@SteveW, if I've assigned LPs 0,2,4,6,8 and 10 to P3D how can I assign separate ones to these other executables? There would be some sharing surely?

Currently they're all on 11. Not ideal but what alternative is there?

As you said, their demands are very small so they don't add much to the throughput of any core in any case, but running on only one LP each can cause waiting within their own multi-threading states. You are loading them on LP11 which is core 6 and that already has a task on LP10. The background task take seconds to complete so it's not a problem to load more work on them when you have several cores to do that work. I would suggest you distribute those exe's on any of LPs 5, 7, 9, and 11. We are making use of affinity to corral that throughput away from the main task because we want to give that maximum throughput, but for any other exe's running along with the sim, we also have to be sure those exe's have freedom to run without concerns.

2 minutes ago, SteveW said:

I would suggest you distribute those exe's on any of LPs 5, 7, 9, and 11.

Thanks, I'll do that via SimStarterNG. Otherwise it's a manual job each time they're launched.

Yes, you wouldn't do it manually because affinity needs to be done correctly by having those exe's only 'see' the required LPs available when they start so their threads can be properly distributed on those LPs.

@SteveW, I appreciate your posts today.  In my P3D forum conversation with Ray yesterday, I was having a hard time understanding why certain AffinityMask settings were being suggested over others in an attempt to mitigate the 100% usage of thread 0 issue.  Now I understand it is most optimal to reserve the even numbered logical processors for P3D and to try to divert add-ons to the odd numbered logical processors.

@Ray Proudfoot, looking at my testing and screen shots of charts, you were correct in the first place.  10101010101 (AffinityMask 1365) would be the best allocation for my six-core 12-threaded processor since all the work is being done by the even numbered logical processors.  I take it SimStarter NG can require third party software to run on the odd numbered logical processors.

Edited June 18, 20214 yr by TextRich

1 hour ago, TextRich said:

  I take it SimStarter NG can require third party software to run on the odd numbered logical processors.

Yes. In the Run section you would create an entry for P3D and then modify it to select the logical processors. You can opt for 0 2 4 6 8 10 or 1 3 5 7 9 11. Both achieve the same result but have a different AM number. Same goes for any other executables taking into account Steve’s advice posted earlier.

I’m sorry for curtailing our discussion on LM yesterday but I didn’t want the topic to deviate from the posted problem.

1 hour ago, SteveW said:

Yes, you wouldn't do it manually because affinity needs to be done correctly by having those exe's only 'see' the required LPs available when they start so their threads can be properly distributed on those LPs.

Yes, I agree whole heartedly! For this purpose, my preferred utility is Process Lasso Pro:

https://www.geekiest.net/post/Process-Lasso-Pro-Review

The digital purists amongst us often frown on this approach, but I have used it for years and, for the most part, have found it can be used effectively as is without any user interventions. However, as stated, I do exploit its automated affinity masking capabilities to allocate LPs for use with those 3rd party exe’s. The exception, of course, is the Prepar3D.exe which should have the ‘affinitymask=‘ entry under the user-added  [JOBSCHEDULER] section of Prepar3D.cfg.

Prior to using Process Lasso Pro, it was necessary to create individual bat files for each 3rd party exe (thanks Steve). This had the advantage of being able to select which exe you wanted to run with P3D at each run and know that its pre-defined affinity mask would ensure proper allocation of LPs. The same can be accomplished with Process Lasso running in the background and starting a 3rd party exe shortcut as and when required. I imagine SimStarter NG has similar capabilities in this respect.

Mike

 

 

Edited June 18, 20214 yr by Cruachan

Gents, whilst all this discussion about AM is helpful it’s deviating away from the topic title. I hold my hand up to being guilty too.

If any of you would like to conduct the Traffic BGL test I eagerly await your findings. Is 5.2 a worse performer than 5.1 for you?

36 minutes ago, Ray Proudfoot said:

Gents, whilst all this discussion about AM is helpful it’s deviating away from the topic title. I hold my hand up to being guilty too.

If any of you would like to conduct the Traffic BGL test I eagerly await your findings. Is 5.2 a worse performer than 5.1 for you?

I didn't do the same test as you outlined, it would have meant too much fiddling for the time I have available today, but I did do the same flight with the same settings with all AI turned off and then with AI turned on (BGL and AIGFP) and while I saw a small drop in FPS - no more than I normally see - I did not see any difference in CPU load patterns. 5.2 appears to be on a par with 5.1 in general performance terms, perhaps even slightly improved, running with the same settings that I ran in 5.1. I have yet to try EA on + volumetric clouds off. 

I did get the 'freeze on view change with HTML text on' problem, but just once and only on one PC. 

7 hours ago, SteveW said:

One of the main principles of HT is that it reduces the workload of thread-swapping by maintaining the context of those threads and reduces cycles required to load each context in the non-HT setting.

I think I would instead say that "One of the main principles of HT is that it can reduce the workload of thread-swapping..."

It all depends on how the code is written.  If the program is skillfully coded to minimize thread-locking and contention with HT in mind, HT can make a significant positive difference.  If not, it can actually negatively affect performance, especially where performance is defined not only in terms of average throughput, but also in terms of maintaining low variability in output timing (minimal or acceptable jitter).  There are other considerations, as well, such as the impact of sharing the low level memory cache between two LPs vs having the entire cache available to a high-priority thread running on that core.  So HT can reduce workload and/or improve performance, but its presence does not guarantee that.  Often it does not, due to code that does not optimize for (or even consider) parallelism in general and HT specifically.

Up to this point, my experience with use of HT in P3D on CPUs with 6+ physical cores (i7-8086K, i9-9900K, i9-10900K), has presented issues with smoothness when compared to running with HT off, even with carefully managed affinity masking.  My last two four-core CPUs (i7-7700K, i7-4790K) seemed to benefit from HT, though.  It seems that P3D works best in an environment with at least six LPs, and better still if those LPs are dedicated physical cores.

On 6/14/2021 at 8:39 AM, Ray Proudfoot said:

There's been no official response from LM about any change in CPU/GPU processing but given there were lots of complaints about excessive VRAM usage I suspect they've moved some processing from the GPU to the CPU hence why my performance seems worse with 5.2 than with 5.1. Look at my settings and ask yourself why does CPU1 stay so high with so little to do?

I was very hesitant updating, however once FSL updated their buses I decided to give it a go. Evshade and Envtex updated as well. 

With EA, CPU is getting more work but not at 100% as reported (10700K). But GPU (2080TI) get work a bit more as well, in my case I would say bit balanced. However with EA on. Since EA is NOT fixed or at least not to my liking, I am driving without and everything is as needs to be. CPU and GPU.

Overall looks much better or at least this is my imagination, LOL. What I noticed is that FPS is 10% down but I am testing now with 4096 and everything to the right just to bring to the MAX to see initial behavior. Had only once black screen and once half black screen (?) whatever that was but was able to save it with view change. Not sure if Content repair is needed, so if someone can advise I would appreciate.

It looks that all depends on individual System and way how is installed. This time I decided to go with full installation as advised by ORBX and no issues with add-ons or sceneries.

And at the end I still don't like this waffle EA and driving me nuts. 

My short report and current experience with 5.2.

Edited June 18, 20214 yr by cyyzrwy24

I have resolved the problem and here's how simple it was. There's one critical setting that has solved things. My thanks to @SteveW for pointing out the typo.

[JOBSCHEDULER]
AffinityMask=1365

I had entered this in P3d.cfg but crucially included a space in the section name. So the AM was ignored. So by default every logical processor was being used. This resulted in my 6-core, 12 logical processor CPU being swamped with demand from P3D on the critical core 0. Hence why even very low scenery settings were still causing cpu 0 to hit 100% and remain there for my test. That resulted in loss of fps and generally very poor performance.

It only recovered to under 100% when I removed all Traffic BGLs including the default stuff. But who wants to fly in a sterile environment?

Having had the typo pointed out to me I corrected it and fired up P3D again. What a transformation. With the AM setting my CPU is now using the following logical processors for P3D.

0,2,4,6,8,10.

The second logical processor on each core is not being used. This is especially important for core 0.

My aircraft is now located 300ft above SimWings EGLL and is slowly slewing through 360 degrees. CPU0 rarely hits 100% and on average is around 80-90%. There are 85 AIG Ai in the sim. I use FSUIPC to reduce Ai if 30fps cannot be achieved.

I'm well versed with updating P3D and feel the wipe and reinstall is only required if you have serious problems you can't resolve.

So there you have it. My BIOS has HyperThreading enabled but whether it has or not the importance of the AffinityMask setting cannot be overstated. It would be interesting to know if those of you with the same problem have that setting. If you don't then you need to calculate the AM value depending on how many cores your CPU has and whether HyperThreading is enabled or not in your BIOS.

Remember it's recommended Prepar3d.cfg is deleted and a new one allowed to be generated but crucially it won't have the AM setting in it. That has to be added manually.

Glad to hear you got it sorted to your satisfaction, @Ray Proudfoot.

For what it's worth, on my client PC (which runs just P3D with Wideview / Widetraffic, FSUIPC and GSX as add-ons, a copy of Active Sky, and a piece of software that I wrote which I know takes almost no CPU), I'm still seeing 100% Core 0 constantly with the other physical cores (HT is disabled) pretty much maxed now too, with AI on or AI off. But as @w6kd and @SteveW have pointed out, CPU load patterns are affected by so many factors that frankly it could just be down to my settings and my CPU itself (with fewer cores and less cache) vs my other PC, which doesn't see full load on all cores (but does on Core 0) despite having a lot more programs running alongside P3D. 

I used to use Process Lasso to set CPU affinities but since I re-built my host PC and didn't put Process Lasso back on there, I honestly haven't noticed any performance difference. In fact, 5.2 definitely is smoother and gives me higher FPS on the same settings as 5.1. Not *much* higher, but noticeable. The only issue I have now is a regular short stutter every 10 seconds or so on my client PC and I had that before 5.2, and I'm pretty sure it's down to some regular process that I've just not identified yet (no FSUIPC autosave on either PC, BTW). 

I'm on my way to EDDB in the sim now...

Edited June 19, 20214 yr by neilhewitt

I thought if you have Hyperthreading off in bios then you don’t need to create AffinityMask? 

@neilhewitt, you may be onto something there. I spoke to @Pete Dowsonand he also reported that on a test PC with HT disabled there were no performance issues with no AM setting.

This requires the expertise of @SteveW to clarify things. AM only needed for HT machines Steve?