14 minutes ago, Ray Proudfoot said:

Thanks Steve, I’ll try that tomorrow.

I typo'd in my post on the core numbers by starting from 1 and said cores 3 and 5, when I meant to say 2 and 4, starting from 0, so I edited the post, the LP numbers were correct.

By moving your other programs to LP5 and LP9 (from 9 and 11) should hand 10% to your terrain loading on core 5 by moving that 10% on LP11 to LP9 and the use of LP9 to LP5. That would equate to 3 1/3 % extra terrain loading for no cost.

I went back to maximum settings and LPs 2, 6, and 10 increased a lot.

For experimentation, I then I added three more cores, total of 9, and changed the AM to 87381=01,01,01,01,01,01,01,01,01. What happens is I get 3 more terrain loading Tasks on 12, 14,and 16, and the LPs 2, 6,and 10 continue as they were just as in Ray's screenshot.

However, I notice also that the main task on LP0 has increased utilisation to nearly 100%, which is to handle the extra messaging from the 3 added cores. If I add more cores I am in danger of saturating the system and causing problems on the main task.

So this confirms that with these settings and having too many cores I will lose performance.

8 minutes ago, SteveW said:

However, I notice also that the main task on LP0 has increased utilisation to nearly 100%, which is to handle the extra messaging from the 3 added cores. If I add more cores I am in danger of saturating the system and causing problems on the main task.

So this confirms that with these settings and having too many cores I will lose performance.

Steve, so how many cores do you allocate to Prepar3D.exe finally? Is it 8 or 6 or else? 

Thanks.

Edited June 23, 20214 yr by Dirk98

1 minute ago, Dirk98 said:

Steve, so how many cores do you allocate to Prepar3D.exe finally? Is it 8 or 6 or else? 

Thanks.

On this system, a 9980Xe + NVLink 2 x 2080Ti, I have settled on 8 cores and a limit of 28fps. I have maximum settings apart from traffic set to 50% and cars and boats set to 10%. Performance of P3D is very good.

It has been a time well spent doing some experimentation with Affinity Masks (AMs) in Prepar3D because even though we are now at version 5.2 the proper use of AMs has proved to be essential. With this system, whether in HT mode or not, the performance without the appropriate AM is appalling.

13 minutes ago, SteveW said:

So this confirms that with these settings and having too many cores I will lose performance.

Steve, with an AM setting is there an ideal number of cores? Could someone have too many as you suggest?

5 hours ago, craigeaglefire said:

there seems to be much discussion about Affinity Mask settings - for P3D 5.2 - when many are happy with 5.2 performance now without such AM tweaking,

only once was I able to get 100% core0 usage when Super Traffic Board was used - have used it subsequently with no problem…

Process Lasso is used for CPU management & works very well,

no need for any AM tweaking here.

You mentioned that you didn't need AM tweaking but you said nothing about your HT On or Off, and what your CPU was. Therefore your statement about what you used and what you didn't made no sense. 

Edited June 23, 20214 yr by Dirk98

1 minute ago, Ray Proudfoot said:

Steve, with an AM setting is there an ideal number of cores? Could someone have too many as you suggest?

Yes Ray, there are always an optimum number of cores for any Prepar3D system no matter what settings or added scenery or aircraft. It is a basic fact of life that there are always optimums.

11 minutes ago, Ray Proudfoot said:

Steve, with an AM setting is there an ideal number of cores? Could someone have too many as you suggest?

I'm not Steve but I can assure the more Cores on your CPU the better overall. Just cut off the excessive ones with AM. 8-10 cores are the sweet spot for Prepar3D. Another important parameter one should watch when choosing a CPU is a single-core performance. 

Edited June 23, 20214 yr by Dirk98

Just now, Dirk98 said:

I'm not Steve but I can assure the more Cores on your CPU the better overall. Just make sure to cut off the excessive ones with AM. 8-10 cores are the sweet spot for Prepar3D. Another important parameter one should watch is a single-core performance.

What you can do is add cores into the AM one by one and with a stopwatch record the time to load a scenario. What I find is that the time to load the scenario reduces at first by a lot, and eventually by a very small increment.

It is when the increment in the reduction in load time becomes very small that you have gone beyond the limit of the system to support all those cores running in parallel.

There is a finite throughput to any system and that will be divided down by adding cores. The result of too many cores brings the main task down, and it sees 100% earlier.

41 minutes ago, SteveW said:

What you can do is add cores into the AM one by one and with a stopwatch record the time to load a scenario. What I find is that the time to load the scenario reduces at first by a lot, and eventually by a very small increment.

If you load the same scenario a second time would caching come into play? Maybe a different scenario each time. But if one was located in London and another in Timbuktu would that be another factor to take into account?

Dirk suggests the optimal is 8-10 which suggests I made a very wise choice of CPU. 😁

14 hours ago, Ray Proudfoot said:

If you load the same scenario a second time would caching come into play? Maybe a different scenario each time. But if one was located in London and another in Timbuktu would that be another factor to take into account?

Dirk suggests the optimal is 8-10 which suggests I made a very wise choice of CPU. 😁

You have to load the chosen scenario a couple of times first and then stick to the same scenario and settings increasing the cores one by one by only adjusting the affinity mask. Other scenarios and different settings take different times.so would spoil the experiment which is to compare the time of the same circumstances with only changing core count.

16 hours ago, SteveW said:

I went back to maximum settings and LPs 2, 6, and 10 increased a lot.

For experimentation, I then I added three more cores, total of 9, and changed the AM to 87381=01,01,01,01,01,01,01,01,01. What happens is I get 3 more terrain loading Tasks on 12, 14,and 16, and the LPs 2, 6,and 10 continue as they were just as in Ray's screenshot.

However, I notice also that the main task on LP0 has increased utilisation to nearly 100%, which is to handle the extra messaging from the 3 added cores. If I add more cores I am in danger of saturating the system and causing problems on the main task.

So this confirms that with these settings and having too many cores I will lose performance.

Like my experiment, in your setup for 6 cores your Task Manager graphs show that your first core is running below 100% with your settings which limits your fps by vsync of your 30Hz monitor. If you then add a couple of cores (01's) with the AM, don't change anything else, load the same scenario and see if the first core becomes too heavily loaded. If it has reached 100% you would need to reduce settings or fps limit to get the core down from 100% because the system bandwidth had been reached, you can't get more throughput for free from nowhere. Remember that if you don't run the same test your experiment is spoiled. For example if your weather program has advanced in the day and weather changed, or your time has advanced and your traffic has changed, the experiment is spoiled. For experimentation, to make valid comparisons, I use a system that makes sure the scenario sets off with identical loads each time.

3 hours ago, SteveW said:

You have to load the chosen scenario a couple of times first and then stick to the same scenario and settings increasing the cores one by one

Yes, this is important: before doing any benchmarks launch and exit the same scenario in P3D a couple of times first.

19 hours ago, SteveW said:

Yes Ray, there are always an optimum number of cores for any Prepar3D system no matter what settings or added scenery or aircraft. It is a basic fact of life that there are always optimums.

Back in the good ‘old days:

Mike 😉

I've repeated a test loading the same scenario at KORD with lots of Ai. I stopped FSUIPC removing traffic to achieve 30fps. No weather was loaded.

With 6 LPs selected (AF1365) the load times were all within 1 second averaging 1m 10s

With an AM of 341 (one LP less) the load time was 1m 17s.

With an extra LP (0,2,4,6,10 plus 11) on top of 1365 the load time was 1m 08s.

Given I appear to have the ideal number of LPs I can't experiment much really. I will move other executables onto other LPs as suggested by Steve a couple of days ago. Other than that things are pretty much optimised.

Rather than adjust scenery settings to keep core 0 from peaking at 100% I use FSUIPC to reduce Ai to achieve 30fps. Seems to work well enough.

I'm about to fly EGBB-EBBR replicating the flight that started all this concern 11 days ago. I imagine the experience will be far better this time.

A change of settings, scenery type, location, altitude etc. can have a big effect on the outcome of the tests. What I suppose would be best is to choose a scenario that fits in with the flying you prefer. For example if you like to land at Heathrow mostly, then have a scenario of your aircraft coming into land there. Then use that in repeat tests to find your best settings. Generally when near the ground in heavy scenery is probably where you need the preferred setup for takeoff and landing. At altitude or over the sea is not going to be so hard on the system.

@SteveW,

Just landed at EBBR and although 30fps was maintained bar the approach into 01 I was curious why cpu0 stayed at 100% for the majority of the flight.

Turns out I'd forgotten to reset my FSUIPC Ai preferences so hundreds of Ai were at EGLL and nearby airports. That could also explain why fps dropped below 30 on approach to 01 EBBR as no Ai could be removed.

I don't fly into EGLL that much but have set my scenery settings for that airport as it's the hardest region I have with London so close.

The biggest impact after Scenery Complexity for me is the number of Ai. I'd rather see another 30 aircraft than a few more buildings.

A satisfactory outcome after nearly two weeks of work. 👍