A Frame-Time Analysis of P3D v3 -- Effects of CPU Affinity, Frame Lock, and HT

[DylanM 1,069]

Interesting stuff Steve. Would you happen to have a test with no AM, and another with all cores enabled (with HT) using 4095 for comparison?

[SAAB340 28]

AM=12 is interesting how fps performance holds even though it hosts the first two sim threads. In the cases above the sim is running on a turbo enabled Intel CPU, it's not an overclocked PC. This CPU performance rises to meet demand. The graphs in Task Manager can show 100% but the core may only be at 80-90%.

Interesting that AM=12 didnt kill performance. When using AM=12, what load are you getting on core#2 and core#3 (the cores assigned)?

[SteveW 1,792]

SAAB340, on 11 Nov 2015 - 11:54 AM, said:

Interesting that AM=12 didnt kill performance. When using AM=12, what load are you getting on core#2 and core#3 (the cores assigned)?

LP2 is 100%, LP3 is up and down like a yoyo as the main thread stops and starts. The performance on screen is jittery, so AM=12 is not recommended. The turbo mode was at near full on, and little else happening across the CPU. The fps is not too badly affected, but autogen falls back and general smoothness suffers.

 

DylanM, on 11 Nov 2015 - 01:48 AM, said:

Interesting stuff Steve. Would you happen to have a test with no AM, and another with all cores enabled (with HT) using 4095 for comparison?

I'll try it out later Dylan.

[SteveW 1,792]

So remember testing a 6 core HT enabled 12LP turbo mode (not OC'd). No [JOBSCHEDULER] section, AM=0, and AM=4095 all equate to (11,11,11,11,11,11), all 12 LPs are active for P3D and all gave an apparently high fps by a few, but showed the worst results in terms of smoothness. As the fist core is more heavily loaded it produces a sawtooth with the turbo mode going up a couple of notches and back.

 

The AMs that masked an LP of each core worked best, especially the first encountered core must have just one LP active or the sawtooth appears to show a high fps but very poor stability. No more than four cores were needed or smoothness suffered. AM=340=(00,01,01,01,01,00) consistently outperformed other masks.

 

The apparent increase in fps with both first core LPs utilised by the primary sim threads together is at particular cost to smoothness, and could be seen with jitter on screen.

 

With HT disabled made no difference to the sim, except that the machine was noticeably sluggish, not just after booting but after using it for a bit. The results with no AM or AM=0 or AM=63=(111111) were similar to HT enabled and AM=340.

 

AM=30=(011110) gave the best result with HT=Off, and is equivalent to 340 with HT enabled. 340 and HT enabled was better overall.

 

HT re-enabled added back a bit of zest to overall performance of the PC. I would say to use HT enabled. If the Overclock is getting hot look at reducing the number of LPs utilised by the sim.

 

With HT enabled, mask one LP of the first encountered core at least. 4 cores was enough, 5 cores was too much. With four core CPUs, an AM masking the first core and one of the first LPs is recommended, e.g. AM=212=(11,01,01,00) if there are processes running on core 0 before the sim starts. Once the sim is running fully, the jobscheduler will usually keep other tasks off that core.

 

Look at the light blue line for real performance, "fps - Delta (Av)", how close it is to the red line, "Average fps".

 

 

[jrs182p 23]

Interesting stuff Steve. I'm interest in conduction these sort of tests on my i7 4790k. May I ask what your using to record this data and plot it to a graph? And also how you replicate a flight exactly to test these different masks?

[AviatorMan 253]

Interesting stuff Steve. I'm interest in conduction these sort of tests on my i7 4790k. May I ask what your using to record this data and plot it to a graph? And also how you replicate a flight exactly to test these different masks?

Looking forward to your results for the 4790k

[GSalden 4,655]

Steve, could you try 3392 too ?

Thanks,

 

Gerard

[BamaKevin 8]

So using an i5 4690k should I use an AFM?

[jrs182p 23]

So using an i5 4690k should I use an AFM?

None in my opinion. I5 4690k has no hyper threading I believe. So default should be fine. Four physical cores.

 

I currently run my i7 4790k with hyper threading off and no am as this seems to work best for me. But I would like to test HT on and off with and without AM's using Steves process and seeing the data.

[SteveW 1,792]

Steve, could you try 3392 too ?

Thanks,

 

Gerard

Hi Gerard, with 6 core HT=On, 3392=(11,01,01,00,00,00) results same as 212=(00,00,11,01,01,00). These work well almost as good as 340 and 980.

 

 

So using an i5 4690k should I use an AFM?

Does not need an AM, but with some addons you may find AM=14=(1110) smoother.

 

 

I currently run my i7 4790k with hyper threading off and no am as this seems to work best for me. But I would like to test HT on and off with and without AM's using Steves process and seeing the data.

Getting repeatable results on the PC is impossible. However we can get close by ensuring we start a flight from a saved file. We have to run it a few times before taking the trace run. There are various programs to record the frame rate etc. but I use the latest IF10 Pro version (not out yet, current version only does simple frame rate trace). I start the flight, let the flight off of pause at the same point after starting up (when the text message disappears). IF10 injects non-random weather and AI, and makes a .csv file with all the readings. I open the .csv file with Excel, select the columns and use the Insert Line Graph feature to produce the graphs. In turbo mode fps can't reliably be used to compare AMs, but we can compare the smoothness. Turbo mode affects smoothness with HT enabled unless we get a good AM. With the 4790 HT enabled try 212 and try 85. 212 works better with some exe addons, I believe some may recommend masking the first core if they run before the  sim. Note also I keep AA at minimum to avoid the GPU capping the frame rate.

 

 

 

[BamaKevin 8]

Steve thanks for that I have been testing 84 and 14. I may need to just lock my frames at 30 and see if that gets rid of the micro stutters.

[SteveW 1,792]

Hi Kevin. If it wont hold or exceed the monitor refresh rate at all times you see jittering. Instead if you  have a 60Hz monitor try locking at 20, 30, or 60. If you tried AM=14=(1110), try no AM at all.

[BamaKevin 8]

Will do Steve

[DylanM 1,069]

Wonderful stuff Steve (and the OP) and thanks much for the added scenarios...loving the empirical approach to this!

[jrs182p 23]

Thanks Steve. I tried 212 with HT on obviously. And it seems nice and smooth. I need to set up some proper tests with data to compare it to no am and HT off. Maybe I will get chance to do this over the weekend. I previously tried 84 and it causes me to get blurred ground textures. Have not tried 85 yet.