Are you getting even usage across all Cores?

[Sethos 4,302]

2 minutes ago, TechguyMaxC said:

Who are you again, person that has 452 posts and has been a member here for all of a year?

Who are you, Steve's cheerleading squad?

[SteveW 1,792]

Guys, HT and P3D are not my idea.

Everything I say can be found in the docs from Intel, MS, and LM. I am trying to help get across what is a very tricky subject. No need for rudeness.

[Sethos 4,302]

Just now, SteveW said:

Guys, HT and P3D are not my idea.

Everything I say can be found in the docs from Intel, MS, and LM. I am trying to help get across what is a very tricky subject. No need for rudeness.

You're right, apologies. 

[SteveW 1,792]

16 minutes ago, Daedalus said:

Steve, I trust you. Would it be easy for you to give me your so far "best recipe" for AF setting for an i7 7700K, overclocked at 5Ghz? I'd be very pleased. I want to do more tests.

Depends how far up the heat curve you are. HT off can mean very high frequencies are attainable and the loss of HT is overcome. Remember that P3D (and FSX) work monolithically so are unaffected by HT on or off, however the benefit off HT On is from the operating system and use of resources like networking. On a relaxed overclock HT is beneficial, but apps like P3D need care. It's all in the manuals.

Remember that the sim sees the LPs made available through the mask:

HT Off AM=0=1111 will be less heat than HT On AM=85=01,01,01,01 at the same frequency. But at the same do-able frequency HT On will outperform Off especially when you have SimConnect clients and a busy system.

 

Win 10 Game Mode can perform better but remember you may need to set addon exe apps to Game mode as well, and so best to try it out.

 

[ComSimPilot 623]

Thank you very much Steve. So, if I can manage the same overclock with HT-on and an AF=85 would perform better. I'll certainly try your suggestion for AF=85. I have tried to turn off unnecessary services that eat performance and don't use non-simulator related programs when flying. Expect a browser sometimes.

Will also try your Game Mode suggestion. Thanks a million :)

[SteveW 1,792]

20 minutes ago, Sethos1988 said:

Oh look, another snooze fest wall of text from SteveW that is all theory. Surprise.

Not seeing any personal test of 8 core / 16 thread CPUs with V4 in that thread from 2014. So that's still a no.

So i've not personally tested all CPUs but at least I've had plenty of feedback on systems with up to 16 cores, still the same results are had from any set of four LPs enabled on four different cores usually does the business, unless something else is up. At least we all know that the main core is the one rendering the fps and that depends on frequency - done and dusted. All that remains is the background processes and they can only go so far on the hardware.

[SteveW 1,792]

...It stands to reason that if HT Off appears to be better then it is either too hot with HT enabled, or errant processes are spawning twice as many threads bogging down the CPU - and that would include apps like FSX and P3D without an AM.

 

Going to six or more cores from four means we can quit screwing around trying to shoehorn addons into the mix. With v4 using more memory it will spend less time with I/O and memory shuffling and shows an improvement over v3 that way, but otherwise we can operate it the same way. As more cores are added the problems experienced with HT enabled will increase.

 

23 minutes ago, Daedalus said:

Thank you very much Steve. So, if I can manage the same overclock with HT-on and an AF=85 would perform better. I'll certainly try your suggestion for AF=85. I have tried to turn off unnecessary services that eat performance and don't use non-simulator related programs when flying. Expect a browser sometimes.

Will also try your Game Mode suggestion. Thanks a million :)

Yes otherwise HT would be worthless!

However I stress again that addon exe apps make more of a nuisance with HT on than off, so require more care. Try out the batch techniques to corral addon exe apps away from the cores used by the sim, or at least only allow them onto cores doing the background work.

[denali 74]

2 hours ago, loafer87gt said:

I tried some experimenting using Project Lasso as well as disabling hyperthreading, and I found that the best performance was with the default Windows affinity mask settings and running Windows 10 game mode to give the sim priority. That being said, V4 is still a no go on my system as I just can't get acceptable performance no matter what the detail settings. The stutters are just horrible, and far worse than anything I ever experienced with 3.2 or 3.4 . Regardless of the detail settings, about every second the game has a millisecond freeze which ruins the immersion as smoothness of the sim. I even tried installing it on my system SSD drive to see if it was a disk performance issue. Still stuttered like crazy though. I am running an older 6 Core / 12 Thread 3930K at 4.8 Ghz with 32 GB or ram, and a GTX 1070 video card. Sadly, I will be uninstalling this version of the sim until I get a newer system capable of running it. Kind of disappointing as this is a vanilla install of the sim, and it runs much significantly worse than V3 with FTX Vector, FTX Global, and Active Sky.

I'm using a 3770k with 2x 780 SLIed.  V4 is running better than I've ever had it.  I get over 30fps across 3 1080p screens, which is more than 6fps over what I usually had at 24, so a 25% increase is what I'm experiencing.  I use mostly photoscenery and AS16/ASCA.  I am very happy with it.  However, I have just renewed Windows 10, and just before that I installed V4 and it wasn't so good.  

[denali 74]

Oh, and HT on with 64 bits.  Don't ask me why.  I've forgotten so much as I keep learning.  But I know it's right.

[DaveCT2003 1,820]

6 hours ago, Nytro said:

No, v4 has still no multicore support, neither HT support.

This was true of FSX prior to SP1 (later, slight improvements in SP2).

The linked article below contains a direct quote from Phil Taylor, a member of the ACES team who worked in multi-core/thread support with Intel.  Phil himself set me strait on this back in 2008.

My very best wishes.

LINK:  https://software.intel.com/en-us/articles/microsoft-flight-simulator-x-soars-to-new-heights-with-multi-threading

 

[SteveW 1,792]

We can allow the sim to split over more and more LPs but no matter how many we use we cannot get more performance after a certain point.

We can use a stopwatch to time the reloading of a scenario and we can compare the fps we get for a particular scene if we keep the view. So it's fairly easy to find that limit. Remember to run tests a few times for each setting because loading P3D takes less than half the time if it's already been loaded once.

The difference between HT On and Off is small, the difference between most reasonably logical AMs is small, in fact difficult to measure. So what's changing when we try different AMs? The layout of other processes - look to those for your performance problems.

[SteveW 1,792]

...Try an AM to enable at least four LPs and watch which ones are used for loading and compare with what they do during normal run of the sim. When you look closely you'll see what Phil's on about - the sharing of work can be seen across those LPs during those different phases of the sim to utilise less cores effectively. Back then the dual core had recently came out with HT giving four LPs!

[joemiller 269]

3 hours ago, Sethos1988 said:

Who are you, Steve's cheerleading squad?

Guys come on.. Do you realize we have only a handful of folks with the amount of/ and knowledge that Steve has?  Hey, even developers such as Oliver and other guys come in here to post their hard-earned knowledge so we can all make our dream to fly, a reality-  jeeez, say Thank You! (Listen, and Learn). 

[SteveW 1,792]

It's not such a commercial asset knowing all that jazz these days. The article Dave presents is interesting because not only does it describe how the sim makes logical use of the available LPs through the mask, showing how processes share LPs efficiently and also how carefully the code has been made to run on more cores when available. Since the code has been carefully designed to split out to make sections leaner, it makes sense to put those on cores of their very own. However, the background tasks are slightly better suited still to breaking up and can utilise pairs of HT LPs very well, basically due to the work on FSX all those years ago.

[glider1 90]

Does the core usage graphs measure actual physical usage of the core or just allocation of processor time?

What I'm saying is I think the fact that core-0 is pegged at 95%-100% doesn't mean the core is actually working at 100%, just that it is allocated 100% to distribute load to the other cores as needed. This makes sense to me. You always need one logical process that is continuously allocated for the purposes of shedding load.

If the theory is correct, core-0 should not run hotter than any other core even if the performance graph indicates it is loaded to 95%-100%.

Would be good to get clarity on this, because the story about "the old outdated archaic single core processor FSX code still existing in P3D" continues to this day.

The reason I would like some clarity, is that it would be wrong from an engineering point of view, to physically stress only one core on the silicon die more than other sections of the die, because this would reduce the mean time before failure (MTBF).

@SteveW or anyone else, got a comment on this?