| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Active Research
Forum: OpenPandemics - COVID-19 Project
Thread: Hyperthreading.... |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 53
|
|
| Author |
|
|
RTorpey
Advanced Cruncher Joined: Aug 24, 2005 Post Count: 67 Status: Offline Project Badges: 
|
A 15-20% gain with HT is possible with newer CPU's - the older CPU's actually performed worse with HT enabled (at least in a graphics renderfarm environment).
|
||
|
|
hnapel
Advanced Cruncher Netherlands Joined: Nov 17, 2004 Post Count: 82 Status: Offline Project Badges: 
|
I always run 75% core use, leaving whatever open for the GPU and F@H currently. I run the i9-9900K at 63% [(10/16)x100=62.5%] So it uses 10 logical cores, I also run F@H on the GPU that needs a CPU process and so it leaves some overhead for the OS, my general idea is not to run it maxed out all the time. |
||
|
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: 
|
WCG gives badges based on time not points or the amount of work processed. The more cores, logical or virtual means more time towards badges.
----------------------------------------
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
----------------------------------------  [Edit 1 times, last edit by nanoprobe at May 27, 2020 2:54:28 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I always run 75% core use, leaving whatever open for the GPU and F@H currently. I run the i9-9900K at 63% [(10/16)x100=62.5%] So it uses 10 logical cores, I also run F@H on the GPU that needs a CPU process and so it leaves some overhead for the OS, my general idea is not to run it maxed out all the time. Whatever the percent is you enter, the resultant number of threads is rounded down, so 64.777% will also become 10 logical cores on a 16 core machine. Entering 99.99% will in effect always utilize all threads minus 1 for all machines linked to the particular profile i.e. 1 on a duo, 3 on a quad, 7 on a octo etc. The past hard entering of a cores number became no good when quad and higher got into the mix. The function on the WCG device profiles became ignored for clients after 6.2 but it's left there because there's still client with 6 or less out there hard as that may be to believe. Think you can see that at BOINCStats. |
||
|
|
Speedy51
Veteran Cruncher New Zealand Joined: Nov 4, 2005 Post Count: 1326 Status: Offline Project Badges: 
|
Using a Ryzen 9 3900X with HT off (12 cores) task completes in roughly 1 to 1.03 hours, running for 12 to 14 hours a day I can complete 98 tasks with HT on (24 cores) some tasks can take up to nearly 2 hours, completing 113 to 154 tasks a day (12 to 14 hours) . On Windows 10 and one court is reserved for the GPU so I am either running 11 or 23 tasks. To me it is more beneficial to run with HT on in terms of number of tasks and also one time returned
---------------------------------------- [Edit 1 times, last edit by Speedy51 at May 28, 2020 10:52:31 PM] |
||
|
|
Bobzilla
Cruncher Joined: May 31, 2020 Post Count: 31 Status: Offline Project Badges:
|
I also am running a Ryzen 9 3900x with 64GB ram and run times are 1 hr 2 min to 1 hr 52 min on 24 cores. It said was returning 280 to 380 tasks per day. I will also be building a 3950x later this year. I have granted access to 3TB storage if they need it. 20TB storage will be avail when 3950x is done.
|
||
|
|
Aurum
Master Cruncher The Great Basin Joined: Dec 24, 2017 Post Count: 2391 Status: Offline Project Badges: 
|
So, I did a test on my 3900X (12/24 cores/threads, RAM at 3600 MHz, no GPU work) for OpenPandemics WUs. Vezixig, nice test. How long did you run it? I've noticed some OPN WUs are much longer than others although many are similar.SMT on (24 WUs parallel): 106 minutes per WU (324 WUs per day) at 81°C CPU temp consuming 142W/h (10,45W per WU) SMT off (12 WUs parallel): 61 minutes per WU (283 WUs per day) at 74°C CPU temp consuming 122W/h (10,00W per WU) I just got my first i9-9960X (2x8 GB 2666 GHz RAM, Noctua NH-U12S, 3 TB spinner) yesterday and started to test it. Today I disabled HT in the BIOS and started a fresh tranche of OPN WUs with an ARP thrown in for good measure. It's early but eyeball results indicate WUs run in about 1.5 hours versus 2.5 hours. My first impression is disappointment that an i9 isn't much better than an i7 or an E5. I set Prefs to 100% CPU, stopped GPU work, Suspended Project WCG and Ran CPU Benchmarks in BOINC Manager: HT enabled: 5153, 121251 HT disabled: 5836, 171446 Specs say 3.1 GHz with Turbo 4.5 GHz. While running it settled on running at 4.0 GHz. watch -n 5 sensors shows all 16 CPU temps stay within 69-72C range. First no-HT WU finished at 1:13, 2nd at 1:17 and the 14th at 1:20 hours. The lone ARP started with HT but I can see it sped up with no HT.   ...KRI please cancel all shadow-banning[Edit 1 times, last edit by Aurum420 at Jun 7, 2020 4:32:20 PM] |
||
|
|
Aurum
Master Cruncher The Great Basin Joined: Dec 24, 2017 Post Count: 2391 Status: Offline Project Badges:
|
The idea behind hyperhtreading is that two (hardware) threads share the same computational resource such as a single floating point unit. So what resources are shared with various WCG projects??? Is there an a priori prediction whether HT helps or not??? Maybe uplinger could shed some light on this question....KRI please cancel all shadow-banning |
||
|
|
Aurum
Master Cruncher The Great Basin Joined: Dec 24, 2017 Post Count: 2391 Status: Offline Project Badges:
|
BoincTasks has a Tasks/day column where it "Shows tasks that have arrived in the last 24 hours." I'm thinking a fair HT test would be if I took pairs of computers with the same CPU and in the BIOS I enabled HT on one and disabled it on the other and let them run OPN WUs for 1.5 days. Pairs I could run: i7-6950X, E5-2673 v4, E5-2686 v4, E5-2678 v3, i7-5960X, E5-2676 v3, and E5-2683 v3 in order of descending sSpec.
----------------------------------------Edit: It just dawned on me what a useless number that is, who cares how many tasks arrived as it only matters how many tasks completed. I don't see where WCG provides enough data to do this. Any suggestions? ...KRI please cancel all shadow-banning[Edit 2 times, last edit by Aurum420 at Jun 7, 2020 5:26:57 PM] |
||
|
|
|