Sorry Eric but AM=0 is a nonsense.

From reading what SteveW has posted, Affinitymask=0 is the same as Affinitymask=15. It's 1,1,1,1 so you are using all four physical cores 0,1,2,3 when HT is not enabled. That's what I'm trying to do, use all 4 physical cores as I don't want HT enabled. I put all the other ancillary programs on cores 1,2,3.

Interesting, I have an i7-3770k OC to 4.5GHz, with HT enabled. I tried to apply an affinity mask of 84 and textures became a blurry mess with that setting, as soon as I took that setting out it was back to being sharp again. Looks like that setting is a no go for me.

 

I only have one machine for simming and I use 2 displays. 1 for my sim and the other for my flight apps running alongside it. Not sure how much it will affect my overall performance in P3D while flying when I have google chrome with about 15 tabs open, Active Sky Next, PFPX, Aivlasoft EFB, Pro-ATC, Remote CDU, FSwidgets network monitor, EZDOK, FSUIPC, and various other apps running simultaneously while flying.

 

Do you think this would cause a big hit in performance in P3D? I'm not sure how much CPU all those apps would consume while running alongside P3D.

no AM=11,11,11,11 = 8 LPs enabled, 84=01,01,01,00 = 3 LPs enabled

Steve so what would you recommend for me? Should I have an AM setting in my P3D.cfg or no? I tried 84 and it was blurries everywhere.

where'd you get 84 from?

I used the calculator here:

 

http://www.gatwick-fsg.org.uk/affinitymask.aspx?SubMenuItem=hardware

 

and it gave me a value of 84... 

if you want to optimise that 4 core with HT you need to make sure your addon exes avoid core zero, and assign an AM=253 to P3D using 7 LPs one LP of core zero to the sim main job, then 6 LPs from the three remaining cores do the background work. You can improve rendering performance a tiny bit with 85 but will reduce scenery throughput. With 85 the sim works the same as with HT disabled. It's where addon exe's go you need to deal with otherwise your AM testing will fail to give results.

 

No AM will result in 8LPs allocated and be very close to 253 in performance, although with 8 some throughput of core zero is handed to the second sim job and so takes throughput from the renderer.

 

255=0=11,11,11,11 = 8 LPs

253=11,11,11,01 = 7 LPs = core zero on the right with one LP allocated to the sim

85=01,01,01,01 = 4 LPs

84=01,01,01,00 = 3 LPs

 

If you want to free up a core, core zero in the case of 84, then it's still better to assign 4 LPs with 116.

 

116=01,11,01,00 third core from right (core 2) has two LPs allocated, jobs 2 and 3 of the four main sim jobs. Jobs 2 and 3 don't require maximum throughput at the same time, that's why 116 is good for that purpose.

 

To measure changes in underlying performance turn off all AA and vsync functions because these hold up the renderer. A common mistake is to assume a big change in performance when in reality a tiny change for better or worse can improve the relationship with the monitor timing and can appear like better, or worse, performance. Set up AA and vsync options after measuring CPU throughput.

 

If one looks more closely at 85 vs 253

 

01,01,01,01=85

11,11,11,01=253

 

With 253 the second third and fourth cores have jobs on both LPs. When these jobs require maximum throughput during scenery loading, they only get 50% each of the core. With 85, there's one job per core they get 100% of the core. So for scenery loading there's a relationship between allocating single LPs to ganging up LPs to share cores. Different sceneries account for subtle changes in performance and some may benefit from one technique or the other.

These settings with HT work well for blurries.

 

cheers

jja

no AM=11,11,11,11 = 8 LPs enabled, 84=01,01,01,00 = 3 LPs enabled

Steve, can you be so kind and look at my post on page 30, #447 and on this page, #451 and give me your thoughts? The chap I quote in #451 says AM=0 does nothing? Am I doing it right for what I want in post #447?

Eric, your post #451 makes perfect sense. The way an AM=0 is interpreted is usually to say all ones. So 15=0=1111 on a four core. The way you are setting up makes perfect sense as well. You allow the sim to break into four parts, one per core and moved addons above core zero.

Eric, your post #451 makes perfect sense. The way an AM=0 is interpreted is usually to say all ones. So 15=0=1111 on a four core. The way you are setting up makes perfect sense as well. You allow the sim to break into four parts, one per core and moved addons above core zero.

Thank you Steve. I'll keep Affinitymask=0 then.

Steve thanks for the detailed explanation, although I don't understand the technicalities behind this. So without any Affinity Mask settings in your cfg file, what is the default value? 0?

 

And do you recommend an affinitymask setting of 253 for me? Since I run a lot of apps alongside my P3D?

253 vs 255 or 0 or no AM? First check you see the logic behind what Eric is doing, he's making sure the core with the sim main activity, the renderer, is unencumbered by addon exes threads. That alone would de-stress the activity on core zero for your no AM situation. Use 253 to ensure the second sim job appears on the second core, so it avoids the second LP of core zero. This is an improvement over 255 so long as you take care that addons don't get assigned to that freed up LP on core zero.

 

These settings with HT work well for blurries.

Exactly. 254 makes only one LP of core zero available, the twin sister of 253, either one will do as they work exactly the same.

 

 

The recent dozen or so posts serve well to show why so much confusion reigns over the HT/AM situation.

 

 

Take a stopwatch and time how long a scenario takes to load, first with four LPs AM=85 (or 170, 102 etc.), then try with 7 LPs AM=253 (or 254). See how the scenario takes less time to load with 7 LPs? If that's not very much you may find 85 is better for rendering. If it's say 10 seconds or more you may find 253 is better overall.

Steve...I'm wanting try your recommended AM of 253, but have a question on allocating AS2016 (or 'addon exes' in general if you like).  I noticed improvement in AS2016 when going from having it using 8 LPs to only LP 0, 1 and 3, and allocating P3D to 2 and 4-7.  But this is giving me blurries now. Based on above don't think this is correct as P3D needs the 1st core (and only one LP of it) to run best?  So given that the AS2016 documentation says don't spread it out over too many cores due to latency concerns, how many and which LP cores would you recommend to run so it operates best with P3Ds 253?  Thanks for your great explanations of this complex topic...hopefully I am understanding it better now. 

You're welcome!

 

For addon exe apps, with the sim starting on core zero rules are; give addons two LPs and avoid core zero (LPs 0 & 1).