Sign in to follow this  
Egbert Drenth

Affinity Mask - Erratic behaviour P3D

Recommended Posts

Hi,

 

Yesterday I set the affinity mask in P3D config to 85 (i7 6700K, HT enabled, Windows 10 X64)

Never done it before and probably will never do it again...

 

I fired up P3D and did a flight from EHAM to LSGG.

Already at EHAM I had a strange issue. For no reason at all, the spot view automatically opened each 2 minutes or so.

Also I had no trim up, only trim down (set through FSUIPC). This had worked for many years, all of a sudden it was broken.

Finally the AP of PMDG had erratic behaviour. When turning, it suddenly banked 45+ degrees. The same in turns whilst hand flying.

 

Bottom line:

After setting AF I experienced a lot of erratic behaviour in P3D and add-ons. After disabling AF in config, all was back to normal.

Has anyone else experienced this?

 

Share this post


Link to post
Help AVSIM continue to serve you!
Please donate today!

Definitely odd but I don't think it might have much to do with AM. Unless you use fsuipc to set AM then it might be a conflicting set of instructions that is causing this.

Share this post


Link to post

How long before AMs are blamed for world famine?  :lol:
 

Share this post


Link to post

How long before AMs are blamed for world famine?  :lol:

 

 

You tell me...

Fact is that enabling AF on my system causes eratic behaviour. The moment I put it off, everyrhing is normal again.

Most likely it is related to my system, nevertheless I wondered if others might have experienced the same, hence I posted a message here.

Thanks for your informative reply.

Definitely odd but I don't think it might have much to do with AM. Unless you use fsuipc to set AM then it might be a conflicting set of instructions that is causing this.

 

Thanks for the heads up, I'll check.

Share this post


Link to post

Calm down, can't you see the funny side, it's the work of the devil? :P

 

Why don't you show us what you put in your .cfg?


The Boom already mentioned to check your FSUIPC settings. I'm not sure what that does when it set's affinity on addons it starts, but you've not mentioned that you use any other affinity setting tools. I need more than what you say to go on. 

Share this post


Link to post

Calm down, can't you see the funny side, it's the work of the devil? :P

 

Why don't you show us what you put in your .cfg?

 

The Boom already mentioned to check your FSUIPC settings. I'm not sure what that does when it set's affinity on addons it starts, but you've not mentioned that you use any other affinity setting tools. I need more than what you say to go on.

Hi Steve, I use FSuipc4 to start and close ASN along with calling up an AM.

 

[Programs]

 

Run1=AM=3,CLOSE,C:\Program Files (x86)\HiFi\ASNext_P3D\ASNext.exe

 

bob

Share this post


Link to post

So you are setting LPs 0 and 1 for ASN. Do you also have a jobscheduler section in your cfg bob?

Share this post


Link to post

So you are setting LPs 0 and 1 for ASN. Do you also have a jobscheduler section in your cfg bob?

yes that's correct, in my cfg AM=116 (i7 4770 4core HT on)

 

bob

Share this post


Link to post

Seems like a good setup bob.

 

Egbert, why not try it out again and check in task manager what's been allocated to what before flying off. If you are trying AM=85 that would normally produce just as good results, so long as you put your addon on LPs 3 and 7. Otherwise bob's example usually produces good results, worth trying that out too. In the end, strange behaviour will be tracked down to some other thing, assuming appropriate AMs and affinity have been set up.

Share this post


Link to post

Odd idea: you CPU has an error on a single core. AM setting forces this core to be used, without AM, the core is not used (for whatever reason). Or it is related to an somehow instable overclocking which is only an issue when AM forces the use of specific cores...

Share this post


Link to post

Seems highly unlikely that an AM setting would have anything to do with the behavior you describe...

Share this post


Link to post

Odd idea: you CPU has an error on a single core. AM setting forces this core to be used, without AM, the core is not used (for whatever reason). Or it is related to an somehow instable overclocking which is only an issue when AM forces the use of specific cores...

That's quite plausible. He did not mention if he overclocked. It could be the case. I have a 6700k too and by default P3D chooses not to use core 1 on my system. If I force it do so with an AM I get stutters every second.

 

But with any AM that doesn't use the first core it's fine.

Share this post


Link to post

By the way, it's the jobscheduler that chooses where the sim places threads, not P3D. The Affinity Mask in the Jobscheduler section of the .cfg only enables and disables LPs showing the Jobscheduler what it can use for the sim.




I have a 6700k too and by default P3D chooses not to use core 1 on my system. If I force it do so with an AM I get stutters every second.

But with any AM that doesn't use the first core it's fine.


I think what you are saying is that when you allow the jobscheduler to utilise LPs 0 and 1, (core 0 = LPs 0 and 1), then you get less performance? In this case it would certainly be true since you are sharing the core the main sim thread is demanding full uninterrupted throughput from. That's why the popular method of disabling the first LP (setting the rightmost 1 to a zero) in the Mask works well.

Share this post


Link to post

 

Obviously many are interested in AMs because it's perhaps mysterious, but MS and Intel know what they are doing, the jobscheduler section in the cfg is a system setting not a tweak.

 

Let's look at popular AM=254=11,11,11,10. The rightmost bit is masked so the jobscheduler chooses LP 1 for the first thread of the sim. The second and third sim jobs share the next core (core 1 LPs 2 and 3). The fourth sim job appears on LP 4. Other sim jobs start on the remaining unmasked LPs. If we move that pattern left by two LPs we have a very similar situation encountered by utilising AM=116=01,11,01,00 as in bobs example. We don't unmask bit 7 with that, (core 3 LP 7) because it creates five sim jobs, we can have four or six, but five produces poor results. So in theory we could improve on 254 with the AM=238=11,10,11,10, we've got the first thread on a core of it's own, we have shared a core with sim jobs two and three, and sim job four gets a core to itself. Finally, we've opened up affinity for the sim by enabling bits 6 and 7, or rather LPs 6 and 7, these don't contribute to the sim performance but they do allow more choices for the jobscheduler to place sub-processes of the sim.

Share this post


Link to post

By the way, it's the jobscheduler that chooses where the sim places threads, not P3D. The Affinity Mask in the Jobscheduler section of the .cfg only enables and disables LPs showing the Jobscheduler what it can use for the sim.

I think what you are saying is that when you allow the jobscheduler to utilise LPs 0 and 1, (core 0 = LPs 0 and 1), then you get less performance? In this case it would certainly be true since you are sharing the core the main sim thread is demanding full uninterrupted throughput from. That's why the popular method of disabling the first LP (setting the rightmost 1 to a zero) in the Mask works well.

Yes that's what I meant. So it could be that by utilizing the AM of 85 the OP could have opened the sim up to conflicts from background processeson the first LP. Although it's unlikely that they might affect the sim to that degree as described.

Share this post


Link to post

You can run the sim with an AM=3 and it wouldn't do those things. Processes don't conflict via use of cores, they are protected by the way the CPU works.

Share this post


Link to post

Not really a conflict but an error on the first core as someone else described. Either that or conflicting AMs set outside of P3D. I really don't see how else it could be AM related.

 

To OP, can you reproduce this again with the same AM or different AMs?

Share this post


Link to post

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this