs or less.

If the app_config is working you see something like this in the stderr file

Sleep() & wait for event loops will be used in some places
Target kernel sequence time set to 10ms
Number of period iterations for PulseFind set to:50
Maximum single buffer size set to:1024MB
SpikeFind FFT size threshold override set to:4096
TUNE: kernel 1 now has workgroup size of (64,1,4)
oclFFT global radix override set to:256
oclFFT local radix override set to:16
oclFFT max WG size override set to:256
oclFFT max local FFT size override set to:512
oclFFT number of local memory banks set to:64
oclFFT minimal memory coalesce width set to:64
Priority of worker thread raised successfully
Priority of process adjusted successfully, below normal priority class used
OpenCL platform detected: NVIDIA Corporation
BOINC assigns device 0
Info: BOINC provided OpenCL device ID used


Obviously my settings are different since mine is a 1060 host.
ID: 1921473 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 4869
Credit: 313,966,669
RAC: 677,024
United States
Message 1921476 - Posted: 27 Feb 2018, 0:35:38 UTC

Al, did you reread config files in the Manager after editing the app_config? Or did you exit and restart BOINC. Changes in app_config won't be acted upon unless the config files are reread or BOINC is restarted. Only new tasks started from the beginning will show use of the new parameters.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1921476 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 4869
Credit: 313,966,669
RAC: 677,024
United States
Message 1921477 - Posted: 27 Feb 2018, 0:42:14 UTC - in response to Message 1921473.  

What Juan was referring to is when you use Notepad to create and save an app_config.xml file, you have to make sure to set the Save type as "All Files" and NOT as Text files. If you don't do that Notepad will save the file as app_config.xml.txt and BOINC will not recognize it.

A good sanity check to perform on any edited xml file is to open it with a browser and you should see the standard xml formatting. If there is something wrong with the syntax or structure, the browser usually flags it.
Seti@Home classic workunits:20,676 CPU time:74,226 hours
ID: 1921477 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921487 - Posted: 27 Feb 2018, 1:49:10 UTC
Last modified: 27 Feb 2018, 1:56:59 UTC

AL

Check if in the 20-30 first lines of your Event Log shows something like this when you start the Boinc.

Mon 26 Feb 2018 08:47:44 PM EST | SETI@home | Found app_config.xml


<edit> Did you use app_config on your others hosts? Apparently not. If you tune each host you will reduce the crunching time of the WU sometimes for 20-30% or even more.
ID: 1921487 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921489 - Posted: 27 Feb 2018, 1:59:44 UTC

Sorry guys, hadn't done the app_config yet, I was waiting on Keiths reply about the proper lines in the file, and then the daughter got home, dinner needed to be created (Chili, yum!) and now that homework is done, time to head to the shop for a couple minutes and then home to work on that file and get it read. Yep, I had ran into that issue recently about it saving as a .txt before I had gone into windows explorer and changed the settings on a new install to show extensions, show hidden files/folders, etc. Even thought it looked like it wasn't saving as a .txt, because I couldn't see the extension, it appeared ok. Lesson (re)learned.

As to the 2 diff lines, I will try and find you post about the proper one for the AP section and get that put in there as well, then I will correct the setting to make sure it shuts down each time. I always check that box when it comes up the first time, but maybe since it was an upgrade something slightly borked? Oh well, as long as I can fix it now, all is good. Should be back in about a half hour to 45 mins, and then will dig in and get this old gal to hum!

ID: 1921489 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921491 - Posted: 27 Feb 2018, 2:09:52 UTC - in response to Message 1921489.  
Last modified: 27 Feb 2018, 2:11:14 UTC

Try the one posted by Keith in this msg: http://setiathome.berkeley.edu/forum_thread.php?id=82634&postid=1921460

I just suggest you to start with a more conservative CPU usage instead of 4 use 1 to see the Atom top performance of feed the GPU if Ok then increase to 2, 3 or even 4 but not believe 4 will work fine without major lags.

<project_max_concurrent>1</project_max_concurrent>

Post when you done so we could follow.
ID: 1921491 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921498 - Posted: 27 Feb 2018, 2:44:41 UTC

Ok, here is what I put together for it, going to move it over to the Atom machine now:

<app_config>
  <app_version>
    <app_name>setiathome_v8</app_name>
    <plan_class>opencl_nvidia_SoG</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>1</ngpus>
    <cmdline>-tt 1500 -period_iterations_num 1 -high_perf -high_prec_timer -sbs 2048 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 256 -oclfft_tune_cw 256</cmdline>
  </app_version>
  <app_version>
    <app_name>astropulse_v7</app_name>
    <plan_class>opencl_nvidia_100</plan_class>
    <avg_ncpus>1</avg_ncpus>
    <ngpus>1</ngpus>
    <cmdline>-unroll 28 -oclFFT_plan 256 16 256 -ffa_block 16384 -ffa_block_fetch 8192 -tune 1 64 8 1 -tune 2 64 8 1</cmdline>
  </app_version>
<project_max_concurrent>2</project_max_concurrent>
</app_config>


I'll let you know how it goes, think I will try running 2 tasks at a time, living dangerously here!

ID: 1921498 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921499 - Posted: 27 Feb 2018, 2:51:33 UTC - in response to Message 1921498.  

Of course is your choice, but my point is to first know the limit of the capacity of the Atom to feed the GPU, when you find it then you could try to start some CPU tasks.

A Atom core crunch a single WU in hrs, and the 1080Ti in minutes so any gain in the GPU performance counts a lot more than the CPU.

Maybe the Atom could even feed 2 GPU tasks at the time. But that is an unknown territory until someone try.
ID: 1921499 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921500 - Posted: 27 Feb 2018, 2:54:11 UTC

Well, it started up without much incident, was a little laggy when it was connecting to the client, but then it settled in fine. First thing I noticed was it was still only running one GPU task, though I thought I had set it to 2. Here is the beginning of the event log, everything appears to be in order, but maybe I missed something?

2/26/2018 8:48:35 PM | | Starting BOINC client version 7.8.3 for windows_x86_64
2/26/2018 8:48:35 PM | | log flags: file_xfer, sched_ops, task
2/26/2018 8:48:35 PM | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
2/26/2018 8:48:35 PM | | Data directory: C:\ProgramData\BOINC
2/26/2018 8:48:35 PM | | Running under account ATOM ANT
2/26/2018 8:49:04 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 1080 Ti (driver version 390.77, CUDA version 9.1, compute capability 6.1, 4096MB, 3034MB available, 12064 GFLOPS peak)
2/26/2018 8:49:04 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 1080 Ti (driver version 390.77, device version OpenCL 1.2 CUDA, 11264MB, 3034MB available, 12064 GFLOPS peak)
2/26/2018 8:49:04 PM | SETI@home | Found app_info.xml; using anonymous platform
2/26/2018 8:49:04 PM | | Host name: ATOM_ANT
2/26/2018 8:49:04 PM | | Processor: 4 GenuineIntel Intel(R) Atom(TM) CPU 330 @ 1.60GHz [Family 6 Model 28 Stepping 2]
2/26/2018 8:49:04 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 movebe syscall lm tm2 pbe
2/26/2018 8:49:04 PM | | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
2/26/2018 8:49:04 PM | | Memory: 3.12 GB physical, 6.25 GB virtual
2/26/2018 8:49:04 PM | | Disk: 74.43 GB total, 56.86 GB free
2/26/2018 8:49:04 PM | | Local time is UTC -6 hours
2/26/2018 8:49:04 PM | SETI@home | Found app_config.xml
2/26/2018 8:49:04 PM | | Config: use all coprocessors
2/26/2018 8:49:04 PM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 6689894; resource share 100
2/26/2018 8:49:04 PM | SETI@home | General prefs: from SETI@home (last modified 03-Apr-2013 23:59:56)
2/26/2018 8:49:04 PM | SETI@home | Computer location: school
2/26/2018 8:49:04 PM | | General prefs: using separate prefs for school
2/26/2018 8:49:04 PM | | Reading preferences override file
2/26/2018 8:49:04 PM | | Preferences:
2/26/2018 8:49:04 PM | | max memory usage when active: 1599.65 MB
2/26/2018 8:49:04 PM | | max memory usage when idle: 3039.34 MB
2/26/2018 8:49:04 PM | | max disk usage: 37.21 GB
2/26/2018 8:49:04 PM | | (to change preferences, visit a project web site or select Preferences in the Manager)
2/26/2018 8:49:04 PM | SETI@home | Sending scheduler request: To fetch work.
2/26/2018 8:49:04 PM | SETI@home | Requesting new tasks for NVIDIA GPU
2/26/2018 8:49:06 PM | SETI@home | Scheduler request completed: got 0 new tasks
2/26/2018 8:49:06 PM | SETI@home | No tasks sent
2/26/2018 8:49:06 PM | SETI@home | No tasks are available for AstroPulse v7
2/26/2018 8:49:06 PM | SETI@home | No tasks are available for SETI@home v8
2/26/2018 8:49:06 PM | SETI@home | Tasks for CPU are available, but your preferences are set to not accept them
2/26/2018 8:49:06 PM | SETI@home | This computer has reached a limit on tasks in progress

ID: 1921500 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921502 - Posted: 27 Feb 2018, 2:58:56 UTC - in response to Message 1921499.  

Of course is your choice, but my point is to first know the limit of the capacity of the Atom to feed the GPU, when you find it then you could try to start some CPU tasks.

A Atom core crunch a single WU in hrs, and the 1080Ti in minutes so any gain in the GPU performance counts a lot more than the CPU.

Maybe the Atom could even feed 2 GPU tasks at the time. But that is an unknown territory until someone try.
Lol especially in this instance. This is one pokey CPU, I think it's going to do all it can to just keep up feeding the GPU, especially if I can get it configured to run 2 tasks at a time. I guess I'm going to be the trailblazer here, huh? Fun stuff!

ID: 1921502 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921503 - Posted: 27 Feb 2018, 3:00:09 UTC - in response to Message 1921500.  
Last modified: 27 Feb 2018, 3:05:15 UTC

No it`s working as expected.

With this setting your host will start 1 GPU + 1 CPU tasks.

But your log file said:

2/26/2018 8:49:06 PM | SETI@home | Tasks for CPU are available, but your preferences are set to not accept them

so go to your preferences and enable the cpu crunch if you wish to do CPU work. Not a good ideia in you case i presume.

For now lets focus on the GPU performance can you check the GPU usage and Temps?
ID: 1921503 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921504 - Posted: 27 Feb 2018, 3:07:16 UTC - in response to Message 1921503.  

Oh, no, I had set up a location called school, which I had done when I had 2-3 of these types of boards running back in the day, and now I remember, initially I did try running CPU tasks as well, and it was fairly Fugly. Passive cooling wasn't up to it, and even when I put a fan on them, the output was miserly compared to the GPU, even back then when I was running a GTX290 (one Hot card) and then up to the 4-500 series. So I created the location School, it made it easy to just turn off all CPU processing, which I found out was a good thing with these boards.

So, as to getting it to run 2 tasks on the GPU, and to nuke the references to running on the CPU, where would one go to accomplish this? I do remember when I ran the Lunatics installer, I specifically remembered to uncheck the CPU portion of the install, but that didn't take care of it?

ID: 1921504 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921505 - Posted: 27 Feb 2018, 3:14:39 UTC

To run 2 GPU WU at the time just change this line:

<ngpus>1</ngpus>

for

<ngpus>0.5</ngpus>

and restart the Boinc or read the config files.

But first runs few Wu with 1 to get some time sample to see what is better.

Did you look the GPU usage and Temps?
ID: 1921505 · Report as offensive
juan BFP Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 6938
Credit: 394,511,430
RAC: 141,443
Panama
Message 1921508 - Posted: 27 Feb 2018, 3:37:21 UTC
Last modified: 27 Feb 2018, 3:42:02 UTC

With project_max_concurrent you could easily control if you do or not any CPU work.

Boinc starts first the GPU WU. If you choose 2 So Boinc will try to start in this order (if you have available WU to crunch in your cache):
2 GPU
1 GPU + 1 CPU
2 CPU

The number of GPU WU is defined by <ngpus> Obviously it will start CPU WU only if you allow the CPU WU to be crunched in your preferences.

Your times look better but still slow for a 1080Ti, try 2 at the time and lets see what you get.
With one your crunching time is in the range of 510 secs.

You need to check the GPU usage. You could use the GPU-Z program to look that easily, or any other equivalent.
ID: 1921508 · Report as offensive
Al Special Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1635
Credit: 360,232,618
RAC: 278,575
United States
Message 1921509 - Posted: 27 Feb 2018, 3:41:50 UTC

I'll hold off changing to 2 tasks till probably morning, to see how it's running with one right now. Here are the settings in EVGA Precision for my overclocking:

GPU Clock 2062
Memory Clock 5022
GPU Temp 30*
Power Temp 35*
Memory Temp 34*
Power Target 127%
Temp Target 65* (lowest, unlinked them)
GPU Clock Offset +72 MHz
Memory Clock Offset +16 MHz (both pretty conservative)
Voltage 1062 (conservative too)

Been running stable at these settings since I installed the card and software. After putting it under some strain by running 2 tasks at once, I am going to start lowering the fans and pump settings, they are currently at 100%, probably can be lowered a bit looking at my current temps, but that can wait.

ID: 1921509 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · Next

Message boards : Number crunching : What was once old is new(er? ish?) again.


 
©2018 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.