8 hours ago, SteveW said:

When talking about % use of a core, we need to know where you are in the sim, loading a scenario, flying at altitude, parked at the gate. It is terribly important to ascertain the type of situation that is required for best performance aspects, maybe that's landing for me...

My tests use exactly the same non-flying situation in each case. It was the comparison which I was asking about. How restricting the number of cores available reduced the one hitting 100% to 70%, equal with the three others. I think you suggested a possible reason in a subsequent message.

8 hours ago, SteveW said:

The reason for 100% on the main task is that it is given too much work to do.

Yes, of course.

7 hours ago, SteveW said:

Use only enough cores to improve the scenario loading time with the main airport.

Why scenario loading?  I only load a new scenario at the beginning of a session. I always start a new scenario with a new session. Surely it is performance at or over the heaviest airport which is important? I know it is to me.

I get superb performance flying most places. There are just certain critical places, and Aerosoft (or UK2000) EGLL is one of the most, if not the most 'critical'.

Most of my settings ARE already really low. Having three scenery Windows is the killer -- with only one, the same size (in terrms of pixels), frame rates double easily, if I let them (they are limited to 30 max via VSync). But for a 210 degree view on my curved screen three windows are needed to avoid extreme distortion towards the extremities.

I have sorted out the settings and CPU arrangement to achieve a usable performance (averaging 20 fps at EGLL and reasonably smooth), but that's with the settings lower than I want -- especially AI Traffic, which I value highly for realism at airports if not in the skies. All I am hoping for with my newer system is for enough overhead to enable me to push the settings up a bit. If i can do that and achieve at least a smooth 20fps I'll be a happy bunny. 😉

Oh, BTW, all the above applies to DAYLIGHT use. Trying to use the sim at night is another matter -- performance far too low. I actually purchased Black Marble after reading the reviews, thinking that would give me better results, but, no, the BM layers made it even worse and affected daytime too. Chris Bell did try to help, even with specially made files he supplied, but there was still a noticeable adverse effect. But I'll be trying that add-on again with my new system, just in case. I hope too, then, to be able to make use of my purchased True Earth Netherlands and the upcoming GB equivalent.

Ah, well, one can only hope ... 😉

Good flying all -- I think i've said enough in this thread now.

Pete

 

 

Edited January 20, 20197 yr by Pete Dowson

1 hour ago, Pete Dowson said:

That's what I tried with my current 7900X system, despite it being supplied by the builders (Palicomp, UK) ready overlcocked on all cores to 4.8GHz (my request).  The AISuite software got up to 5.2 which failed, and retreated to 5.0. I tried that for a while but got blue screen crashes in P3D. Not immediately, but within an hour each time. I think P3D is more of a test than that carried out by the automatic software, which doesn't run long enough in any case.

I then tried 5.0 on core 0 and compensated by reducing two others to 4.6. That worked, but gave me no better performance. So I ended up restoring the original as-supplied settings.

For my new 9900K system next month I'm relying on a friend's reknowned expertise to get me a good and stable overclock.

Hi Pete,

Perhaps I just got lucky or this particular CPU is/has been a good example. I’ve overclocked a few times using the ASUS DIP5 utility and each time it got up to 4.8GHz and promptly crashed. It then backed off to 4.7GHz on Cores 0-3 and 4.6GHz for the remaining 4. This configuration persisted until I reverted to defaults before installing Windows 10 under a Dual Boot arrangement. I have now performed an all-core clock at 4.6GHz. In each case stability reigns. Fingers crossed that continues!

I wish you every success with your new system. Those that know how to accomplish a good manual overclock can squeeze that bit more from the CPU than any automated tool. I choose the latter simply because it’s easier and much less time consuming. I may not be so fortunate next time around if I ever decide to upgrade again....who knows!

BTW, regarding Black Marble don’t forget to redownload the latest NG installers from your shop account. There have been quite a few improvements since those early days following release. I think Chris has it nailed now. I’ve installed the lot including the excellent Stargazer package. Great product!

Regards,

Mike

What we are seeing is that the problems associated with a heavily loaded sim do not allow the sim to work in the same way as it does when loaded less. With Pete's system showing no improvement from 5 - 5.2 shows that there's a limit elsewhere unhelped by that.

Having a setup working well and then moving to a location that overloads it means quite simply that's the wrong setting for that location.

5 hours ago, Pete Dowson said:

 

Why scenario loading?  I only load a new scenario at the beginning of a session. I always start a new scenario with a new session. Surely it is performance at or over the heaviest airport which is important? I know it is to me.

 

 

 

Seems you misunderstood me Pete.

The loading speed test is suggesting you use that heaviest airport- as you say, what would be the point of the test of the easy airport - that makes no sense whatsoever.

Loading the heaviest scenario a few times (so it's cached) and then comparing the time it takes to another setting is *the biggest hit the system takes* during the sim flight.

Moving to another location with more throughput requirements is a basically another test another setup. The theory holds for that too.

 

13 hours ago, SteveW said:

Use only enough cores to improve the scenario loading time with the main airport. If only one or two seconds are gained by adding another core, that's probably excess to requirements. Too many cores allocated will eat into shared cache, slow down the entire process, and make more heat.

 

 

The heaviest place you will visit.

12 hours ago, GSalden said:

So can we see it like this :

- HT on : lower OC = less FPS, but less fluctuation in FPS = smoother

- No HT : higher OC = more FPS , but with larger fluctuation in FPS = less smooth

?

Although we say 'lower fps' it's going to be the same HT On or Off with the same OC. HT Off and an increased overclock appears to allow slightly more fps because the items in the scene are loaded.

But when new items are loaded then we do actually see less fps with HT Off. Again, setup not showing this behaviour is redlined somewhere, then, who knows what the performance characteristic will be.

 

 

 

It was the same with FSX, we could keep adding LPs (or cores) and the sim loaded faster and faster. But as more are added, eventually the sim loads half a second faster with one more core.

So what do we do? If we have lots of cores do we allow that core? Maybe.

If we have few cores, then it is better to not dedicate the core to a task.

Look at the Task Manager graphs and we see most LPs go up to 100% during loading the scenario. These LPs 'blip' like that during the run of the sim as parts are loaded.

If we allocate too many LPs to loading (on that particular machine) that takes throughput from the system and it produces less fps as the linkage is flawed. If that main task is 100% it's going to like it even less.

So on any particular machine the balance is found very quickly by the loading speed test (at the heaviest airport you will be visiting).

We can have multiple setups and use a different one for fast flying (pass over terrain quickly)  and gliding (turning all the time) might require a slicker fps setting. Study level sim? Study the plane not the scenery!.

1 hour ago, SteveW said:

Although we say 'lower fps' it's going to be the same HT On or Off with the same OC. HT Off and an increased overclock appears to allow slightly more fps because the items in the scene are loaded.

But when new items are loaded then we do actually see less fps with HT Off. Again, setup not showing this behaviour is redlined somewhere, then, who knows what the performance characteristic will be.

 

 

 

With my system ( 7940X ) and HT on there is a lot of extra heat generated. Therefore I need to OC to 4.6 Ghz max on the 8 P3D cores instead of 4.8 Ghz with HT off. Then the max framerate is lower ( + - 2 - 2.5 fps ).

And with 3 4K displays connected to 1 gpu  I need all power I can get 🤔

Edited January 21, 20197 yr by GSalden

Yes, the more cores there are, the more HT enabled will increase heat. Those spare cycles are used to do real computations and that means running P3D tasks more often.

If we use heat generated as a guide to the work done by a CPU that might not give us the right idea. Simply because it may be set to do too much of the wrong work and not enough of the right work. It is that simple. Too much heat and not enough performance in P3D = wrong setting.

 

...too much heat with HT enabled, we reduce the amount of cores where we might double the tasks for P3D (and maybe some addon).

For example; with lots of cores and P3D simply allocate '01' on each P3D core, this still leaves the system to utilise the HT mode more effectively and allows the OC to push the fps up....

9 minutes ago, SteveW said:

Yes, the more cores there are, the more HT enabled will increase heat. Those spare cycles are used to do real computations and that means running P3D tasks more often.

If we use heat generated as a guide to the work done by a CPU that might not give us the right idea. Simply because it may be set to do too much of the wrong work and not enough of the right work. It is that simple. Too much heat and not enough performance in P3D = wrong setting.

 

Will try with a lower OC + HT on.

What would you suggest as the HT on value for 1847 ( HT off ) ?

If you want to try what I use now and always used in the past (also with a 4 core HT CPU), you have to use an AM enabling all "real" cores but not the "HT cores". Means 85 for a four core HT processor (0101 0101), means 1365 for my current six core HT processor (0101 0101 0101) and would mean 89478485 for your 14 core HT processor (0101 0101 0101 0101 0101 0101 0101). Sounds rather strange to me, but that's what I get with this binary code.

Your AM=1847 results in 0111 0011 0111, which does not make sense on a 14 core CPU without HT to me, as it covers "only" 12 of your 14 cores.

But I might be wrong, with this huge amount of cores, maybe the calculation is different?

42 minutes ago, AnkH said:

Your AM=1847 results in 0111 0011 0111, which does not make sense on a 14 core CPU without HT to me, as it covers "only" 12 of your 14 cores.

But I might be wrong, with this huge amount of cores, maybe the calculation is different?

Every 0 at the left is being left out in the AM calculation.

AM=1847 is 000111 0011 0111

The P3D cores are the 8 fastest cores of the processor.

 

God lord, I have never seen such a long subject regarding AF.

@SteveW, there is clearly a market for this m8, why don't you just write an add-on that calculates all these for the end user and configure it on their behalf? you will get lots of money .

S.

51 minutes ago, simbol said:

God lord, I have never seen such a long subject regarding AF.

@SteveW, there is clearly a market for this m8, why don't you just write an add-on that calculates all these for the end user and configure it on their behalf? you will get lots of money .

S.

Yes please.  We must be on Volume 21 of AM discussion.  

End the pain with an app😛

 

3 hours ago, GSalden said:

Will try with a lower OC + HT on.

What would you suggest as the HT on value for 1847 ( HT off ) ?

1847 = 11100110111 HT off

If we want HT On we first convert as such

01,01,01,00,00,01,01,00,01,01,01 = 1377557

So now we can use HT on and the sim, as far as it is concerned, continues just the same - it has no idea of the changes.

That's 8 cores. How much time difference loading the main location with 6 cores and 8?

01,00,00,01,01,00,01,01,01 = 66837