FSL Concorde - Core usage confusion

[lownslo 592]

2 hours ago, Ray Proudfoot said:

I’m sure I’ve read in these pages assigning all cores to P3D is the correct method. The individual main three threads can then be assigned to specific cores. LM recommend 2 and 4 are not changed but suggest the one assigned to MainThreadScheduler can be changed from 0 to 6. Seems sensible to keep the OS clear.

I wasn't aware LM recommended that "2 and 4" (do you mean the second and third of the 3 ThreadScheduler entries?) be unchanged.  I have been running them for months on my P3Dv5.3HF2 on 2 of the E Cores of my 13900KF.  Doing so allows me to run 2 more P Cores as terrain loaders.  And I use 4 of my E Cores in my P3D affinity mask (they too are terrain loaders).

2 hours ago, Ray Proudfoot said:

I turned off HT to keep temps lower. Would enabling HT with the extra 8 E cores improve performance given I limit fps to 30? Are those E cores improving performance? How much does it impact on increased temps?

For the past few years I've been using HT on my Intel CPU's because I found the feature offered me the best performance (mostly smoothness, not so much FPS) for FSX and P3D 64 bit.  But when I built my 13900KF system several months ago I flew for many hours a number of test flights I've used over the years, and the CPU performed that same for me with HT on or off.  Of course, the temp was lower with HT off.  So for me 24 cores works best.

As well, turning on HT will have no effect on your E Cores; only the P Cores have HT.

One of the tuning tasks I use for my flightsim box is to make sure all cores are unparked.  Core Parking is a power saving tech (a tiny, minuscule amount of power saving😖).  For me it isn't worth the latency it brings.  You can check your Core Parking configuration using a simple, standalone app titled Disable CPU Core Parking Utility by Coder Bag.  You can easily move the slider to the right to disable Core Parking.  Note that you have to make the setting for every Windows Power Plan you utilize.

The reason I bring up Core Parking is your E Cores not working for you.  Perhaps they are parked by your OS?  For us though, the E Cores can be useful; they can do work for us.

[Ray Proudfoot 8,979]

@lownslo, I followed the guidance here. Yes, second and third entries.

https://www.prepar3d.com/SDKv5/prepar3d/getting_help/advanced_configuration.html
 

The reason for my post was questioning the disparity between Task Manager histograms and numbers shown in RTSS. Bob has explained the reason and in future I’ll trust TM rather than RTSS.

My system is stable and performance is generally excellent. I’m not looking to make any further tweaks. I restrict fps to 30 as that prevents the CPU from running flat out and causing micro-stutters.

If you plan on buying Concorde I’d be interested in hearing how you get on with it. The most resource hungry aircraft I’ve ever bought.

[Bob Scott 7,721]

8 hours ago, Ray Proudfoot said:

Really? I thought LM started at zero. The default values for the three main threads were originally 0 2 and 4. When I was having lockups I took LM advice and switched 0 to 6. They suggest that in the SDK if you wanted to keep 0 for the OS.

No, sorry...I got that one wrong--they do both start at zero.  But that really has me puzzled now, as the core usage displayed by TM in your first screenshot shows that Windows is not applying P3D's affinity mask.

8 hours ago, Ray Proudfoot said:

Ah, that’s interesting. The monitoring is setup in Afterburner with RTSS used as a display. I still have HWInfo. Can that display something similar to RTSS?

It doesn't do the onscreen display like RTSS, but it does display a couple of core loading metrics--core usage (the percent of time that the core is busy) and core utilization (same as task manager but unlike the TM metric, it isn't capped at 100%).  I run HWInfo linked via network to my auxiliary PC (the one that Radar Contact, ActiveSky, etc run on) and I can watch a wide variety of metrics on the second PC without interfering with the sim display.
 

8 hours ago, Ray Proudfoot said:

Bob, is enabling HT worth it in respect of increased performance or am I likely to hit temp problems?

I don't think it's likely to help much, if at all, with P3D on a 8+16 core 13900.  I keep it off so I can run my main thread at 6 GHz--with HT on, the temps spike high enough that I can't keep it stable.  HT will make your CPU run hotter--whether that presents an operational impact depends on your configuration, cooling system capacity etc.  It also requires some care with the affinity mask to prevent P3D from trying to run threads on a virtual processor paired with the main thread.  Of course with the 13900, you have to ability to do per-core hyperthreading, so you could turn HT off just on the core running the main thread and turn the others on and see.  On my 13900KS box, with the overclock dialed back and HT on, even with an appropriate AM or HT disabled on the main thread's core, I get microstutters at times.  With HT off I do not.

[Ray Proudfoot 8,979]

6 hours ago, Bob Scott said:

No, sorry...I got that one wrong--they do both start at zero.  But that really has me puzzled now, as the core usage displayed by TM in your first screenshot shows that Windows is not applying P3D's affinity mask.

Could you expand on that please Bob? The first row shows cores 0 to 5. The second 6 to 11. Only cores 4 and 5 are working flat out. Here's my JobScheduler ...

[JobScheduler]
AffinityMask=16777215
P3DCoreAffinityMask=16777215
MainThreadScheduler=6
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

Core 5 is the strange one. I've loaded the Carenado PC12 at the same EGLL gate for comparison. Cores 5 and 6 are the most active. Cores 8 onwards are doing little or nothing. Is the P3D engine pushing work onto core 5?

6 hours ago, Bob Scott said:

I don't think it's likely to help much, if at all,

I agree. Heat could become an issue and with Concorde being so demanding I'd rather not push things.

6 hours ago, Bob Scott said:

I run HWInfo linked via network to my auxiliary PC (the one that Radar Contact, ActiveSky, etc run on) and I can watch a wide variety of metrics on the second PC without interfering with the sim display.

I'll investigate that, thanks.

[Bob Scott 7,721]

4 hours ago, Ray Proudfoot said:

Could you expand on that please Bob? The first row shows cores 0 to 5. The second 6 to 11. Only cores 4 and 5 are working flat out. Here's my JobScheduler ...

[JobScheduler]
AffinityMask=16777215
P3DCoreAffinityMask=16777215
MainThreadScheduler=6
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

Core 5 is the strange one. I've loaded the Carenado PC12 at the same EGLL gate for comparison. Cores 5 and 6 are the most active. Cores 8 onwards are doing little or nothing. Is the P3D engine pushing work onto core 5?

It looks to me like your main thread is running on core 5 rather than core 6, and I have no idea why core 4 would be loaded up like it is--the Concorde dll, perhaps?  Is this behavior reproducible?  Is there anything else running in the background when you see this kind of CPU utilization?

The only time I've seen weirdness like this is if I disturb P3D's processor affinity with an external utility like the task manager or Process Lasso, at which point the OS then does what it wants with thread allocation without further regard to P3D's internal affinity settings.  I find myself wondering if the Concorde add-on is manipulating thread affinities, and in so doing is negating the affinities set by P3D.

[Ray Proudfoot 8,979]

@Bob Scott, I didn't get a chance for a Concorde / default aircraft comparison today with nothing else running. I'll do it tomorrow.

[jcomm 5,939]

This may look silly, and uninformed... but I risk writing it here 🙂

Try using NO Affinity settings, and set HT Off and let P3D and FSLabs do their job... I'm curious to find out the possible outcome...

[Ray Proudfoot 8,979]

@jcomm, I did when I switched HT off. It came up with this...

[JobScheduler]
AffinityMask=16777215
P3DCoreAffinityMask=16777215
MainThreadScheduler=0
RenderThreadScheduler=2
FrameWorkerThreadScheduler=4

I changed the MainThread to 6 which is in line with LM recommendations for JS settings. It's a logical choice to keep the OS core as free as possible.

If I was to set it back to 0 I don't expect any difference at all. Concorde is extremely demanding of resources. As to whether FSLs do anything behind the scenes with settings I wouldn't know how to check and certainly wouldn't expect them to answer if the question was asked over there.

[Ray Proudfoot 8,979]

Here's two identical setups - one for the PMDG737, the other for Concorde.

In both, weather is clear skies with no ASP3D running, Traffic at zero, engines running ready for take-off. Airport is KILM, Rwy 24.

Here's the PMDG737...

And here's Concorde...

Thoughts?

[jcomm 5,939]

28 minutes ago, Ray Proudfoot said:

Thoughts?

Hmmmm, the Concorde flies higher... That's proportional to the peaks in the cores 4 and 5 🤓

OTOH, the PMDG strives to do it's VNAV / LNAV calculations correctly, and sometimes fails miserably, whereas the FSLabs Concorde is a lot more effective and polished, showing on core 6 😜

Sorry... couldn't resist .... For me it looks like Modern Abstract Art ...

Edited February 9 by jcomm

[Ray Proudfoot 8,979]

@jcomm, no worries. I see nothing unusual in the use of cores. The PMDG uses 2,4,5 and 6. Concorde uses 4-7. Whether these change during the startup procedures I don’t know.

But once airborne the remaining cores are used to load scenery.

[kmax59 213]

Hi Ray,

I've also noticed a quite weird behavior of the CPU cores with Concorde.

I think FSL is dealing with the cores used by P3D when you're flying Concorde and it's overriding your CFG's AffinityMask and choosen cores. just like when you're using their Bus.
With the A32X series you can change that by using the UMpolicy file and force it to use the cores you want, it's been a great tool for me in the past versions of the bus that were much heavier than they are now.

In my perf topic with my detailed report on the FSL forums I've asked if it would be possible to add such a thing for Concorde so we could control it better as well, but I did not get an answer on that point, and that still leaves me with an unusable Concorde btw. I can't fly an approach or taking off at more than an average 15 fps, pretty hard for the eyes, even harder for the controls.

[Ray Proudfoot 8,979]

@kmax59, I’ve just read your post on performance over there. It must be bitterly disappointing not to be able to get decent performance from your system. I have never used AMD hardware so have no idea how it compares to Intel / Nvidia.

As you say, what’s the point of making such a complicated Aircraft with all its systems if it’s basically unflyable around major airports unless you spend thousands on a system.

Even with my really expensive system I have to lower Ai to get 30 at Heathrow. And scenery complexity is set to Dense, nothing higher.

They are so secretive about their processes I doubt they’ll make any meaningful changes.

I’m sure many of us would have been happy with a slightly modified version of the 32-bit we enjoyed for years. And with no option to take over control of many systems we’re reduced to being a spectator on our flights.

Their reply will be interesting.

[Ray Proudfoot 8,979]

@kmax59, I’ve just read through that topic again. How depressing. Two posts stood out…

Andrew Wilson on 25 August… “performance issues are our main priority at this time. I anticipate we'll be in a position to issue another update next week that should fix significant issues and resolve a number of performance related issues.”

And on 5 January… “.. our focus is primarily on MSFS at the moment.”

Over four months between the two statements.

[Ted Striker 338]

This is disappointing news. I've been looking forward to purchasing the new Concorde but I think I will just keep flying the P3Dv3 version. It is the only plane I fire up P3dv3 for these days. Yes I have the scenery level reduced but I am so busy just flying this beast that I don't have time to notice.

Ted