Multithreading?

Message boards : Number crunching : Multithreading?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile KA1J

Send message
Joined: 11 Oct 00
Posts: 20
Credit: 432,783
RAC: 2,486
United States
Message 1913116 - Posted: 15 Jan 2018, 1:58:05 UTC

Hi there,

I've just now put together a new computer with lots of very nice options, It's based on the Asus Prime Z370-A motherboard and the processor is an i7-8700K with 6 cores and 12 threads running at 4.7GHz. With all the abilities of this beast I'm wondering if SETI@home's/BOINC software is set up to take advantage of all these cores and the hyperthreading?

I've tried to find an answer but moderate searching has come up blank.

Anyone have the answer?

Thanks!
ID: 1913116 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4230
Credit: 242,054,475
RAC: 363,160
United States
Message 1913118 - Posted: 15 Jan 2018, 2:03:32 UTC - in response to Message 1913116.  

You can run seti on that with no problem but I would not recommend crunching any work on the built in GPU. If you have an open slot, crunching on a dedicated GPU is much, much faster than on the CPU. The server will try to run work on all your cores if you let it, best to restrict it to maybe 75% unless you aren't using the computer for anything. Depending on the GPU you get, you could run multiple work units on it instead and free up some CPU cores.


Zalster
ID: 1913118 · Report as offensive     Reply Quote
Profile KA1J

Send message
Joined: 11 Oct 00
Posts: 20
Credit: 432,783
RAC: 2,486
United States
Message 1913126 - Posted: 15 Jan 2018, 3:58:43 UTC - in response to Message 1913118.  

Thanks for the reply!

The GPU I've got for it is a mid-range one, a Geforce GTX 1050. I chose that one because it has a 2.0 HDMI output & will be fine for 4K output. It has 640 CUDA cores.
ID: 1913126 · Report as offensive     Reply Quote
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 4230
Credit: 242,054,475
RAC: 363,160
United States
Message 1913131 - Posted: 15 Jan 2018, 4:31:16 UTC - in response to Message 1913126.  

Next thing would be to install lunatics. You have to read each page as you go thru it. You want to install the CPU applications for both multibeam and Astropulses. You will want to skip the ATI section but when you get to Nvidia, the default is for cuda 3.0. You must deselect that and select SoG. Once the installation is complete, any new work will process under the installed applications. Any work downloaded before the installation of the Lunatics will still say whatever application it was going to run under but it will in fact change to SoG, so don't panic when you see that. I would recommend you use the advance view of BOINC and under computing preference section the section of "Use at most" CPU and put the value of 75 in. That way it doesn't use all the CPU cores, otherwise responsiveness of the computer will suffer.

http://mikesworld.eu/download.html

You will want installer

Lunatics Win64 0.45 Beta 6
ID: 1913131 · Report as offensive     Reply Quote
Profile Jimbocous Project Donor
Volunteer tester
Avatar

Send message
Joined: 1 Apr 13
Posts: 1062
Credit: 98,358,516
RAC: 87,542
United States
Message 1913156 - Posted: 15 Jan 2018, 10:07:08 UTC
Last modified: 15 Jan 2018, 10:12:30 UTC

What Zalster said :)
I still haven't been able to find a good answer as to whether enabling hyper-threading helps or hurts. I suspect it's close to a wash, based on what I've seen.
My crunchers that support HT (X5675) are dedicated crunchers, so I have it enabled, and set the number of CPUs allowed to do SETI to the number of available threads minus the count of GPUs installed. This is because with the SoG application it will want a free CPU thread per GPU to get decent performance. Running more than 1 WU per GPU is specifically not recommended with the SoG application.
What I have done is use HWInfo64 on each machine to watch the CPU loading (and temps), and increase it until I'm as close as possible to 100%. Works out about like this, for each machine type I'm running:
CPU          #Cores          #Threads         #GPUs     #CPUs in CC_Config (how many CPU jobs running)
X5675          6                12              2           9
Q9650          4                 4              2           2
E5450          4                 4              3           1

In order to ensure I wasn't slowing GPU processing speed by choking GPUs in favor of CPUs, I have tested reducing CPU tasks to actual number of cores rather than threads. The result was no change in average WU time on GPU, but an immediate reduction in total production due to all the CPU idle time that resulted.
FWIW, the E5450 is a dedicated cruncher, so I don't sweat that the CPU stays pegged at 100%. The Q9650 is my daily driver, but I put up with a bit of intermittent lag to maximize production, and have entered time-critical programs into the exclusion list so that BOINC suspends when I run those programs. Seems like a fair compromise. All CPUs are ~3ghz.
HWInfo also lets me see that the GPUs are running close to 100% also, which again is where I want them.
YMMV, hope this is useful. Feel free to reply back if I can help.
Jim ...
ID: 1913156 · Report as offensive     Reply Quote

Message boards : Number crunching : Multithreading?


 
©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.