• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Ryzen 7950x faster with half of its cores disabled in games

I'm curious about the effect of SMT off, too.
 
That is nothing new. This has been a known fact since Zen 2 and dual core complexes was introdused. That also why Ryzen master has a gaming mode. It simply dissable one core complex for best possible gaming performance. what happens is that if game load are split between the two complex. It increase latensy and lower gaming performance and latensy.

Gaming on multi core complex zen cpu´s like 3900, 5950X and 7950X, can gain from dissable one complex when gaming.

About SMT or hyper threading. Years back it was more a problem with decreased performance when SMT/HT is on. Today however thesae problems are resolved. The only new game i can remember can suffer from performance loss with SMT is Far cry 6 and maybe other games using the same game engine. Else it is not a problem really today. I have never dissable SMT on my CPU´s, only 1 core complex on my 5950X when i game.
 
That is nothing new. This has been a known fact since Zen 2 and dual core complexes was introdused. That also why Ryzen master has a gaming mode. It simply dissable one core complex for best possible gaming performance. what happens is that if game load are split between the two complex. It increase latensy and lower gaming performance and latensy.

Gaming on multi core complex zen cpu´s like 3900, 5950X and 7950X, can gain from dissable one complex when gaming.
AMD or Microsoft should be able to fix this somehow and make software that doesnt need too many cores to work only on one but i think it is a bit tricky to do this automatically and the other option is to make profiles like in GPU drivers but probably thats not practical.
 
I want to chime in on this as I am very interested in this as going beyond 8 cores may be a problem.

Is it just a problem with Windows 11 22H2 and the bug. Or is it a problem in general its just that other Windows versions know to avoid scheduling game threads on 2 different CCDs to avoid it?? Cause if it is the later being a problem, that would mean any game that scales to more than 8 cores could take a severe drop in 1% or 0.1% lows due to latency hit of inter CCD communication which is not good.

So also in such case, a 7700X or 5800X is better than a 7900X or 5900X as it has an 8 core CCD where as the later 2 have dual 6 core CCDs. Though benchmarks for 5900X indicate that it outperforms the 5800X in like every game even 1% lows. Though I wonder if that is because there are almost no games scale beyond 6 cores and 12 threads so it stays on one CCD anyways and the few games that could scale to 8 cores 16 threads would take a hit and be much better on 5800X/7700X??

Or you could get a 5950X or 7950X and lock the game threads to the fastest CCD to not run into the issue.

Or is it even always an issue?? And no I do not mean to the point of a bad stutter. I am talking a big hit to 1% or 0.1% low at any time even if it goes unnoticed. Still unacceptable for it to happen though.

That is nothing new. This has been a known fact since Zen 2 and dual core complexes was introdused. That also why Ryzen master has a gaming mode. It simply dissable one core complex for best possible gaming performance. what happens is that if game load are split between the two complex. It increase latensy and lower gaming performance and latensy.

Gaming on multi core complex zen cpu´s like 3900, 5950X and 7950X, can gain from dissable one complex when gaming.

About SMT or hyper threading. Years back it was more a problem with decreased performance when SMT/HT is on. Today however thesae problems are resolved. The only new game i can remember can suffer from performance loss with SMT is Far cry 6 and maybe other games using the same game engine. Else it is not a problem really today. I have never dissable SMT on my CPU´s, only 1 core complex on my 5950X when i game.


Yeah with Zen 2, isn't it an issue even with the 8 core parts as they only have 4 cores on a single CCX??

Or no because they are on same CCD despite different CCXs where as the Ryzen 9 parts from Zen 2 to Zen 4 all have any more than 8 cores crossing to a new CCD??
 
Hi,
Usually just disabling hyper threading is all it takes to do the same.
 
Hi,
Usually just disabling hyper threading is all it takes to do the same.


Even having both CCDs active? Then less room for threads and more likely they hop from one CCD to another which is supposedly bad for gaming?? Or is it not an issue at all anymore and would run games great as a 16 core 16 thread chip even with dual 8 core CCDs??
 
Even having both CCDs active? Then less room for threads and more likely they hop from one CCD to another which is supposedly bad for gaming?? Or is it not an issue at all anymore and would run games great as a 16 core 16 thread chip even with dual 8 core CCDs??
Hi,
16 cores still doubt you'd be missing threads anytime soon gaming
Just run heaven benchmark it will perform better without hyper threading.
Once you go past a point hyper threading isn't needed and on intel it's a security risk waiting to happen.
AMD might handle threading better IDK.
 
Hi,
16 cores still doubt you'd be missing threads anytime soon gaming
Just run heaven benchmark it will perform better without hyper threading.
Once you go past a point hyper threading isn't needed and on intel it's a security risk waiting to happen.
AMD might handle threading better IDK.


Yeah true with 16 cores no need for SMT. So is there no penalty for cross CCD thread swapping?? Or is it that no game scales past 8 threads so a one 8 core CCD is enough for them while the other CCD handles background tasks?? Or is there no problem at all with game threads spread out between both CCDs despite latency jump for cross communication or switching between CCDs??
 
The issue is that with two CCDs there is an inter CCD latency penalty if any game process handles are on both CCDs. HT on/off makes little to no difference, usually 1-2% plus or minus performance, depending on game (I had 5950X), disabling one CCD however forces all process handles to be on the same CCD (obviously), eliminating the chance for that latency penalty.

Win 11 and AMD will probably eventually fix this.

The 7950X is still faster than the 7700X in stock configuration.

Zen is highly sensitive to latency, it's why extremely quick memory in 1:1 is needed for gaming performance similar to Intel, and why the X3D has such an advantage despite lower clocks, due to the large cache meaning memory is accessed less frequently for latency sensitive things.

This is also why 12/13xxxK is still king for gaming, due to monolithic chip with consistent, low latency and better memory controllers.
 
Yeah true with 16 cores no need for SMT. So is there no penalty for cross CCD thread swapping?? Or is it that no game scales past 8 threads so a one 8 core CCD is enough for them while the other CCD handles background tasks?? Or is there no problem at all with game threads spread out between both CCDs despite latency jump for cross communication or switching between CCDs??
Hi,
High thread count is always best rendering anything image/ video and likely folding...
Otherwise gaming think 12 threads is plenty on majority of games.

Fun fact just running cmd as admin using winsat mem on 14 cores or 28 threads on 9940x is a pretty large difference this is just system showing difference on a easy test.

That hyper threading isn't that great
 
AMD or Microsoft should be able to fix this somehow and make software that doesnt need too many cores to work only on one but i think it is a bit tricky to do this automatically and the other option is to make profiles like in GPU drivers but probably thats not practical.
I fully agree, both companies make parts that can and do park core's into sleep mode all the time anyway but, they are not automatically parking the right core's at the right time or even better holding processes within bounded thread profiles.

It really should in both camps be the case that a game uses all the best core's and uses the rest for ALL other background tasks.

Or whatever suits best in that moment and since games have configs and profiles how f#@£ing hard is it for a OS to know a game does well on upto 8 core's for example.
And to run it on one set of core's.
 
  • Like
Reactions: izy
The issue is that with two CCDs there is an inter CCD latency penalty if any game process handles are on both CCDs. HT on/off makes little to no difference, usually 1-2% plus or minus performance, depending on game (I had 5950X), disabling one CCD however forces all process handles to be on the same CCD (obviously), eliminating the chance for that latency penalty.

Win 11 and AMD will probably eventually fix this.

The 7950X is still faster than the 7700X in stock configuration.

Zen is highly sensitive to latency, it's why extremely quick memory in 1:1 is needed for gaming performance similar to Intel, and why the X3D has such an advantage despite lower clocks, due to the large cache meaning memory is accessed less frequently for latency sensitive things.

This is also why 12/13xxxK is still king for gaming, due to monolithic chip with consistent, low latency and better memory controllers.


Thanks for explanation as that does make sense. Though question. Process handles are subparts of same process correct?? So if a game needs or is abl;e to benefit from scaling more than 8 cores or 6 in the case of a 7900X/5900X, it is fine with no issue on second CCD as long as long as all process handles for a given process stay on the same CCD?? But separate processes themselves can be on different CCDs with no issue?

And same would be true for any app even highly threaded apps perfect at parallelization where double core count gives double performance all else being equal?
 
Hi,
Cores with hyper threading off also run cooler except if they're e cores :laugh:
 
Hi,
High thread count is always best rendering anything image/ video and likely folding...
Otherwise gaming think 12 threads is plenty on majority of games.

Fun fact just running cmd as admin using winsat mem on 14 cores or 28 threads on 9940x is a pretty large difference this is just system showing difference on a easy test.

That hyper threading isn't that great

So in your opinion, SMT and HT is not very good. I used to kind of think the same way with high core counts though some state including I believe dgianstefani in another thread a while back it allows CPU to be much more efficient in multi threading even with lots of cores.

Like how come 4 core 8 thread CPUs were relevant for gaming a lot longer than 4 core 4 thread CPUs if HT/SMT was not good for gaming. Was it mostly because games became more threaded, but those threads did not take up the full 4 cores. And because HT/SMT gives 2 threads per core it allows it to be much more efficient as long as each thread does not have to use more than 50% of a CPU core they can be divided into 2 easily. Now if 1 thread needs 100% of a CPU core and the others saturate all cores to 100% than HT/SMT advantage for gaming would vanish mostly. But because games scale to more threads, but only 1 or 2 saturate a CPU core to 100%, the others benefit form SMT/HT because the other CPU cores are not pegged 100% and can be effectively divided to multiple threads?

So when you say SMT/HT can hurt that would be because maybe another thread is trying to compete with a thread that needs 100% access to a single core, though with scheduling in Windows being proper it should be able to work around that and tell the SMT/HT 1 core is getting this thread no matter what and nothing else touches it??
 
Hi,
Older 4c-8t cpu's were killed with security flaws with intel ht prediction though so newer chips verses older chips is a whole different subject

Not sure to many gamers were willing to disable ht use a quad core just for a bigfoot security risk :laugh:
 
Hi,
Older 4c-8t cpu's were killed with security flaws with intel ht prediction though so newer chips verses older chips is a whole different subject

Not sure to many gamers were willing to disable ht use a quad core just for a bigfoot security risk :laugh:


SO was the HT in games not really helping for games on those older chips was it just the extra L3 cache??

And you say security issues with HT/SMT?? Have they been fixed by Intel in Alder Lake CPUs?? Or still issues??

And is AMD any better with it??

And in your opinion, having a 16 core Ryzen CPU its way better to have SMT off for games even high end ones?? And there should be no issues with inter CCD latency even with game processes for the same game going to different CCD as long as the handles of a particular game process are all on the same CCD??
 
Hi,
Hyper threading was made for older chips starting I believe with core 2 duo so without ht they were shit
Intel micro code threw most for a loop plus win-10 security updates but using this was pretty easy to get any low of performance back a bit

GRC | InSpectre


Newer chips none of that security stuff applies to much except for core isolation security features running amuck in 10 or now 11
If HT makes a difference one would have to test it themselves to see if the difference was slight or large or any at all I just gave a couple examples where ht made scores lower not to many scores in games just kill or be killed :laugh:
 
Hi,
Hyper threading was made for older chips starting I believe with core 2 duo so without ht they were shit
Intel micro code threw most for a loop plus win-10 security updates but using this was pretty easy to get any low of performance back a bit

GRC | InSpectre


Newer chips none of that security stuff applies to much except for core isolation security features running amuck in 10 or now 11
If HT makes a difference one would have to test it themselves to see if the difference was slight or large or any at all I just gave a couple examples where ht made scores lower not to many scores in games just kill or be killed :laugh:
HT was even before that, I had a p4 with hyperthreading.
 
HT was even before that, I had a p4 with hyperthreading.


I remember the first HT chip was the Pentium 4 3.06GHz 533MHz FSB 20 years ago November 2002. I was a senior in high school. I did not have one but remember eying it and wanted to get one and then the 3.0C dropped and I got that one following Spring with the 800MHz FSB.
 
Last edited:
The issue is that with two CCDs there is an inter CCD latency penalty if any game process handles are on both CCDs. HT on/off makes little to no difference, usually 1-2% plus or minus performance, depending on game (I had 5950X), disabling one CCD however forces all process handles to be on the same CCD (obviously), eliminating the chance for that latency penalty.

Win 11 and AMD will probably eventually fix this.

The 7950X is still faster than the 7700X in stock configuration.

Zen is highly sensitive to latency, it's why extremely quick memory in 1:1 is needed for gaming performance similar to Intel, and why the X3D has such an advantage despite lower clocks, due to the large cache meaning memory is accessed less frequently for latency sensitive things.

This is also why 12/13xxxK is still king for gaming, due to monolithic chip with consistent, low latency and better memory controllers.

If you look at 5950X reviews, it was actually the fastest in games despite having two CCDs.

Microsoft borked something when it updated the scheduler to accommodate Intel's Big-Little cores. It's pretty clear it's bouncing threads around CCDs in Win 11 when it shouldn't be.

So in your opinion, SMT and HT is not very good. I used to kind of think the same way with high core counts though some state including I believe dgianstefani in another thread a while back it allows CPU to be much more efficient in multi threading even with lots of cores.

Like how come 4 core 8 thread CPUs were relevant for gaming a lot longer than 4 core 4 thread CPUs if HT/SMT was not good for gaming. Was it mostly because games became more threaded, but those threads did not take up the full 4 cores. And because HT/SMT gives 2 threads per core it allows it to be much more efficient as long as each thread does not have to use more than 50% of a CPU core they can be divided into 2 easily. Now if 1 thread needs 100% of a CPU core and the others saturate all cores to 100% than HT/SMT advantage for gaming would vanish mostly. But because games scale to more threads, but only 1 or 2 saturate a CPU core to 100%, the others benefit form SMT/HT because the other CPU cores are not pegged 100% and can be effectively divided to multiple threads?

So when you say SMT/HT can hurt that would be because maybe another thread is trying to compete with a thread that needs 100% access to a single core, though with scheduling in Windows being proper it should be able to work around that and tell the SMT/HT 1 core is getting this thread no matter what and nothing else touches it??

It makes no sense to disable HT or SMT anymore. There's almost no performance hit in games that don't benefit and a large uplift in games that do.

Hyperthreading and SMT are similar in that they both seek to improve utilization of cores. SMT is slightly more efficient though. A good example would be CP2077. When the game launched it didn't support SMT (and still has some issues with this today). After applying an SMT patch, average FPS would increase by 20 - 30 as well as 1% lows.

So long as a game doesn't try to cram it's main thread and another heavy thread on the same core HT and SMT shouldn't come with a hit. Even if it does you are taking 2% or less at worst.
 
If you look at 5950X reviews, it was actually the fastest in games despite having two CCDs.

Microsoft borked something when it updated the scheduler to accommodate Intel's Big-Little cores. It's pretty clear it's bouncing threads around CCDs in Win 11 when it shouldn't be.
It was the fastest because it had the highest boost clock. With one CCD disabled it would still be faster than the 5800x.
 
Yep, that's what I had. I was in my first IT job and wanted to flex my "wealth" ;)


I was in the early stage of senior year of high school being Fall 2002 than graduated in June 2003 being the end. I waited then got the upgraded Pentium 4 3.0GHz with 800MHz FSB and Intel 875P chip that dropped like weeks before I actually graduated high school. Still have the motherboard in my closet an ABIT something with an 875P chipset and Gigabit LAN.
 
Well with my 5950x I leave smt off and get better clocks and performance... So with the new chips I wonder how that would work instead of turning half the chip off.
 
Back
Top