Hyper-Threading on CPU

Message boards : Number crunching : Hyper-Threading on CPU
Message board moderation

To post messages, you must log in.

AuthorMessage
Prescott

Send message
Joined: 15 May 17
Posts: 14
Credit: 3,462,605
RAC: 0
Message 3494 - Posted: 11 Jun 2023, 14:11:59 UTC

Hi!

I was not able to find an answer so I am posting here to ask!

I know some projects discourage Hyper-threading on CPU's for certain tasks because it makes them slower/produce bad results.

Excluding the issues from heat generation from HT, is there a problem with using HT for this project?

Best,

Prescott
ID: 3494 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1330
Credit: 435,571,206
RAC: 485,858
Message 3495 - Posted: 11 Jun 2023, 17:17:56 UTC - in response to Message 3494.  

Hi!

I was not able to find an answer so I am posting here to ask!

I know some projects discourage Hyper-threading on CPU's for certain tasks because it makes them slower/produce bad results.

Excluding the issues from heat generation from HT, is there a problem with using HT for this project?

Best,

Prescott


There is not a problem with HT, but tasks will run approximately 2 times longer. I did some stats on this several years ago and found that the average run times were about 1.9x (as best as I can remember). So there is a small improvement, but I personally didn't think it was worth the extra load on the CPU. Other things to keep in mind is the code has changed since my original tests and HT on newer CPUs may be better now.
ID: 3495 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Prescott

Send message
Joined: 15 May 17
Posts: 14
Credit: 3,462,605
RAC: 0
Message 3496 - Posted: 11 Jun 2023, 20:51:35 UTC - in response to Message 3495.  

Thanks for the quick reply Eric!

I will give it a try and let you know!
ID: 3496 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1330
Credit: 435,571,206
RAC: 485,858
Message 3497 - Posted: 11 Jun 2023, 21:41:13 UTC - in response to Message 3496.  

If you wanted to do a quick timing test yourself, you can use the "Remaining (estimated)" column in the boinc manager. This value for an unstarted task is a good estimate of the average run time. Just make sure you have run a sufficient number of WUs so that the average is accurate (several hundred WUs should be good). Do it once for single threaded tasks, and then again for hyper-threaded tasks, and then compare the numbers.
ID: 3497 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Prescott

Send message
Joined: 15 May 17
Posts: 14
Credit: 3,462,605
RAC: 0
Message 3500 - Posted: 13 Jun 2023, 3:33:04 UTC - in response to Message 3497.  

If you wanted to do a quick timing test yourself, you can use the "Remaining (estimated)" column in the boinc manager. This value for an unstarted task is a good estimate of the average run time. Just make sure you have run a sufficient number of WUs so that the average is accurate (several hundred WUs should be good). Do it once for single threaded tasks, and then again for hyper-threaded tasks, and then compare the numbers.


Preliminary data suggests a 58% increase in estimated time to completion based on BOINC estimated time to complete after several hundred WUs.

The WU set they have been crunching is 4 Cobblestones less than the previous set and is only searching to 16 Cvec.

Overall, it appears that it is completing more, but without additional data I cannot say for sure.
ID: 3500 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Eric Driver
Project administrator
Project developer
Project tester
Project scientist

Send message
Joined: 8 Jul 11
Posts: 1330
Credit: 435,571,206
RAC: 485,858
Message 3501 - Posted: 13 Jun 2023, 5:19:38 UTC - in response to Message 3500.  

If you wanted to do a quick timing test yourself, you can use the "Remaining (estimated)" column in the boinc manager. This value for an unstarted task is a good estimate of the average run time. Just make sure you have run a sufficient number of WUs so that the average is accurate (several hundred WUs should be good). Do it once for single threaded tasks, and then again for hyper-threaded tasks, and then compare the numbers.


Preliminary data suggests a 58% increase in estimated time to completion based on BOINC estimated time to complete after several hundred WUs.

The WU set they have been crunching is 4 Cobblestones less than the previous set and is only searching to 16 Cvec.

Overall, it appears that it is completing more, but without additional data I cannot say for sure.


That sounds like a nice improvement, definitely better than what I saw years ago.
ID: 3501 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Hyper-Threading on CPU


Main page · Your account · Message boards


Copyright © 2024 Arizona State University