Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
![]() |
World Community Grid Forums
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 53
|
![]() |
Author |
|
hnapel
Advanced Cruncher Netherlands Joined: Nov 17, 2004 Post Count: 82 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
A lot has been said about hyperthreading and I know the usual arguments pro and con, I can only say what I observe here: without actually disabling hyperthreading if I reduce the the number of WCG processes to the actual number of cores (50%) the execution time of the WU's gets better, of course putting more logical cores to work will result in overall more work being done, but if you aim at returning results faster rather than quantity you could consider this. B.t.w. the reason to reduce workload a bit was to reduce heat dissipation and noise, I have an Alienware R8 with an Intel core i9 9900K, this has 8 real cores and x2 hyperthreading.
|
||
|
Sgt.Joe
Ace Cruncher USA Joined: Jul 4, 2006 Post Count: 7745 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
You make an interesting point about heat and fan noise. In my opinion, the problem of heat and heat dissipation is due to inadequate design capacity. It is due to the manufacturers cheaping out on high quality components in order to put out a high performing system which they erroneously presume will not be used for any length of time at its potential capacity. This is especially true for laptops with high end CPU's. If your laptop gets too hot to sit on your lap, why call it a laptop. I don't know the answers necessarily, but do think better heatsinks, higher quality fans, and adequate airflow are some of the keys to addressing this issue. Which brings us back to hyperthreading. If the above issues can be addressed adequately, you will get more throughput overall by about 15 to 20% while using some additional electricity. If I used laptops I would probably not hyperthread, but I only use desktops and servers, so I use hyperthreading.
----------------------------------------Cheers
Sgt. Joe
*Minnesota Crunchers* |
||
|
Macromancer
Veteran Cruncher United States Joined: Sep 6, 2016 Post Count: 994 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I do the same with my dual socket Xeon (X5660) workstation. 12 cores / 24 threads, but only 12 cores running WCG 24/7. If I try to run more than 12 WCG jobs simultaneously i generate a ton of heat, use a lot more electricity and get only about 20% more throughput. It seems the Spectre and Meltdown patches really slow hyper-threading to a crawl on these older generation Xeon CPUs.
On my AMD Ryzen workstations, I run multi-threading all of the time, i.e. at least 75% of available threads running WCG 24/7 with significant throughput gains with very little excess heat and additional electricity usage. Macromancer |
||
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1680 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I always operate my machines at 100% with hyperthreading enabled.
----------------------------------------I usually never use the standard CPU cooler but change it for better cooler, e.g. Noctua. The machines perform well and are quiet. I do never overclock the machines. Cheers, Yves |
||
|
supdood
Senior Cruncher USA Joined: Aug 6, 2015 Post Count: 333 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I'm curious to know how using hyperthreading but at reduced CPU frequencies would compare on throughput, heat, and electricity use to running without hyperthreading at full frequencies. Has anyone explored this?
---------------------------------------- |
||
|
Tamagoch
Cruncher Ukraine Joined: Jul 21, 2007 Post Count: 33 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I always turn HT off lo load physical cores.... if I choose to run 50% with HT enabled I can see how wrongly applied affinity can load one real core with two tasks - and then performance drops (lately tested with distributed.net client with live core throughput display).... maybe it was fixed in the new CPUs but I'm pretty oldschooled on that - only real cores, only hardcore! )))
|
||
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
There's progs like Process Lasso that can make sure 1,3,5 etc are not used by the science apps.
|
||
|
fuzzydice555
Advanced Cruncher Joined: Mar 25, 2015 Post Count: 89 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
The last time I turned off hyperthreading was with an X5650. On that old-old CPU hyperthreading was +20% points for +20% power consumption.
----------------------------------------I can only hope HT has gotten better in the last 10 years, so if you have a modern CPU and you have OK thermals there is not much point in turning it off. The 9900K is already an extremely fast CPU, with or without HT you're returning tasks faster than most of us. The thermal issue are basically caused by this, 9900K is factory overclocked to an insane clock speed. I had the same issue with the 4790K, during crunching it immediately topped out at 99C with a Hyper 212 tower cooler. My solution was to limit clock speeds to 3 GHz. Power consumption went from 126W to 58W while I lost 30% performance... but I gained a silent PC with thermals around 55C :) CPUs get extremely power hungry around the 5GHz mark where the 9900K operates, so you might get very good results with TDP limiting or underclocking. ![]() [Edit 1 times, last edit by fuzzydice555 at May 21, 2020 10:37:29 PM] |
||
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1680 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I use to keep my CPU not higher than 65°C.
----------------------------------------The both i7 - 4770K (Win7) as well as 6700K (Linux Mint) - succeed to keep 65°C as limit without dropping the performance, even in Summer. The Phenom II x6 do not have multithreading capability and stay below 60°C, mostly at 55°C. The Ryzen 7 2700X succeed as well to keep 65°C as limit without dropping the performance, even in Summer. I use good cooler (Noctua) and I do not overclock, but the machine crunch at 100% with hyper/multithreading. I deliberately prefer not to overclock, at least to avoid speeding up CPU and memory ageing, additionally because of energy consumption (for me, overclocking is clause to pure energy wasting). Cheers, Yves |
||
|
CurtisNewton
Cruncher Joined: Feb 24, 2008 Post Count: 25 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.
----------------------------------------This can speed up processing if the instruction queue of one thread has gaps related to that resource or if the queue needs to wait for something (e.g. pagefault and waiting for data getting available). Typically application thats do some floating point stuff and then something else will benefit from ht, since when one hw-thread does not use the fpu, the other hw-thread can. Programs that do only computations and are using fpu completly can even be slower with ht enabled since two threads have to compete for the same resources. Basically, if you have 6 cores with 6 fpus and turn ht on, you will have 12 logical cpus, but still only 6 physical fpus. Btw, CPU vendors never claimed that HT will double speed, typical benefit for average "multi purpose" programs is about 10%-30%. Rule of thumbs: If you run highly optimized code that does computation only, you most likely will not benefit from HT, at least not noticeable. I did some some experiments with fft stuff on 100gb data sets, using all that optimized math libs, and I never benefited from ht. [Edit 2 times, last edit by CurtisNewton at May 23, 2020 7:51:02 PM] |
||
|
|
![]() |