Is it worth REDUCING number of core's in CPU for DCS performance gain?

[Dangerzone]

Just had this thought. Since DCS only uses 1 thread - and I have a 8 core processor - would it be worth reducing the number of core's being used by half to 4 to reduce the heat on the CPU?

 

(My thought is that by reducing the number of active cores, I'm reducing the heat, thus allowing for a greater OC of the chip thus getting better performance out of the active cores). I'm a noob when it comes to OC'ing so have no idea whether or not this is practical and thought I'd ask here what others do?

[DimSim]

You are suggesting either manually setting CPU core affinity, or disabling hyper threading.

I'd suggest not doing either.

 

While there is some chance of overclocking a single core to higher frequency in either case, you're more likely to hamper overall performance and make it erratic (mix of faster and slower).

 

While this sim may be single thread heavy, there are still a lot of other activities going on in the background, including graphics driver, sound, input and OS. While these don't saturate cores, they still consume CPU time.

 

Others may have different opinions, but mine is that overclocking in the present day is unlikely to give you significant gains, yet very likely to decrease your system stability. There were times in the past, where safe overclocks could gain 20% real performance. Best save your money and make upgrades to your GPU or CPU over time. Or enjoy what you have and upgrade your HOTAS etc.

[Csgo GE oh yeah]

Hyperthreading is usually a bad idea for gaming. So i would turn that off (if your cpu has it) . Otherwise, no i don't think you should disable physical cores .

[Raven (Elysian Angel)]

Just had this thought. Since DCS only uses 1 thread - and I have a 8 core processor - would it be worth reducing the number of core's being used by half to 4 to reduce the heat on the CPU?

No.

On Ryzen CPUs with crazy core counts it can reduce latency. In such instances it is beneficial in gaming to turn off one of the CCDs. It's called Ryzen Game Mode.

[twistking]

Hyperthreading is usually a bad idea for gaming. So i would turn that off (if your cpu has it) . Otherwise, no i don't think you should disable physical cores .

generally this is not true. there have been instances where specific games reacted positively to disabling hyperthreading, but this could be seen of a quirk of that specific game, or that specific OS version it ran on. most of the time you can expect the same or better performance with hyperthreading enabled. if you're going for extreme OC however, this is a completely different question.

[Dangerzone]

OK - I'll keep all CPU core's active, and will play around with hyperthreading to see what happens there. Thanks!

[Cytarabine]

I use Process Lasso to make sure DCS only has affinity for the physical and not logical cores. It's probably just a placebo but it seems to help.

[Csgo GE oh yeah]

generally this is not true. there have been instances where specific games reacted positively to disabling hyperthreading, but this could be seen of a quirk of that specific game, or that specific OS version it ran on. most of the time you can expect the same or better performance with hyperthreading enabled. if you're going for extreme OC however, this is a completely different question.

 

Only in Far Cry 5 does HT positively influence gaming performance. In all other games = stuttering . Google "hypertreading gaming on/off" or something like that. HT sucks for gaming.

[Eldur]

Just funny how there's always the hint to disable HT, but yet people kept on buying CPUs with HT for gaming. When I bought mine, I did check the benchmarks. My 3570k was like 180€ and the 3770k which is identical except for it having HT and more cache (8MB instead of just 6) was like 350€, but had like 0.5% more fps in benchmarks. And I always kept recommending just getting the i5. Well, later on the i7 (now i9) came with higher clocks which then would be an actual reason to get those instead. But back in the day there literally was none, except for shelling out more cash that you could use other-wise-ly for a better graphics card or more mem.

[Sn8ke_iis]

The Hyperthreading issue depends on the chip.

 

If you have a 7700K keep HT on, if you have 8700K, 9900K, or 10600K you won't necessarily gain anything in performance by turning it off but you can overclock the core to a higher GHz and still be stable. Anytime Gamer's Nexus does tests with HT or SMT off there is usually a performance gain but depends on the game. In general games don't use hyperthreading of cores like Blender or Cinebench does.

 

If your cooling system is limited it might be worth a try to turn off some cores but if you lock affinity to 3 cores with Process Lasso the other cores stay idle and cool most of the time. Doubt there would be much difference but you never know for sure until you test it.

 

And regarding OCing, YMMV but if you bought a K chip from Intel and a Z motherboard and RAM with an XMP profile you are leaving performance on the table if you don't use them. The limiting factor for most people will be cooling. And right now rather than spend money on cooling you can just buy a better CPU or GPU and then cool that when you get bored.

 

[Thinder]

On 11/16/2020 at 2:52 PM, Raven (Elysian Angel) said:

No.

On Ryzen CPUs with crazy core counts it can reduce latency. In such instances it is beneficial in gaming to turn off one of the CCDs. It's called Ryzen Game Mode.

 

You don't need a crazy core count to do that, you can limit it to the number of cores you want but by itself, a package like Ryzen Master will do it automatically, select the best core and give it priority over the other, on the 6 cores, it translate by 4 remaining with one with the highest speed.

 

Latency is already reduced compared to Zen 2 by the Zen 3 architecture and the more direct way for each core to access the cache, so the goal here is to increase an individual core speed withing the factory-set limits of the processor, but there are ways around those settings and people did find settings allowing for 4.8GHz all corps.

 

As I understand it was done with liquid cooling, but we can expect faster core speed from this processor as players experiment and find better settings.

 

 

[Raven (Elysian Angel)]

10 hours ago, Thinder said:

You don't need a crazy core count to do that

No you are right, you don't. But it was the general idea behind it: some applications that are particularly latency-sensitive can benefit from disabling one of the CCDs entirely. But it depends on the application, so it shouldn't be done as a general rule-of-thumb for gaming.

 

And yes, one of the reasons Zen3 is so much better than Zen2 is because of the architectural changes that dramatically improved latency, which translates into IPC improvement and increased single-core performance.

[Worrazen]

On 11/16/2020 at 1:33 AM, Dangerzone said:

Just had this thought. Since DCS only uses 1 thread - and I have a 8 core processor - would it be worth reducing the number of core's being used by half to 4 to reduce the heat on the CPU?

 

(My thought is that by reducing the number of active cores, I'm reducing the heat, thus allowing for a greater OC of the chip thus getting better performance out of the active cores). I'm a noob when it comes to OC'ing so have no idea whether or not this is practical and thought I'd ask here what others do?

 

It's a long time ago DCS only used 1-2 threads, it does benefit from a quad-core CPU for a number of years now, and the ratio of workload has improved as time passed, there's around 4 significant threads and a bunch of terrain-texture-loader or "IO" threads which are responsible for texture streaming and mission loading. However that doesn't mean that work on the main thread was necessairly shifted and spread around, as time passed I think more work has been added overall and DCS became more demanding, yet that doesn't mean it's not single-thread bottlenecked, certain things will bottleneck the main thread and this will continue to be so even after Vulkan API will be supported.

 

 

 

On 11/16/2020 at 9:54 AM, Csgo GE oh yeah said:

Hyperthreading is usually a bad idea for gaming. So i would turn that off (if your cpu has it) . Otherwise, no i don't think you should disable physical cores .

 

That actually depends on game behavior, some don't like it enabled at all while some don't care whether you have enabled or disabled and will be very similar.

 

 

 

On 11/16/2020 at 3:52 PM, Raven (Elysian Angel) said:

No.

On Ryzen CPUs with crazy core counts it can reduce latency. In such instances it is beneficial in gaming to turn off one of the CCDs. It's called Ryzen Game Mode.

 

Yeah, disabling CCXs may no longer be necessary on Zen3 based CPUs, but I'm afraid the higher core-count Zen3 CPUs that are multi CCD would still have higher latencies between the CCD despite all the improvements, tho the OS thread schedulers could start to catch up. 

 

A bit of a background on this for others who aren't familiar:

 

So Zen 3 aren't 2 CCXs in a CCD anymore being glued toghether separated by circuitry, now the CCX is a proper octa-core the size of the whole CCD and all the cores are more equal*, but the latency pentaly between CCDs would still remain ofcourse, but OS thread schedulers should hopefully improve with such topologies in the future to make this a non-issue.

 

The root problem isn't actually with the amount of cores themselfs or the fact that the Zen1+2 octa-core CCDs were made up of two quad-cores (CCX) inoptimally glued together (kinda fake octa-core), but the fact that due to default OS thread scheduler behavior and default program settings, the main thread of a game would jump between CCXs and on even higher core-count CPUs between CCDs, requiring a lot of time to complete before executing again, resulting in a stutter. 

 

There can also be inconsistency in this happening, to some people and not others, or some times yes, sometimes no, because not all the DCS (games) process threads are so latency critical, work may be delayed but it won't stop the program from executing if minor threads are caught up in a inoptimal Context Switch (CS = thread starts executing on a different CPU from where it stopped on) such as the IO threads, you won't really notices those trees or textures taking a bit longer to update to higher-definition LODs and your FPS will remain stable, but when the main thread gets caught up in this inoptimal context switch (aka "thread bounce/jump") being instructed to move from CCX1 CPU2 to CCX2 CPU3 for example, a move between CCXs (or CCDs) requires more time for the CPU to do this kind of operation before the thread can start executing again so the thread can do nothing but wait, the caches have to copy over to the CCX2 CPU3 context (area) and the L3 stuff isn't shared so there's more operations invovled there, so you end up with latency big enough that you notice in practices as a stutter. 

 

Disabling a whole CCX or CCD is the fast and easy solution, but more of an emergency one, but also overkill. Setting process affinity of the DCS process to the same CCX or CCD would be the way to go.

 

 

On 11/17/2020 at 11:10 AM, Cytarabine said:

I use Process Lasso to make sure DCS only has affinity for the physical and not logical cores. It's probably just a placebo but it seems to help.

 

Tho, that's only in regards to HTT/SMT on vs off, but not in regards to Zen1+2 CCX/CCD and Zen3 CCD topology, if were talking about it as well.

So if you have a Zen based CPU, optimizing affinity for the CCX/CCD is more important than SMT, but you can do both, only pick one CCX/CCD and then also only pick 1 logical CPU per 1 Core.

 

 

 

On 1/1/2021 at 1:20 AM, Raven (Elysian Angel) said:

No you are right, you don't. But it was the general idea behind it: some applications that are particularly latency-sensitive can benefit from disabling one of the CCDs entirely. But it depends on the application, so it shouldn't be done as a general rule-of-thumb for gaming.

 

And yes, one of the reasons Zen3 is so much better than Zen2 is because of the architectural changes that dramatically improved latency, which translates into IPC improvement and increased single-core performance.

 

I don't know why the heck would Ryzen Master go startight to the nuclear option of disabling a whole CCD, must have been so bad it was a real rush to get the gamers calmed down?

 

Or do you actually mean disabling affinity to a CCD, that's a WHOLE other thing ... because the former means to disable the CPUs physically, to shut them down, so nothing else can use it ... ?!?!

 

I personally don't have an AMD PC so I don't know some of the details, but I will have Zen3 PC soon-ish so I'll see some of it. I wasn't that following all the Zen1+2 events and gaming news in general either so I guess I'm not familiar with the scope of this CCX/CCD latency issue in practice, I've just focused on the technical aspects, reviews and explanations.