Good Framerates - major stutters / pauses

[SteveW 1,792]

See ShezA post. Here he uses the sister LPs for other exe apps which utilises the HT effect of finishing threads faster. For HT Off these apps still run on those cores, but without the benefit of HT. With HT we direct the apps onto those 'spare' cores:

19 hours ago, ShezA said:

Got it. So for me with temp issues the setup settles on

1. P3D on core 0-2,4,6,8,10
2. Other tasks (ASP3D, etal) on 3,5,7,9,11.

Best balance of temp and fluidity. Thanks for the help in further refining my setup. Cheers.

If you had read and understood this post above, that's explaining it:

20 hours ago, SteveW said:

The way to think about cores is that with HT disabled you have one logical processor per core, that's one virtual core per core. With HT enabled that is two LPs per core, two virtual cores. The HT core is hard to understand. Let's look at one core running two identical tasks. These tasks take a certain time to complete. These tasks are time sliced at a certain frequency so they appear to run at the same time. In order to affect a good result with  HT enabled, those two tasks must be directed one each per LP. When that happens the two tasks finish sooner than they do with HT disabled. Those two tasks are helped to finish sooner because the swapping of the tasks to appear simultaneous is improved. Two tasks can be stored without swapping over their setup and registers (context) each time slice. With the way the simulator works, we can't improve the performance of the main task as it is monolithic. We can improve the performance of the background tasks pulling in the data and constructing the scene by using some of them in pairs on the later cores. If we use those spare LPs for other tasks that too is an improvement overall to the system throughput. We need not increase heat with HT if the setup is organised properly.

The problems spoken of around the forums is because they didn't set up their system correctly. Poor performance and excess heat described by those HT Off No AM folks is due to improper setup.

[Micko 89]

Hi all,

I followed Steve's advice for AM and put P3D on 0, 2, 4, 6....,22. Very pleased with the result, sim runs much smoother in heavy Orbx Eu England like London / Manchester and best of all, temperature reduced by about 10%. Thank you Steve for all the advice given on that topic appreciate it.

Mick

[SteveW 1,792]

Thanks Mick. You can also try reducing the P3D affinity by one or two LPs - 22 and maybe 20, which might give a bit of breathing room to the rest of the system - but it's a tricky one to determine the outcome.

[duckbilled 310]

Yes, many thanks for the advice. I tried 21845 and 21844. Both were better than HT off and HT off was better than HT on with no AM. 21844 seemed to beat 21845, so I'll stick with that for the time being. 

 

 

[SteveW 1,792]

9 minutes ago, duckbilled said:

Yes, many thanks for the advice. I tried 21845 and 21844. Both were better than HT off and HT off was better than HT on with no AM. 21844 seemed to beat 21845, so I'll stick with that for the time being. 

 

 

Seems good.

01,01,01,01,01,01,01,00 = 21844 ignores first core zero

00,01,01,01,01,01,01,01 = 5461 ignores last core 7

You might want to try 5461 which is exactly the same but ignores the last core rather than the first - depends on system but can help with faster flying in some cases. Again, very hard to determine without the PC in front of me.

I have written extensive custom monitoring tools which have highlighted the way the performance goes. So the theory is sound.

Edited June 2, 2020 by SteveW

[Ianrivaldosmith 5,317]

10 minutes ago, SteveW said:

Seems good.

01,01,01,01,01,01,01,00 = 21844 ignores first core zero

00,01,01,01,01,01,01,01 = 5461 ignores last core 7

You might want to try 5461 which is exactly the same but ignores the last core rather than the first - depends on system but can help with faster flying in some cases. Again, very hard to determine without the PC in front of me.

I have written extensive custom monitoring tools which have highlighted the way the performance goes. So the theory is sound.

I was told that one couldn’t exclude core 0 and that P3D needed this core active? 

[GSalden 4,654]

12 minutes ago, SteveW said:

Seems good.

01,01,01,01,01,01,01,00 = 21844 ignores first core zero

00,01,01,01,01,01,01,01 = 5461 ignores last core 7

You might want to try 5461 which is exactly the same but ignores the last core rather than the first - depends on system but can help with faster flying in some cases. Again, very hard to determine without the PC in front of me.

I have written extensive custom monitoring tools which have highlighted the way the performance goes. So the theory is sound.

Steve, what do you use on your cpu. Didn’t you have a 16 core ?

Now with V5 I might take a look again at HT. Currently still using ITB where the main thread is put on core 8 and all other take on the 13 remaining cores.

All addon apps have been put on the 6 less strained cores.

[SteveW 1,792]

5 minutes ago, Ianrivaldosmith said:

I was told that one couldn’t exclude core 0 and that P3D needed this core active? 

In general yes, especially with fast flying. I usually advise to include LP0 as can be seen in several posts above.

[SteveW 1,792]

1 minute ago, GSalden said:

Steve, what do you use on your cpu. Didn’t you have a 16 core ?

Now with V5 I might take a look again at HT. Currently still using ITB where the main thread is put on core 8 and all other take on the 13 remaining cores.

All addon apps have been put on the 6 less strained cores.

Hi Gerard,

I've got an 18 (9980Xe) and a 10 (9900X). Problem for me is they never stay in tune because they are usually under test. I use HT enabled. When I first got them I found HT enabled was better (slightly) than the optimum HT disabled settings.

[SteveW 1,792]

We can corral the other exe apps away from the main cores. So those cores have threads to complete. With HT enabled we can put the other exe apps onto the sister LPs thus those threads complete faster:

 

[SteveW 1,792]

What P3D and FSX do in terms of threading, and how they do it, requires an understanding of Windows programming and threading at an expert level considered a commercial asset, so understandably most of what's read about Affinity Masks (AMs) and FSX/P3D threading are guesswork.

[fogboundturtle 2,708]

LM do pinned core 0 at 100%. I am not sure if they are just doing dummy load to make sure core 0 is always at the highest frequency. It does spread the rest of the load to other cores. I am getting a pretty decent performance on DX12 with my RTX2080ti. I was getting some 60fps smoothness on the Majestic Dash 8 last night. I am playing at 2560x1440p.  I haven't that in years. 

[Dirk98 354]

5 hours ago, duckbilled said:

Yes, many thanks for the advice. I tried 21845 and 21844. Both were better than HT off and HT off was better than HT on with no AM. 21844 seemed to beat 21845, so I'll stick with that for the time being. 

 

 

Did you try 1555 instead of 21844? Why not having P3D on LP0 instead of on LP15? I recall there was a discussion that P3D should be preferably on Core0 or LP0 (in case of HT).

Thanks.

[sho69607 29]

p3dv5 has a major problem with stutter with autogen trees. If you turn the tree autogen to off the stutters stop. Something is off if my FPS is 45-60 but I am getting stutters every two seconds. I have resorted to using v4 for the time being. Sure my FPS is much lower, but I have an otherwise smooth flight experience.

[sho69607 29]

I want to add that this may be related to ORBX. With my ORBX sceneries disabled the stutters are gone in V5.