P3D V5.4 is available

[lownslo 594]

After several hours of testing a few scenarios I'm not seeing much of a difference from 5.3HF2 and 5.4.  VRAM usage is almost precisely the same, and performance is as well.  The JobScheduler section of the P3D cfg in 5.4 certainly works differently from 5.3HF2... clearly not as good as the former version.  Took me a while to find something that works for me, but 5.4 is running fine here now.  Hoping TOGA gets on board soon.

[Ray Proudfoot 9,217]

3 hours ago, lownslo said:

The JobScheduler section of the P3D cfg in 5.4 certainly works differently from 5.3HF2... clearly not as good as the former version. 

Are there different entries? If so, is there anything in a readme to explain how best to use them for your CPU and cores in use.

It’s difficult to accept LM would make something worse after such a long time refining the process.

[VHOJT 202]

Updated - have noticed little/no difference in performance - but haven't extensively tested.  Flew FSL A320 with EZCA3 and ActiveSky.  All worked well.  I only updated the client.

[IanHarrison 535]

9 hours ago, Patco Lch said:

Same here.

Must be. I installed it without defender running and Malwarebytes didn't complain.

Had it going for 2 days without problems.

[ark 1,306]

22 hours ago, ark said:

WIn10 Defender finds 2 files with Trojan:Script/Wacatac.H!ml (sever) when I download and try to install the new FFTF.

False positives?

Al

I had a small update to WIn10 today, so after that I tried redownloading and installing the new FFTF and now Win10 Defender did not find any problems.

Al

[lownslo 594]

15 hours ago, Ray Proudfoot said:

Are there different entries? If so, is there anything in a readme to explain how best to use them for your CPU and cores in use.

It’s difficult to accept LM would make something worse after such a long time refining the process.

The entries are the same as 5.3HF2.  The problem is that the 3 Scheduler entries are not flexible like 5.3HF2.  With that JobScheduler you could assign any of the 3 Schedulers to any core you wished.  This allowed the end user to assign the MainThreadScheduler to the strongest core (different for every CPU), and the other 2 Schedulers could be assigned in any order around the Main.  With the 5.4 update the MainThread Scheduler must be assigned to the first core, with the other 2 Schedulers following.  For my system, that means my temps are higher because my first core is not the best performing.  If I don't assign the MainThreadScheduler to the first core my flight loading times run from just under 2 minutes (with the first core running the Main) to 1 Hour and 50 minutes (with the Main assigned to my strongest core)!!  And the sim runs like mud when the Main is assigned to my strongest core!

If there's a readme I surely would like to have it.  But the method I used to assign affinity in 5.3HF2 worked perfectly well here (great performance and low temps), 5.4 not quite so well (maybe comparable performance but more heat).

[Ray Proudfoot 9,217]

8 hours ago, lownslo said:

With the 5.4 update the MainThread Scheduler must be assigned to the first core, with the other 2 Schedulers following.

I have the first core assigned as that. I suggest you post on the LM forum and whilst you probably won’t get a reply from them others could join in with their thoughts. That might then elicit a response from LM.

Bear in mind this is likely the last update to v5 with v6 probable before the end of the year.

[G-RFRY 1,765]

There must be a main core Asobo explained this when asked about multi-threading, A core must in control before handing of work to other cores, That`s why core O is the main core CPU`s are made with the main core to achieve the highest clock. Disable core O and another core becomes the main core. Even the latest multithreaded games have a main core that dishes out the workload.

This is why testers found the latest CPUs disabling the E cores and just having the P cores will not drop performance in games, the E cores are not needed and just increase the heat buildup and can achieve better clocks with just the P cores.

Edited June 17, 2023 by G-RFRY

[Bob Scott 7,848]

On 6/17/2023 at 1:49 AM, G-RFRY said:

There must be a main core Asobo explained this when asked about multi-threading, A core must in control before handing of work to other cores, That`s why core O is the main core CPU`s are made with the main core to achieve the highest clock. Disable core O and another core becomes the main core. Even the latest multithreaded games have a main core that dishes out the workload.

This is why testers found the latest CPUs disabling the E cores and just having the P cores will not drop performance in games, the E cores are not needed and just increase the heat buildup and can achieve better clocks with just the P cores.

but...the program's main thread is *not* constrained by the OS to run on physical core 0 (or logical processor 0)--it just happens to be the default (initial) assignment.  P3D v5.3 is an exceptional case in point--the program is written to allow locked assignment of the main thread to a user-specified processor, and that will then, in turn, assign the other two scheduling threads to the processors of choice enumerated in the config.  There are most certainly programmatic means of allowing a program to specify its own processor affinity...IOW, I can write a program that will run on whichever logical processor I choose.  Most software does not use this capability in order to maintain universal compatibility across a wide variety of CPU configurations, but in the case of P3D, allowing configurable (via the config file) installation-specific affinity makes sense when you are trying to optimize performance so that the main thread runs on the strongest processor. 

When you couple this with the capabilities built into Intel's 13th+ generation CPUs, where specific individual CPU cores can be configured separately from their peers (i.e. clock multiplier, hyperthreading), it allows the all-important main thread to be constrained to run on a core purposefully configured in hardware to run faster than its peers without having to clock the entire CPU up, which keeps heat dissipation to manageable levels.

[G-RFRY 1,765]

I`m only going by the best testers finding example the latest 13000K the first three cores can achieve the highest clocks. and when OC`ing on all cores you run the risk of exciding the clock and a BSOD lower than the first three cores can achieve.

[Luke 789]

20 hours ago, G-RFRY said:

I`m only going by the best testers finding example the latest 13000K the first three cores can achieve the highest clocks.

How many CPUs did they test? Don't make assumptions based on extremely small sample sizes. Silicon quality between cores should be essentially random.

Cheers

[klimpen 45]

Which affinity mask do you think is best 

for i5 13600k then?

[shermank 175]

Why not let P3D cfg do the calculations and setting. That’s how I run my sim and it is perfect, no Burris, no stutters only smooth as butter performance.

 

To each his own but as I have read and followed this thread I am seeing what IMHO is a lot of wasted worrying and time.

sherm

Edited June 19, 2023 by shermank

[G-RFRY 1,765]

14 hours ago, Luke said:

How many CPUs did they test? Don't make assumptions based on extremely small sample sizes. Silicon quality between cores should be essentially random.

Cheers

This is why Derbauer buys retail samples as well.

[Ray Proudfoot 9,217]

I've just installed v5.4. Previously I was running 5.3HF2 and stayed with that version on my new computer whilst I installed all 150+ airports and several aircraft.

I then enabled EA because this graphics card could cope with it. Whilst it took some time to adjust to the new perspective I noticed the clouds were flickering in the distance. But only when they were above me. Once above them they were fine.

I posted on Hi-Fi about it and was advised to update to 5.4. Doing that fixed the problem but I also wondered about any fps improvement.

I setup a test positioned above T5 at EGLL pointing east towards the airport and London  beyond. VSync was off, no weather and Ai was set to zero.

With v5.3HF2 the fps varied between 69-72.

With just 5.4 Client installed it was 67-73

With Content it was 67-74.

Finally with Scenery it was 68-75.

A small improvement but the annoying flickering cloud issue has been resolved.