CPU/Affinity question

Hello everyone,

I have this intel cpu which has 4 cores/8 threads at max. I remember coming across the statement that DCS uses only 2 cores: 1 for main and one for sound.

Now, I can run my CPU at:

1 core@3.8 GHz

2 cores@3.6 GHz

3 cores@3.5 GHz

4 cores@ 3.4 GHz

I can run all this setups with or without HT.

What would be the best choice for me? Which combo would give best results in DCS? I understand that in general 4 cores HT on is best system wise but if DCS has say only 3 threads wouldn’t it be wiser to keep it at 3 cores config?

Thanks all for your input and clarification!

I'd like to know this too. On last patch, the patch note said something about CPU affinity being set to 0 is fixed.

And I noticed the DCS process using all 4 cores pretty evenly. And now I'm wondering about HT. Is it better to have HT on or off?

I tried this before on my 9600K...

The clock speeds were higher, but the gains with less cores close to yours - it showed no real advantage with this little of improvements...

"Results":

1-3 cores showed - although higher clocks - LESS performance - through loading very noticeable and having firefox, chrome or windows ( ;) ) running at the same time gave noticeable dips in fps...1 core no chance, 2 cores were barely playable...

Comparing 4 to 6 cores I only loose 100MHz (4.7 vs. 4.6 GHz) show no noticeable difference in fps with slightly longer loading times...

Not that detailed but it shows where we are going...

That said, bumping up RAM from 2666MHz to 3400 MHz gave a massive boost in avg and low fps!

Also clocking up from 4.1 to 4.6 another but not as much as the RAM...

I tried this before on my 9600K...

The clock speeds were higher, but the gains with less cores close to yours - it showed no real advantage with this little of improvements...

 

"Results":

1-3 cores showed - although higher clocks - LESS performance - through loading very noticeable and having firefox, chrome or windows ( ;) ) running at the same time gave noticeable dips in fps...1 core no chance, 2 cores were barely playable...

Comparing 4 to 6 cores I only loose 100MHz (4.7 vs. 4.6 GHz) show no noticeable difference in fps with slightly longer loading times...

 

Not that detailed but it shows where we are going...

That said, bumping up RAM from 2666MHz to 3400 MHz gave a massive boost in avg and low fps!

Also clocking up from 4.1 to 4.6 another but not as much as the RAM...

Hey, nice detail! Could you please share your numbers (i.e. how much increase you got from the RAM upgrade?)

I personally went from 4.2 stock speed of my i7-6700K to 4.6Ghz across all 4 cores.

I am using this with GTX 1080Ti and I gained 10FPS average in The Channel compared to the same mission running at stock CPU speed. Testing was done with previous beta version.

I tried this before on my 9600K...

The clock speeds were higher, but the gains with less cores close to yours - it showed no real advantage with this little of improvements...

 

"Results":

1-3 cores showed - although higher clocks - LESS performance - through loading very noticeable and having firefox, chrome or windows ( ;) ) running at the same time gave noticeable dips in fps...1 core no chance, 2 cores were barely playable...

Comparing 4 to 6 cores I only loose 100MHz (4.7 vs. 4.6 GHz) show no noticeable difference in fps with slightly longer loading times...

 

Not that detailed but it shows where we are going...

That said, bumping up RAM from 2666MHz to 3400 MHz gave a massive boost in avg and low fps!

Also clocking up from 4.1 to 4.6 another but not as much as the RAM...

Hey, thanks for sharing the experience. I guess I'll stick with 4 cores. Now got to figure out HT thing.

What CPU is this?

I found that disabling HT on my CPU (i7-6700K) has no impact on DCS at all.

I keep it on as it helps me with other tasks I use my PC for, but I have also seen situations where HT off would give better FPS.

It also depends on your GPU and the software itself - in this case DCS.

What CPU is this?

I found that disabling HT on my CPU (i7-6700K) has no impact on DCS at all.

I keep it on as it helps me with other tasks I use my PC for, but I have also seen situations where HT off would give better FPS.

 

It also depends on your GPU and the software itself - in this case DCS.

<rant>

There is massive religious nonsense at this point, as people apply optimization techniques that applied to old CPUs to new ones that work totally differently. For example, locking threads to CPU cores made sense when switching cores would lose the cache. But if you have a CPU that doesn't has a shared cache, then you're actually better off letting your 100% thread hop from core to core, so the cores have less busy cycle and therefore can stay at max clock without getting thermally throttled. I find it best to stop redneck engineering and just let Intel do the work :)

But then I am using an i9 that can't even overclock (9900 no K) and it doesn't need to because it just reaches max clock anyway (as DCS hops from core to core.)

</rant>

I got a mobile intel i7 7700HQ which is not the best for sims but i have to get along with it. My subjective feeling is that DCS has much less stutters HT off compared to HT on. The peak FPS is better with HT on on the other hand.

I personally went from 4.2 stock speed of my i7-6700K to 4.6Ghz across all 4 cores.

I am using this with GTX 1080Ti and I gained 10FPS average in The Channel compared to the same mission running at stock CPU speed. Testing was done with previous beta version.

When quoting the performance increase, it's better if you say "from 80 to 90fps" or whatever it was, instead of +10fps. Because the gain is gonna be completely different based on where you were before (graphics settings, GPU overclock, etc.)

That said, I am glad to hear you got a nice increase. I would question the world's consistency otherwise. I know we are CPU bound in the higher end graphics cards with DCS, but I am also very curious to hear the numbers about the RAM clock increase alone. I have no way to test that myself at this point, as both my machines have the same RAM.

It does make sense though that it would matter for DCS! Are we going to see Ryzen's getting competitive again with 5GHz RAM? No idea but exciting.

People who said they tested this should state when they did this test. There was change to affinity on last beta.

Well, I had a few goes with different setups and 4 cores HT off gives the best result. HT puts terrible stutters in DCS for me.

Back in the old days of early multicore CPUs it was indeed recommended to use something like CPU Lasso to do this. However, in recent years you're MUCH better off letting Windows 10 handle it. I wouldn't suggest touching it at all.

Not all programs play nice together. Games based on the Unity engine are very core and hyper-threading sensitive. For example, if the Steam VR service happens to run on the same threads as the unity VR game, that it launched, the performance of the VR game will suck big time. Windows does not know which programs belong to which, but if a program launches another program, windows will run that program in the same thread. I use CPU Lasso to assign the steam programs one set of cores and the games that it launches to another set of cores and the gains in performance are dramatic!

Many games do not work well with hyper-threading enabled, so turning this off in the bios can increase gaming performance as this allows turbo boost to run the non-threaded cores consistently at high turbo speeds. The less cores and threads that are running the less heat is generated on the CPU. I can maintain a 4.3 GHz across 10 cores without hyper-threading, but only 4 GHz when hyper-threading is enabled which gives me 20 cores.

Another benefit of CPU Lasso is that you not only assign which process to specific cores, but you as assign the priority of those processes, but NEVER give a process “Real-time (Highest) priority as this is reserved for the Windows software and hardware systems.

With everything that I just said, unless you know what your doing, it is best to just let Windows manage the processes.