Do Helper Threads actually improve performance?

[kgor93]

I have been playing around with Helper Threads on both my laptop and my desktop. My main concern is the speed of auto-simming since that is the most cpu-intensive process.

Haven't noticed any difference with/without helper threads.

My conclusion: if they do make things faster, it's only marginally. So IMO, disable threading.

[Matt Arnold]

Every machine is different. It certainly will make a different if you have a computer with many cores, but on many machines there may be literally no difference.

It will also vary a lot depending on how your league setup is. For example, if you are generating a lot of box scores and game logs, they are quite slow and not as easily threaded as other parts of the game, so you may not notice a difference there. However if you are simming a league without storing any of those, then you may notice a larger improvement due to threading.

But in general, my take on threading is to leave it on whatever the recommended setting is, unless if you understand enough about your machine as to why you should try it on another setting.

But also, as I sort of alluded to above, if you're most concerned about speed of simming, then I would definitely recommend checking through the various league options, since those can make a larger difference than enabling threading. Enabling all the options (game logs, recaps, box scores, etc...) is probably a 10X difference from doing everything to doing nothing, whereas threading is maybe a 2X factor in how much it can realistically speed things up.

[NoOne]

I just want to point out one drawback to "leave it on recommended" for # of threads.

i have a 4core cput wiht intel's multi-(hyper-?)threading -- whichiver the tm is. 4 physical cores and a little gimmick that uses the 'math' portion of the chip as it's own 'core', in layman's as best i know.

anyway... in this situation, recommended uses "8" which slows down everything significantly even compard to just "1" thread, at least with my pc.

if i use "4", everything is much faster than "1" thread.

so, it's better to say set it to the # of physical cores. unless this turns out to be a localized issue.

This type of cpu feature is generally defaulted "on" in the bios/uefi. so, 'recommended' will use 2x physical cores for anyone with a similar architecture (gimmick that makes a chip "double" physical cores - AMD has this too) unless they changed this bios controlled feature, which isn't likely.

[Matt Arnold]

Originally Posted by NoOne

I just want to point out one drawback to "leave it on recommended" for # of threads.

i have a 4core cput wiht intel's multi-(hyper-?)threading -- whichiver the tm is. 4 physical cores and a little gimmick that uses the 'math' portion of the chip as it's own 'core', in layman's as best i know.

anyway... in this situation, recommended uses "8" which slows down everything significantly even compard to just "1" thread, at least with my pc.

if i use "4", everything is much faster than "1" thread.

so, it's better to say set it to the # of physical cores. unless this turns out to be a localized issue.

This type of cpu feature is generally defaulted "on" in the bios/uefi. so, 'recommended' will use 2x physical cores for anyone with a similar architecture (gimmick that makes a chip "double" physical cores - AMD has this too) unless they changed this bios controlled feature, which isn't likely.

Yes, this could count as "understand enough about your machine"

Although in general, it also depends on how much else you run on your machine at the same time. For example, if you're not running anything else, then I would expect that a 4 core/hyperthreaded box should still run ever so slightly faster with 8 threads than with 4. However, if you're running a bunch of other stuff too, then the 8 threads can slow it down enough to cause issues.

So yes, if you don't feel that threading helps much, then try half the recommended value.

[kgor93]

So I’ve actually been testing this extensively recently.

My systems:
i7-4770HQ Mac
4 cores 8 threads
16GB RAM
SSD
~12000 Geekbench 4

i7-4790k Windows
4 cores 8 threads
24GB RAM
SSD
~15000 Geekbench 4

Temporary rented system:
C4.8xlarge Amazon EC2
19? (what it said under computer) cores 38 threads
60GB RAM
SSD
~38000 Geekbench 4

What I did for the tests: Nothing too scientific, I just started auto-simming my franchise, which has basic MLB feeder leagues and the independent/European leagues combined into one secondary major league.

The results:

On my desktop and laptop there was a noticeable difference between no threads and threading. But it wasn’t too significant. It was a slight boost. Nothing spectacular.

Though I should note, when switching the threading mode, I should have restarted the game. I did not do this and the number of threads decreased but was still about a dozen.

On the EC2 instance with 38 (yes you read that right) helper threads enabled, it was SLOOOOOWWWW. CPU usage never climbed above 50% despite being fully allocated. In addition, it was taking multiple minutes to sim single days. Decreased to 16 and noticed a slight performance increase. CPU usage stayed around 17-25%, but still quite slow.

I can go on further but I’m short on time so I’m going to wrap this up.



My conclusion: Helper threads do improve performance, but not drastically. I would not enable more than 8, and ideally 4. Any more seems like a waste.

One last thing: I would like to revisit the EC2 instance. I’m thinking maybe because of the massive number of threads it needs some time to warm up. Like it’ll start slow and pickup speed.

[NoOne]

Originally Posted by Matt Arnold

Yes, this could count as "understand enough about your machine"

Although in general, it also depends on how much else you run on your machine at the same time. For example, if you're not running anything else, then I would expect that a 4 core/hyperthreaded box should still run ever so slightly faster with 8 threads than with 4. However, if you're running a bunch of other stuff too, then the 8 threads can slow it down enough to cause issues.

So yes, if you don't feel that threading helps much, then try half the recommended value.

it slows it down enough that it's obvious to the eye, even slower than 1 thread as a comparison above. sim a year and it's painfully obvious. (8threads, recommended, on a 4core chip with multithreading(tm) cpu enabled)

i wouldn't say anythign, otherwise. it's like the trade deadline or end of MiL seasons everyday or worse. (ie whatever most intesnive 'days' are) i knew to avoid it... it took a week or so to realize my cpu's multithreading was erroneously and magically re-enbled. thought my league was just getting 'too big' before i noticed.

when i run 8 threads on 4-cores (with mutlithreading turned off in bios) -- it does work fine (i keep it at 4, fwiw).. no faster or slower that i can tell without a stopwatch. this cpu feature is defaulted on and most will not turn it off. just not sure if all pc's with this cpu-feature experience the problem. few things benefit, numerous things are negatively impacted.

(cpu feature, not ootp threading -- ootp threading is all a good thing)

there were quite a few people intially complaining about the slow-down. if like me, it was this cpu feature... this crappy, crappy, useless feature, lol.... (it does actually help with video encoding, but runs hot as he!! even with a water-cooling.)

[kgor93]

Originally Posted by NoOne

it slows it down enough that it's obvious to the eye, even slower than 1 thread as a comparison above. sim a year and it's painfully obvious. (8threads, recommended, on a 4core chip with multithreading(tm) cpu enabled)

i wouldn't say anythign, otherwise. it's like the trade deadline or end of MiL seasons everyday or worse. (ie whatever most intesnive 'days' are) i knew to avoid it... it took a week or so to realize my cpu's multithreading was erroneously and magically re-enbled. thought my league was just getting 'too big' before i noticed.

when i run 8 threads on 4-cores (with mutlithreading turned off in bios) -- it does work fine (i keep it at 4, fwiw).. no faster or slower that i can tell without a stopwatch. this cpu feature is defaulted on and most will not turn it off. just not sure if all pc's with this cpu-feature experience the problem. few things benefit, numerous things are negatively impacted.

(cpu feature, not ootp threading -- ootp threading is all a good thing)

there were quite a few people intially complaining about the slow-down. if like me, it was this cpu feature... this crappy, crappy, useless feature, lol.... (it does actually help with video encoding, but runs hot as he!! even with a water-cooling.)

I have 4 cores as well. An i7-4790k. I find that if I use 4 helper threads it’s faster than a single-threaded instance but with 8 threads it gets ungodly slow. I also experimented with putting my entire ootp on a ramdisk. It makes initial loads and saving slightly faster but for the actual simulating there isn’t a change.