Does keeping Hyper Threading enabled make any sense for home systems? Especially on AL/RL CPUs...

[XoR_]

Hyper Threading splits core in to two threads and each thread has the same performance as without HT if second thread remains unused. When such second thread being used then each thread on given core can get around 60-70% performance of single thread. This is what of course leads to overall performance improvements HT brings.

The possible issue is that when having HT enabled is that running background application which uses lots of CPU time will get scheduled on the same cores as game is using through additional hardware threads Hyper Threading creates.
It does indeed lead to decreased performance in games when running heavy background tasks and can be easily confirmed running heavy background application and checking FPS in game.

Disabling HT solves this issue because game has higher priority and it will always get as much CPU time as it needs.

Intel for Alder and Raptor Lake with Microsoft came with solution to prevent background application to run on P-cores when playing games but this sounds like excessive solution which often can lead to underutilization of CPU and much longer run time of background tasks than if it was allowed to use free P-core resources. In this case background performance can be better with this Thread Director thing disabled and Hyper Threading also disabled to not affect game performance.

I see massive increase in power consumption and fan noise with HT enabled for rather unimpressive performance boost in multi-threaded applications. This alone makes HT rather hard sell and performance issues with Thread Director or without it, in games or even in desktop applications, make it even more pointless to have HT enabled. The only advantage is that if I really need certain multi-threaded task to get done as quickly as possible and I either keep it foreground or ignore decreased performance when doing other things then HT will give some extra performance.

What are your opinions on this topic?

 

[toast0]

HyperThreads are kind of meh. Super great when it was single core vs two hyperthreads, and there's use cases where it helps a lot, but if you're experiencing negatives, turn it off and don't worry about it.

If you've got all the speculative execution security mitigations enabled, hyperthreading probably just makes a lot more cache flushing work. For security purposes, you probably shouldn't let two threads from different processes run on the same core at any given time, but I don't know if any OS scheduler does a good job with that, and if they do, I'd wonder if they are going to try to squeeze your 8-thread game (or whatever is high priority for you) into 4 cores with HT, rather than taking up 8 cores if you have them.

 

[XoR_]

HyperThreads are kind of meh. Super great when it was single core vs two hyperthreads, and there's use cases where it helps a lot, but if you're experiencing negatives, turn it off and don't worry about it.

Pentium 4 line of processors had only one redeeming feature and it was Hyper Threading
Single core single thread processors were really bad at multithreading and would often get too busy handling high priority task or I/O request to process user interface leading to mediocre user experience. Second thread, even if not true second core/processor, gave Pentium 4 with HyperThreading room to breathe and in user responsiveness these processors was much better than later otherwise much faster Athlons 64. Especially in cases with lots of background network and I/O operations.

For dual cores having HT already brought scheduling issues but at this point benefits still outweighed the issues. It got only downhill trent from there. Cannot imagine 32 threads improving user experience vs 24 threads on 13900K or 32 threads improving it vs 16 threads on Ryzen 9.

If you've got all the speculative execution security mitigations enabled, hyperthreading probably just makes a lot more cache flushing work. For security purposes, you probably shouldn't let two threads from different processes run on the same core at any given time, but I don't know if any OS scheduler does a good job with that, and if they do, I'd wonder if they are going to try to squeeze your 8-thread game (or whatever is high priority for you) into 4 cores with HT, rather than taking up 8 cores if you have them.

I actually didn't think about security at all but as you mentioned it I remembered reading about
Example article https://www.techradar.com/news/inte...-you-disable-hyper-threading-linux-dev-claims

---
Not sure running 8 thread game on 4c/8t CPU but I remember seeing case where game would run fine on 6c/6t CPU but would stutter on 4c/8t CPU - probably because games are programmed in such a way to prevent using two threads of the same core so do not take advantage of HT threads. In this specific case where you run out of cores it might be detrimental to performance to not allow to use HT as fallback. GPU drivers can use HT and GPU drivers can often be very CPU intensive thus in some CPU configurations HT or SMT might still make sense - though results from various tests indicate its hit or miss with performance on eg 8c/16t CPU allover the place with HT off vs on.

For 13600K I deemed HT completely pointless and I wouldn't want HT threads to be used for anything.
Would maybe keep HT enabled but change affinity of everything to not be allowed to use HT unless I specifically need it to but it itself isn't always good solution as programs could change CPU affinity of their own process and/or expect certain amount of threads to be available and ot having them run worse. GZDoom in true color software mode run straight worse with HT than without it. E-cores do improve performance. Given why this happen in this specific game only shows what HT threads are worth vs E-cores in situations where game actually tries to use HT cores.

Also HT increases power consumption so much I would need to drop clock from where it is at stock rather than OC to get to the same fan noise levels as I have with OC. Cinebench MT scores are still lower but almost everything runs much better.

 

[uOpt]

You can also run out of cache by running 2 threads on one core, especially L1 data cache. Or you can run out of RAM when you run 2x as many processes. It is not rare to get overall lower throughput with HT. Normally I measure around 18% benefit.

I have hyperthreading off in my server, mostly for security reasons (some hardware attacks are possible with HT) and my gaming machine is a 9700K which doesn't have HT in the first place. I don't miss it. I have it on in my FreeBSD machine that does `make world` regularly and has enough RAM.

 

[zandor]

It really depends on what you're running. Can be anywhere from +30% or so to it's slower turn it off. It's the same at work. We have some production apps at work we run with hyperthreading and some we turn it off for. In both cases the app runs better with it's preferred HT setting. I don't think you can make a general argument for gaming either unless maybe you have a lot of cores. I wouldn't be surprised if a 7950X or i9-13900K generally ran games better with HT off, but I bet it's much more of a mixed bag for smaller chips.

 

[xDiVolatilX]

This is a great question. Looking for more answers myself. I have a gaming system at home with a 13900KS & I have disabled 12 E-cores also disabled hyperthreading. I have not seen the CPU be utilized enough in any game to turn the features on. If there is any evidence of more than 12 cores being beneficial for gaming or hyperthreading on I will entertain turning them all on, but it just seems like a bunch of voltage to a lot of unused cores?

 

[GotNoRice]

I think that it makes sense to leave hyperthreading enabled personally. Modern operating systems, particularly Windows 11, are much better at handling thread scheduling than older operating systems. HT is easy compared to other things that the OS has to deal with these days, such as scheduling threads on E cores vs P cores on Intel chips, or putting threads on the correct CCD on newer AMD chips.

Go back in time a decade or more and HT was far less useful, especially on chips that already had multiple cores. It's the reason why the i5-2500k was so popular, as opposed to the i7-2600k. Back then, average core usage was much lower and there was almost no real-world performance difference between 4c/4t and 4c/8t. But fast forward to today, and there is now a significant, tangible difference between 4c/4t and 4c/8t. I had an i5-2500 in my HTPC and it couldn't handle software decoding 4K x265 (GPU in HTPC is too old to do it in hardware). But swapping that out for a cheap Sandy Bridge Xeon with Hyper-threading was enough to make the difference. Both chips used the same architecture; almost the only difference between the chips was Hyper-threading, but that gave it a good ~30% boost when all of the threads were being used.

Of course, you could say that with core counts increasing again, hyper-threading has become less necessary again, but one thing that is for sure is that core count usage is on the rise. The number of things typically running in the background while gaming is also on the rise. I would at least run some of your own benchmarks and only disable it if you can conclusively show that hyper-threading is slowing things down, otherwise you're just throwing away free processing power. There are times, such as right after booting up, when a surprising number of tasks can be running at once.

 

[kalston]

As a gamer I found that disabling it was more often very harmful (last few releases of Total War, Battlefield (64/128p matches) & Ubisoft open world games saw a 30%+ drop and terrible 1% lows for example) rather than just leaving it enabled and losing 0-5% in a couple of older games that already ran okay.

Major caveat: testing done with 4c/8t (4790k), 6c/12t (8700k) and 8c/16t (11900k) CPUs only!

I went from Rocket Lake to AMD 7800 X3D now so I have no idea about 12th and and 13th gen Intel with their ecores. I just know that ecores are way better than HT, so I too would think that HT will be useless/harmful to most people. But then you might still be able to find some games that scale past 14 cores (13600k with HT disabled since that example was given) or maybe HT will help you because otherwise some cores get pegged at 100% or something.

 

[daglesj]

Seen several videos on YouTube that basically state it makes it more worthwhile to enable HT that the odd occasions it benefits switching off.

 

[XoR_]

This is a great question. Looking for more answers myself. I have a gaming system at home with a 13900KS & I have disabled 12 E-cores also disabled hyperthreading. I have not seen the CPU be utilized enough in any game to turn the features on. If there is any evidence of more than 12 cores being beneficial for gaming or hyperthreading on I will entertain turning them all on, but it just seems like a bunch of voltage to a lot of unused cores?

Your settings might as well be most optimal - but only if you overclocked your CPU and found disabling these 12 E-cores actually hindered getting maximum clock at lowest possible voltage.
HT I would not even consider enabling - more (potential*) issues than benefits.
E-cores imho you should consider enabling - but only if you do from time to time need more processing power by applications which can use all these E-cores all the while having them enabled wouldn't make computer loud enough for it to be worse compromise than having these application finish whatever they are doing slower.

For most people modern CPUs have more cores than they can possibly utilize let alone need and getting higher model with more cores just to get cores with larger caches and able to run higher clocks make perfect sense.
It even makes more sense to disable HT/SMT and some cores to make fan not get super loud and allow for higher OC (because single thread performance matters in almost everything done on computer) than pretending one needs all these cores for "productivity" - because let's be clear, most people do not do anything productive on their PC. I rarely do anything productive and even described scenarios of running long running demanding program in background application in background barely affects me. Its more having to Prime95 AVX2 validate OC for me that makes HT so terrible of an option and why I disabled it.

*) If HT can be an issue depends on Windows 11 settings. By default Win11 is configured to push all background applications to E-cores when game is running. This is actually good behavior for mostly game-oriented users as HT in this case won't interfere with games even if running heavy applications. This is why even if imho HT makes no sense on Alder/Raptor Lake it makes it less of an issue on these CPUs than on other CPU architectures

 

[LukeTbk]

Like you say maybe more on the AR/RL side
Maybe need a recheck, but on AMD side back with the 3900x
https://www.techspot.com/review/1882-ryzen-9-smt-on-vs-off/
You could gain 1% average fps and loosing 1% in 1% average low FPS by disabling SMT. Some specific title seem to have an effect, but overall as a general rules seem not worth it.

Same for the 5xxx series:
https://www.anandtech.com/show/1626...-multithreading-on-zen-3-and-amd-ryzen-5000/5

And those suite benchmark test run on perfect fresh installed machine with little else than the game running when it run, the ideal SMT off scenario to shine.

Also on AMD side, it is very clear that you go 6 or 8 core for gaming only type of machine and in those case SMT off became even less obvious, if you go above a 7800x or x3d, you usually love having a lot of threads.

From I bit of reading, it seem it almost always make sense for home system to have HT on, it is more for special servers, supercomputer and other specialized task where it become a more interesting question (Security, very limited to not enough memory bandwith per core anyways and so on).

 

[Keljian]

30%. That is the difference in performance without HT on average.

The thread scheduler on intel uses HT as a fallback. First Pcores, then Ecores, then HT if P cores are active. Without it, your pc ends up less power efficient for light to moderate loads.

 

[XoR_]

And those suite benchmark test run on perfect fresh installed machine with little else than the game running when it run, the ideal SMT off scenario to shine.

Actually the reverse would be true - the more background processes the higher chance Windows will schedule background process on the same core as one used by game if you have SMT/HT enabled. This is also the only explanation.

Also on AMD side, it is very clear that you go 6 or 8 core for gaming only type of machine and in those case SMT off became even less obvious, if you go above a 7800x or x3d, you usually love having a lot of threads.

Yeah... there is also lots of talk about 'productivity' in discussions about 16+ core CPUs like we should believe typical people habits in using computers suddenly changed and now everyone is so in to 'productivity' they actually need all these cores... I do not buy it.
Of course having twice as many threads in task manager looks better... until you actually need to check CPU resource usage which makes it all confusing having HT/SMT

From I bit of reading, it seem it almost always make sense for home system to have HT on, it is more for special servers, supercomputer and other specialized task where it become a more interesting question (Security, very limited to not enough memory bandwith per core anyways and so on).

Definitely people will claim HT is great but this is mostly because of Cinebench scores and having no knowledge how it might actually reduce performance.
In the end disabling feature which is there to win benchmarks for CPU manufacturers is hard sell without solid proof and proof not being provided by typical benchmarker - who themselves are usually not knowledgeable enough to say anything about results and not able to figure out test scenarios to challenge typical benchmarks - also because simple to consume content sells better and avoids making some people triggered - as if you show that this feature which wins you higher Cinebench score is kinda problematic and you now have to consider your use cases and maybe disable it... no one want to hear that so this is not the content you will likely hear.


Anyways, dry talk is probably not the way to go about presenting issue so I need to put more effort in to it...

 

[LukeTbk]

in discussions about 16+ core CPUs like we should believe typical people habits in using computers suddenly changed and now everyone is so in to 'productivity' they actually need all these cores...

One aspect is that people that do reviews nowadays are into productivity a lot, they do not type html text in notepad anymore, they capture, edit, and compress videos they upload, so it became a part of the focus to almost all of them, I would imagine in the Apple world to be even more prevalent. Maybe it get too prevalent

But who that not a lot in "productivity' and in cinebench score is relevant to me care for a second about 12-16 core CPU, reading those review and in the market to buy one ? It would make obvious sense for people that review those products to cover that.

Actually the reverse would be true - the more background processes the higher chance Windows will schedule background process on the same core as one used by game if you have SMT/HT enabled. This is also the only explanation.

But say you have 8 core, they will in some game be all used a bit or a lot by the game, so any background task will use the same core as one used by game, do you it to happen with SMT/HT on or off ?

 

[XoR_]

But say you have 8 core, they will in some game be all used a bit or a lot by the game, so any background task will use the same core as one used by game, do you it to happen with SMT/HT on or off ?

If game uses all CPU time of all cores then without HT background process will stal. With HT it will execute normally albeit at reduced performance and at the same time it will reduce performance of the game.

High priority tasks are different in that they execute no matter what and in this case without HT they will interrupt game threads to execute and with HT they will execute in parallel. This is the only scenario which makes HT beneficial in games.

Personally I think its much more likely to experience lower performance in games with HT by running heavy task(s) in the background (or lots and lots of smaller tasks) than improvements given by HT/SMT in handling high priority requests/tasks. Especially if game doesn't use 100% CPU time in which case random high priority request won't take any CPU time from the game. Eight cores pretty much guarantees (at least today) you have plenty of CPU time to handle background tasks and this will also be true when games can use 8 threads because usually actual core utilization in multi-threaded games is less than 100%.

More extreme example of these issues was present on very first Pentium 4 with HyperThreading and HT was improving system responsiveness exactly because it allowed high priority tasks to be executed in parallel. Even given that at the time PC architecture was pretty terrible compared to todays and we had shared buses, used keyboard/mouse which used interrupts and of course had only one core this still didn't mean games ran better with HT enabled. Opposite was universally true and it was much more likely background processes stealing CPU performance from games lowered performance and in some cases significantly.

Say you had lots of Internet Explorer windows running in background and using eg. 40% of CPU time. Without HT this was non-issue because IE didn't get any CPU time in this case because game used all available CPU time (except higher priority request of course). With HT your IE happily executed on virtual copies of your core and this did reduce performance available to the games - therefore people would notice HT hindering performance and sometime opt for hard solutions like disabling HT.

BTW. Windows 7 mitigated one issue with HT but it was different issue than what I describe. It was related to multi threaded applications which couldn't use as many threads as hardware threads being scheduled on threads of single core before first trying to schedule them on threads from different cores. I mention this because I have impression some people thing Windows patched all issues with HT. This is far from being true.

And on the topic of Windows fixing things: Windows 11 on Alder/Raptor Lake kinda fixed issue I described but this solution is not solution but workaround and is whole different bag of worms and can lead to serious performance degradation of background tasks - good for games but terrible for actual productivity... at least games run well so I guess all is nice and dandy?

One aspect is that people that do reviews nowadays are into productivity a lot, they do not type html text in notepad anymore, they capture, edit, and compress videos they upload, so it became a part of the focus to almost all of them, I would imagine in the Apple word to be even more prevalent. Maybe it get too prevalent

Typically productivity stuff is done either on different PC than what they use for benchmarks or at different time. Not relevant.

When assessing things like HT it would make perfect sense to modify test scenarios to test some of the typical scenarios which some people might experience otherwise conclusions might be wrong.
It is obvious getting eg. 130% better performance when rendering video is great but if you make article about if HT is good for your gaming experience and only test scenario where HT cannot really show its flaws (scenarios which personally I experience and I can imagine lots of people do too) then such benchmarks do not tell the whole story and conclusions from them might be misleading.

That said I think this whole topic is too complex for most people to even bother trying to figure how it all works let alone know enough how it works to come up with scenarios. Not to mention most reviewers/benchmarks want to make predictable boring content and not reveal things which might make them themselves uncomfortable or their audience so even if original thought crosses their mind they will just ignore it.

But who that not a lot in "productivity' and in cinebench score is relevant to me care for a second about 12-16 core CPU, reading those review and in the market to buy one ? It would make obvious sense for people that review those products to cover that.

Lots of people buy computers for gaming only and if they need 16 core CPU or Intel's 24 core CPU doesn't really matter - if you can afford product which will give you more performance in what you want and also have some more oomph on top of that the better.

This however also means because all these CPUs are recommended to 'productivity' while cheaper CPUs being recommended for gamers and "productivity" being new/old buzzword then at least some people will say that they got the more expensive product because of 'productivity' even if the use cases these people have in mind are purely hypothetical. When AMD released 5800X3D and people jumped to it from their 5950X do you think how many out of these people used "productivity" as justification when buying 5950X? I would say lots and lots of people. Its only natural.

 

[LukeTbk]

Typically productivity stuff is done either on different PC than what they use for benchmarks or at different time. Not relevant.

Well obviously, I mean that the people that do review now have a lot of interested in their stuff, because it is what they do as a job.

A bit like some reviewer past more than half the time during the review of a CPU on what they do in games because they like games a lot.

 

[XoR_]

CPU set to 5GHz P-cores 4.3GHz E-cores. Power plan set to best performance to avoid pushing background tasks to E-cores*
Game Cyberpunk 2077 Ultra RT but low resolution of 720p with FSR Ultra Performance to make it CPU limited
7-zip is 32MB benchmark running on all available hardware threads so 20 with HT and 14 without HT
Each test was executed 6 times and averages were taken except "actual min" value which is lowest recorded fps value

As can be seen with both HT enabled and disabled there is some performance some when there is heavy background tasks activity.
This loss is however clearly higher with HT being enabled.

Of course disabling HT entirely is not the most optimal way to maximize performance.
So can however be just using defaults* with regards to productivity tasks.
Adjusting process affinity with eg. Process Lasso and/or scripts can give users best flexibility and ability to give more performance where it is actually needed at any given time. It is however most bothersome and involved way to approach this issue.
Personally I just want one click solution to get more stable and predictable solution and disabling HT is this solution. Helps with fan noise and/or overclocking. I will post power related results when I get to make proper measurements.

*) If it didn't register by now: on Alder/Raptor Lake the issue with default settings is background process will be prevented from using P-cores when Windows decides this is necessary to keep foreground application most responsive. This translates to possibility productivity task being limited to E-cores when user decides they might rather do something else with their computer than stare at progress bar. This surely improves game and responsiveness of any foreground application performance but is terrible solution for actual productivity.
More info: https://www.xmg.gg/en/news-tips-and-tricks-rendering-e-cores/

BTW. This default behavior on latest Intel systems is imho best kept as it is by default on mostly gaming systems which only see productivity applications very occasionally and of low time critical as is imho how typical home computers are being used by gamers. In any way someone who uses computer for actual productivity probably noticed this issue already and mitigated it d

 

[pippenainteasy]

As a gamer I found that disabling it was more often very harmful (last few releases of Total War, Battlefield (64/128p matches) & Ubisoft open world games saw a 30%+ drop and terrible 1% lows for example) rather than just leaving it enabled and losing 0-5% in a couple of older games that already ran okay.

Major caveat: testing done with 4c/8t (4790k), 6c/12t (8700k) and 8c/16t (11900k) CPUs only!

I went from Rocket Lake to AMD 7800 X3D now so I have no idea about 12th and and 13th gen Intel with their ecores. I just know that ecores are way better than HT, so I too would think that HT will be useless/harmful to most people. But then you might still be able to find some games that scale past 14 cores (13600k with HT disabled since that example was given) or maybe HT will help you because otherwise some cores get pegged at 100% or something.

Once you have enough physical cores to cover all workloads, HT only becomes a detriment. If you try gaming on a 13900k for example you'll find in 95% of cases you'll see a 2-5% framerate regression with HT on vs off. Most games only use 8 threads but some (like the frostbite engine) scale up to 12 threads.

 

[LukeTbk]

This loss is however clearly higher with HT being enabled.

I feel one would need to look at the time took / speed 7-zip goes ? Outside wanting to purely go faster in a single specific app scenario of course (if you make something use all threads that probably the main thing you want ito go fast and not cyberpunk).

One obvious possible issue is comparing in one scenario 20 threads of s-zip versus only 14 threads in the others, making it an obvious case where one should hurt the rest of the system much more.

I feel like you need to either use 20 or 14 (or maybe more interesting-realistic in the case of someone having not much but some background task, 6-7) but at least about the same if not the same for both.

 

[GDI Lord]

Can be anywhere from +30% or so to it's slower turn it off. It's the same at work. We have some production apps at work we run with hyperthreading and some we turn it off for.

We get up to +80% with HT on with some workloads, never dipping below 20-ish% from what I've seen. Possible, but I haven't seen it.

Are you able to share what workloads do you experience a negative return on, please?

 

[Prince Valiant]

CPU set to 5GHz P-cores 4.3GHz E-cores. Power plan set to best performance to avoid pushing background tasks to E-cores*
Game Cyberpunk 2077 Ultra RT but low resolution of 720p with FSR Ultra Performance to make it CPU limited
7-zip is 32MB benchmark running on all available hardware threads so 20 with HT and 14 without HT

To be clear, is 7zip processing something on a different drive than the game is running on? (I snipped most of the post to not clutter with the quote.)

 

[toast0]

We get up to +80% with HT on with some workloads, never dipping below 20-ish% from what I've seen. Possible, but I haven't seen it.

Are you able to share what workloads do you experience a negative return on, please?

I've needed to turn off hyperthreads when doing some haproxy optimizing in tcpmode. But I also ended up having to ignore cores for that one, dual 14-core machines, but the nics only had 16 queues, so lots of idle cores because cross core communication in a highly network i/o boubd workload is much worse than idle cores. Would have been better to use a right sized single socket server, but had to reuse what we had. That's a fairly unusual use case though; more often there's a mix of network i/o and cpu processing and memory fetching, and HT can often help. Easy enough to turn it on and off to test though.

 

[kalston]

Once you have enough physical cores to cover all workloads, HT only becomes a detriment. If you try gaming on a 13900k for example you'll find in 95% of cases you'll see a 2-5% framerate regression with HT on vs off. Most games only use 8 threads but some (like the frostbite engine) scale up to 12 threads.

Yup that's basically what I was saying. With the CPUs I have used, I never had enough physical cores that disabling HT made any sense because of the dramatic loss in several different games that I play - while I gained only a puny 0-5% by disabling HT in the other titles.

But starting with Alder Lake, the number of physical cores went up dramatically so I assume gamers are better off without HT now. But I have not tested this theory myself. And I went for a 7800 X3D upgrade so I'll leave it up to other people to do the testing for now.

 

[zandor]

We get up to +80% with HT on with some workloads, never dipping below 20-ish% from what I've seen. Possible, but I haven't seen it.

Are you able to share what workloads do you experience a negative return on, please?

I can't share specific details, but they're low latency applications. End to end processing time is very important but the machine is lightly loaded compared to what it could handle in terms of throughput. So we're using a totally different scale to measure compared to a lot of applications. It's not how many things per second you can process, but rather how many microseconds it takes to process one request and response.

 

[III_Slyflyer_III]

Technically on AMD over here right now, but I have generally always left HT on with Intel and now SMT on AMD. In my specific setup, having 2 CCDs on a 5950X, the "penalty" of having SMT on in what I have seen is way less than having SMT off and incurring a latency penalty (adding stutter) by having tasks for a game cross over to the other CCD if using more than 8 cores.

 

[D-EJ915]

We get up to +80% with HT on with some workloads, never dipping below 20-ish% from what I've seen. Possible, but I haven't seen it.

Are you able to share what workloads do you experience a negative return on, please?

On X299/C621 using HT on 10+ core CPUs would nuke perf in cpu heavy games like FF14. Regular desktop chips didn't have that issue so it seemed to be something with that specific architecture since the 6950x didn't have that issue either.

 

[SvenBent]

I would argue it make no sense to turn it off.

the drawbacks fomr Hyper threading can easily be fixced software wise and this way you don't have to restart to get the benefits/drawbacks enabled/disabled

My tool Project mercury you can enable a fix for "SMT conflicts" which will activare on your main application ( games) so that they dont utilize hyperthreading but elevate it for the back ground task.
This way you main application gets "RAW cores" and it can help speed up a lot of games.

Here is one review of it
https://www.reddit.com/r/BattleRite/comments/97vv24/serious_performance_boost_on_ryzen_with_project/

and its 100% free. i made it mostly for myself and juit wanted to share it for other people who uses their computer both for heavy CPU usage and for games

 

[bigdogchris]

I'm not aware of situations where you would want to disable HT unless you are trying to eliminate variables while OCing. There's also discussion regarding bare metal hypervisors as well. But I think overall, it should be left on unless there is a known valid reason to turn it off.

HT does help. I clearly remember back when the first single core hyperthreaded CPU's came out. The performance boost over single core to multitasking was substantial.

 

[III_Slyflyer_III]

Contrary to popular modern belief, there is no reason to leave SMT/HT on in modern systems when GAMING is the primary task once you understand how modern CPUs work.

Check this post out, I can prove it across multiple games if anyone wants the results... I now leave my system with SMT off at all times, unless I need SMT for a specific workload. Plus you get the benefit of a much cooler system as well!

https://hardforum.com/threads/x3d-or-is-it-amd-single-core-torture.2028234/page-2#post-1045665053

 

[LukeTbk]

Contrary to popular modern belief, there is no reason to leave SMT/HT on in modern systems when GAMING is the primary task once you understand how modern CPUs work.

Well I have no doubt that it would be the case for a 16 core CPU with no background task, but is there any reason to buy a 16 core CPU when gaming is the primary task, that just a weird premise, how much better than a 7800x3d with SMT on would a 7950x do at gaming ?

 

[III_Slyflyer_III]

Well I have no doubt that it would be the case for a 16 core CPU with no background task, but is there any reason to buy a 16 core CPU when gaming is the primary task, that just a weird premise, how much better than a 7800x3d with SMT on would a 7950x do at gaming ?

Well, that's a different architecture and 2 different CPU types (X3D vs. Not). How memory starved is AM5 and does the 3D Cache matter as much on that platform? Would also depend on CPU and Memory configurations too I am sure. Some people (like me) tune every last drop out of a PC, others just hook it all up, go and call it a day.

For strictly gaming, different strokes for different folks really. My PC gets used for work as well as gaming. But even if it was ONLY gaming, I generally always spring for the absolute best CPU at the time when I move to a new platform, it is a hobby for me that I dedicate money for, so the money is not a problem. Someone else may not have that same thought and would rather save the cash with a lower end CPU (nothing wrong with that). The problem remains though; SMT/HT splits physical threads, I imagine it can still cause micro stutters depending on the game even on lower core count CPUs, where the cross-over is though, I could not tell you.

 

[LukeTbk]

ONLY gaming

Even if it would not cause any problem, I would imagine it would be totally useless for someone with 16 core that optimize the machine for best game performance, from OS to background task, making the question under those scenario less interesting.

The difference is wild too,was the all core OC the same frequency ? between the 107 and 127 fps ?

I generally always spring for the absolute best CPU at the time when I move to a new platform

But was it the best for gaming ? 7700x seem to often "beat it"

And at lease tout of the box OC:
https://www.overclock3d.net/reviews...n_4_ryzen_7_7700x_and_ryzen_9_7950x_review/15
https://www.overclock3d.net/reviews...n_4_ryzen_7_7700x_and_ryzen_9_7950x_review/18

 

[III_Slyflyer_III]

The difference is wild too,was the all core OC the same frequency ? between the 107 and 127 fps ?

Yes, both runs were at 4.725Ghz all core OC (with the exception of the very first one which was PBO).

But was it the best for gaming ? 7700x seem to often "beat it"

I built this setup when the 5xxx series and Intel 11 Series were out... Intel 12 had not even come out yet and AM5 was probably still in the engineering / testing phase at AMD... lol. Built this setup in March / April 2021 timeframe.