Test: Setting CPU Affinity

@Hadwell: I thought 0,2,4 & 6 were the actual physical cores? DCS was thrashing core 7 (which I thought odd if it was a logical core?). Knowing Windows utilizes core 0 (and thread 1) I allowed it to stay on that physical core 6 (utilizing its thread 7),,,,, I am no expert, it just seemed sensible to me at the time. it took forever to then change all the other processes NOT to used the cores I assigned to DCS.

@Sandman1330: yes, I found it odd that it appeared to be pegging a logical core myself, but what the heck do I know? By the way, I could not find a way to change all the other processes to cores 0 thru 3 without changing the rules for EVERY specific process individually? What a pain staking task that was! Do you know of a way to do it say, globally or collectively?

I will experiment a little further and see what happens when I turn HT off in the BIOS

@Hadwell: I thought 0,2,4 & 6 were the actual physical cores? DCS was thrashing core 7 (which I thought odd if it was a logical core?). Knowing Windows utilizes core 0 (and thread 1) I allowed it to stay on that physical core 6 (utilizing its thread 7),,,,, I am no expert, it just seemed sensible to me at the time. it took forever to then change all the other processes NOT to used the cores I assigned to DCS.

yeah you're right it starts at 0, not 1...

UPDATE! I went for the Process Lasso approach as a last ditch effort with Hyper threading enabled. I set DCS to dominate cores 4,5,6,7. So 2 dedicated physical cores and 2 HT. I understand that DCS basically runs off of one core, with some understanding that it uses a 2nd to handle audio, so thats how I set it up anyway, all other process use cores 0,1,2,3. WHAT A DIFFERENCE! 1.5 runs silky smooth, the only thing with it is when I get into or over thick forestry at low and high altitude do I get stuttering. I will have to figure out what settings to tinker with to address that.

 

2.0 ran silky smooth throughout the desert,, I did not run over to the strip yet to check its performance, but Im heading in a good direction it appears. I monitored my core temps during this and I did not see anything different from before with temps. this is to be continued until I get the stuttering gone. Thank you all again! Paddyman, thanks for the tip sir!

Glad you made it :beer:

Interesting to see how differently cores and workloads are handled across the bench.

Process Lasso seems to be a top candidate for those who actually should have all it takes to run DCS smooth but somehow end up like Fakum.

Is there any way a programmer can fix this, aka tell DCS to run on physical cores only, or is this beyond ED's realm and in the depth of Windows & Intel/AMD. I have no real idea.

Thanks BitMaster! Although I have made it past a significant hurdle, I still have to resolve the stuttering related to the the forestry etc. I do feel that there is some sort of underlying issue with my new PC. Something probably needs to be changed or adjusted but I dont know yet what that is. I personally feel that this hardware should not be experiencing this issue. I was very tempted to just put my old build back in place. Its a shame to spend SO MUCH TIME trying to get this running,,, was almost ready to go back to ARMA!

Hi guys!

I´m playing around with Prosess Lasso Pro too so which cores do i have to set with my 5930k six-core cpu? HyperThreading is on so there are threads from 0-11.

Would be really great if this one could make better performance with my rift:music_whistling:

I couldnt tell you anything more than I understand NOT to set it to cores 0 & 1 because Windows uses that core primarily, then I set my core for the one that DCS was using while I was running it. I ran everything I typically run when flying DCS, I went into an instant action mission, then I Active Paused DCS, then alt-tabbed to bring up Process Lasso, and the core that stood out big time, was the one being used for DCS, so I allowed it to continue usin that and I allowed it to also use another core just for itself,, then I pain stakenly set all the other processes to the other cores via RULES

Excellent to hear Fakum, delighted it worked for you!!. I bet its a huge relief after all the work you put in.

I spent hours trouble shooting mine as well. I knew the hardware was capable of smooth VR performance but every few seconds i was getting stutters for no reason and FPS drops. I installed Process Lasso, done a basic setup and on my first play it was such a difference i almost didn't believe it. No random stutters and higher FPS.

To be honest i didn't even go as far as you and allocate all the cores individually. It was a while ago so i don't remember the details offhand, but i just used some of the basic settings. The most important being setting the DCS.exe as the high performance application. I also made DCS.exe a high priority for CPU, memory etc, and i believe there was something about power profile..

I actually went and bought it (i am not affiliated in any way) as a silent thank you for solving my problem. Thinking about i could probably wring some more performance out of it by doing the CPU affinity / core separation, but i didn't see any easy way to do it. I wish there was an option to "reserve x,y,z cores" for high priority applications and let everything else run on the other cores.

Thanks paddyman, interesting info!

"setting the DCS.exe as the high performance application. I also made DCS.exe a high priority for CPU, memory etc"

I will go look into how to do that. Many thanks sir! I am going to experiment a bit more with this and also see the results with HT off. Never had HT before, and DCS was smooth with the old i5 2500k,,,

I have HT off because i got a better overclock and the CPU is 8 physicals anyway so i can't imagine needing more. You will get better performance in DCS with less cores at higher clock frequency than lots of core at lower frequency.

From what i understand setting the bitsum highest performance power plan mode allows the CPU to be ran at full speed at all times, so no parking the cores and waking them back up. It is only turned on when the high performance process is running. Just right click on the DCS.exe and classify it as a high performance process. Then when DCS starts it automatically starts the power plan.

I would definitely do this, but after that you need to research. As i read trying to do too many permanent tweaks (setting always) and rules can lead to conflicts and might cause you other issues.

During my Overclocking I was experimenting with HT on & Off also, with it off my temps came down an average of 6-8°C on AVG, but my scores dropped substantially as well.

I have just set the prioritys for memory and CPU etc as you suggested. I actually set it for DCS, TrackIR and Helios. I will fiddle. I have been exporting profiles for sake of having saved profiles to restore back to when experimenting. I will be turning off HT today and setting up a profile for that as well to check performance. I do have one question for you though. You have more physical cores than I do, as I only have 4. My HyperThreading profile utilizes 4-7 for DCS, I selected that many because I understand it as DCS primarily only uses 1 core, but perhaps a second for Audio etc. So it was easy enough to give it cores 4-7 because in my mind, everything else should work well with the other remaining 4(if you will). Now once I turn HT off, I only have 4 cores, Im not sure If i should let DCS have 1 or 2 of them? What did you do, set up for 1 or 2?

@Sandman1330: yes, I found it odd that it appeared to be pegging a logical core myself, but what the heck do I know? By the way, I could not find a way to change all the other processes to cores 0 thru 3 without changing the rules for EVERY specific process individually? What a pain staking task that was! Do you know of a way to do it say, globally or collectively?

 

I will experiment a little further and see what happens when I turn HT off in the BIOS

I didn’t actually mess with any other processes, just DCS. Like someone else mentioned, I just set it to high priority and limited it to physical cores and let process lasso / windows scheduler do the rest. It seems to be smart enough to keep other processes off the heavily used DCS core.

I do still get some stuttering in Normandy at low level, with shadows cutting in and out (stutters continue even with shadows off). This is mostly on a mission like the A10C CAS mission with a lot of AI movement.

So I turned off HyperThreading and created a new config in Lasso to utilize only core 3 for DCS and changed all other applications to only run on cores 0-2. without anymore testing that I already did tonight, it seems to me that running without HT on was more beneficial. I still get some stuttering getting near trees, but at least I am at a playable position, more tinkering to follow.

I did notice something that may or may not be of concern? See attached screenshot of Lasso, DCS gets a lot of page faults while running. I dont know if this is something i need to understand and resolve or ?

So I turned off HyperThreading and created a new config in Lasso to utilize only core 3 for DCS and changed all other applications to only run on cores 0-2. without anymore testing that I already did tonight, it seems to me that running without HT on was more beneficial. I still get some stuttering getting near trees, but at least I am at a playable position, more tinkering to follow.

 

I did notice something that may or may not be of concern? See attached screenshot of Lasso, DCS gets a lot of page faults while running. I dont know if this is something i need to understand and resolve or ?

Just to clarify, running with HT turned off was better?

Hi Fakum,

Looking at your signature it says you are at 4gz. Have you tried to push it up to 4.2 / 4.3 ghz?. you should be able to achieve this even on air.

As you correctly said DCS only uses two physical cores. The only reason i turned HT off was i could get a better overclock on my CPU with it off (im @4.4ghz). When overclocking your CPU, if you get the same overclock with HT on as with HT off then there is probably no point in turning HT off. You will get some lower temps with HT off but that would be it.

You will get a bigger improvement increasing your clock frequency than core count. DCS likes higher clock rates. I would focus on this first before worrying too much about HT on/off.

If Process Lasso has already given you a massive improvement then your probably getting 95% of the improvement its likely to give you. by all means try tweak it better but as great as it is there is only so much it can do :)

There is also a point when you have to start playing the game :) and enjoy yourself. Honestly i am playing away now with a smooth experience and really enjoying it. you should too ;)

Not sure on your page faults. Perhaps its because you have set high priority memory for DCS, TrackIR and Helios. I just have DCS.exe as high, nothing else. Easy to test.

@ hansangb: That was a very quick comparison last night flying over some forest for about 20 seconds comparing the same flight with HT on & OFF. I will have to do a few repetitive testing to confirm.

@ paddyman: I have updated my sig (thanks for the reminder), I have been running @4.6 during this process. HT Off during overclocking yielded better temps, but lower benchmark scores @ 4.6, (100@46) so for now, HT=ON. When I said I have some tweeking to do, I was meaning graphics settings in NVIDIA & DCS GUI settings (a whole nother subject). I also believe Lasso has provided as much bulk results as I can get from it (and i am VERY grateful). I have to get the remaining stuttering resolved. I will try your suggestion though regarding the high priority settings thank you.

More testing to do, the last 95% is always the most time consuming,, not complaining, just happy I didn't pick this new rig up and smash it against the wall like I wanted too! That would have cost me a new gaming rig and probably a divorce LOL

Which benchmark?

Synthethic test doesn't represent what you will see in DCS. Use DCS as the benchmark task.

The benchmark gives you a better score with the HT on, but DCS doesn't use the HT tech then the test is useless.

Decrease of power consumption and heat output (HT off) gives you the opportunity to overclock more. This is exactly what DCS needs. Don't forget, in the DCS software, the "HT" is off.

All I want is speed.

I use AIDA64, ASUS RealBench,Cinebench,3d Mark, Heaven, Prime 95, Corsair Link, CPU ID & Real Temp to test and monitor. Scores across each dropped with HT off, but so did the temps 6°-8° C as I recall. You make a good point that I think I should consider, and that is, DCS works better at higher clock speeds. I should probably turn off HT and run at 4.7 (Which I was stable at with HT on but I did not like the temps). I think i will look into that, I dont think I will be successful getting to 4.8, but I have not put the time in to see, like I said, I hit 4.7 and stopped as the temps with HT on were benching in the high 80's C.

Only a few lucky 6700k made it past 4.7G, with or without HT.

If you opt for HT off you can also leave Process Lasso aside as you only have real cores then.

As a tip, I use CoreTemp over RealTemp. You might wanna try CT and arrange the TaskBar icons as you wish.

Do not torture your 6700k with Prime95+AVX at high clocks and temps, it will ruin your CPU !

Nothing against a few runs, but stay below 90°C all times. Mine goes 100°C and throttles on newest p95AVX and 5G/3600. It wont crash ( tho it throttles from time to time ) but I fear the smoke signs.

Anyway, I hardly see any difference from the CPU from 4.5 to 5G ( with my 980 ). Overclocking the 980 gives direct proportional more fps = GPU locked.

Maybe with my new 1080Ti it will make a difference, but I doubt. Anything at 4G or even better is enough imho. This 5G thing is an Ego thing foremost, it is no game changer or deal breaker.

I agree with you, I have seen the stats for this CPU, and quite frankly, I am happy that I made it to 4.6 with low enough temps. I will experiment a bit more though and see how she does at 4.7 with HT off, then see if its worth it.

Yes Prime95,,, Its (in my opinion) a bit of overcooking, so I actially only run that for about 5 mins, just to see if it crashes etc. I have read alot about not a good idea to use it at all, except maybe if your doing alot of video editing etc (which I dont do).

I setup a profile for Lasso when I was testing with Ht off, at which time I dedicated core 3 for DCS. I figured it may still be worth while doing that, I think it did actually help alot with the stuttering.

CoreTemp, I do have it, I vaguely recall it may have been providing some inaccurate info? I dont remember?

"Overclocking the 980 gives direct proportional more fps = GPU locked." im not following you on this exactly, but I Do recall I overclocked this 980 while it was on my older PC, but I did not see any FPS increase, so I let it stay running at stock at the time.

........

 

"Overclocking the 980 gives direct proportional more fps = GPU locked." im not following you on this exactly, but I Do recall I overclocked this 980 while it was on my older PC, but I did not see any FPS increase, so I let it stay running at stock at the time.

That means, you were CPU bound. No more fps unless you pump-up the CPU as well.

It's either my way or your way, hardly any system is 100%-100% balanced with CPU-GPU.

One is always faster than needed for the other, sigh !

I hope somebody can help me understand all this correctly, because I'm not terribly experience with tuning and CPU architectures et all. So if any of my interpretations here are wrong in any way, please feel free to jump in and correct me.

Let's start with some basic terminology. Much of what is posted about in this thread is in reference to AMD Bulldozer architecture. In this architecture, an "Interger Core" is equivalent to a Logical Processor (LP) aka VCPU (Virtual CPU) in Intel i7 architecture. IE on an Intel i7 CPU, which implements HTT (HyperThreading Technology aka HT) with 4 physical cores (in AMD terminology a Module), the CPU has 8 LP's presented to the OS.

In Skatezilla's first post in this thread he states: "Main DCS Thread will get an Integer Core, X-Audio Thread will Get It's own integer core. They'll Share FPU Resources of the Module."

From Skatezilla's description of how he is assigning affinity, he is assigning DCS.exe to use two "Integer Core" on the same "Module" aka physical core; in his example, Integer Cores 6 & 7.

In his other examples for setting affinity for Intel CPU's with HT however he's talking about assigning affinity to what appears to me to be two different physical cores. IE even numbered cores (Primary Thread) as opposed to odd numbered cores (Secondary Thread). This confuses me. Why should it be advantageous on an AMD CPU to assign DCS.exe to a single Module (physical core), but on an Intel CPU, to assign affinity to two separate physical cores?

Also, from my understanding, with i7 HT there is no difference in thread processing between the two LP's on a single physical core. Is this correct? If it is correct, why assign affinity to two different physical cores on an Intel CPU, but on the same Module on an AMD CPU?

I have an i7-5820K, which has 6 physical cores, so 12 LP's. So if I want DCS.exe to run on the last two physical "primary" cores (even numbered, so 10 & 8 ) I would start with the binary value "010100000000", which is hexadecimal 0x500, which is decimal 1280. So my entry in autoexec.cfg would be "affinity_mask = 1280". Correct?

If I assign affinity as above, won't I run the risk of other threads landing on these two physical cores, and possibly hindering processing of DCS's main and x-audio threads?

Any help here will be greatly appreciated.

Install ProcessLasso and run DCS on 0,2,4,6,8,10, make your life easy.

PL wont make you gain any fps but iron out CPU hick-hack and mish-mash if you run many other apps aside DCS, like streaming, VR, TS, SRS etc...

I have attached a photo showing PL forcing my miner to use said cores. That may help to work it out for you.

AMD's CCX's make some trouble if you send work across the CCX all the time, its therefor better to keep the work on one CCX. That is I think what Skate means. I dont run DCS on my Ryzen thus I dont know. I do know that 6 cores sleep and 6 cores work quite hard on my 6-core when I play DCS. DCS all-in-all uses at least 6 real cores, forget the old world.

I have poszted pics that show my rig pushing 6 cores to considerable workloads, no VR, no apps aside, clean DCS-NTTR WQHD. SIX cores at work !

I think the pics are in the Ryzen-2 hype thread, have a look to see what DCS can do with your 6-core. It will melt it down

Hi BitMaster, thanks for your answer, but Aristotle sends his greetings--"the more you know, the more you know you don't know".

I really want to understand how DCS is using my CPU, and what I can expect from the steps I take in tuning, and why.

Skatezilla's CPU is an FX-8350. The AMD Zen architecture uses CCX.

One of the questions I have, beyond what I've already asked above, is whether DCS.exe is using two real "threads". In other words, could both the DCS main thread and the x-audio thread both run on the same physical CPU at the same time using HTT? I'm not suggesting to do this. I simply want to know if that is the case, because it has implications.

Hi BitMaster, thanks for your answer, but Aristotle sends his greetings--"the more you know, the more you know you don't know".

 

I really want to understand how DCS is using my CPU, and what I can expect from the steps I take in tuning, and why.

 

Skatezilla's CPU is an FX-8350. The AMD Zen architecture uses CCX.

 

One of the questions I have, beyond what I've already asked above, is whether DCS.exe is using two real "threads". In other words, could both the DCS main thread and the x-audio thread both run on the same physical CPU at the same time using HTT? I'm not suggesting to do this. I simply want to know if that is the case, because it has implications.

I am no coder, but this is how I understand it:

No, they cannot. DCS would need to thread the process into 2 threads, 1 being the game and the other being the sound, to remain on 1 core with 2 logical CPUs. This works with processes that are alike, like apache can fork into dozens of threads, firefox, chrome, etc.. those are examples of good use of HT/SMT. DCS uses at least 6 cores total up to a certain amount, tho 4 of those are not actually DCS but the underlying architecture that needs to spool up to deliver.

DX11 uses more than 1 core, usually 2, so you have 4 already. The other 2 cores, I have no idea, but they are heavily used too, while all cores are physical, ProcessLasso wise.

Add VR, streaming, TS, SRS, etc etc. and you see why multi-core CPU's are on the rise.

Actually none of the cores was totally tilt, one was the "heaviest" but iut still had headroom most of the time, the gPU was tilt K.O. with 99% usage, GPU bound.