XP12 CPU and GPU optimization....

[Jude Bradley 659]

18 hours ago, Beerz said:

file a bug report, xp12 is EA/Beta👌😊👍

optimizations will come in future updates

Bug is already reported, https://www.x-plane.com/kb/x-plane-12-00-release-notes/

No need to file duplicate bug reports.

 

[blingthinger 667]

1 hour ago, jcomm said:

Turned SMT Off in the BIOS.

Had to re-activate the FF A320 U 😕

So far I didn't notice lower Temps in the CPU / GPU - still CPU around the 60s, GPU around the 70+ s which I don't like to see 😕  Fans run like mad.

In general, I've never noticed a worth-while benefit from SMT. I turn it off because I'm frequently compiling code and there is actually a significant temp reduction, but only when the cores are maxed out. Mid-usage temps don't change.

LR warned us the GPU was going to be well loved in 12. Technically speaking, 70s for a GPU is actually not terrible. They usually don't start thermal throttling till high 90s.

My GPU fans scream every time I start the sim. It's at +90% usage most of the time (temps also 60-70s). It caught me off guard too because most apps don't actually use the GPU like Vulkan enables you. LR is saying: all of your GPU cycles are belonging to us now.

[AUA425 130]

2 hours ago, mSparks said:

I was working through understanding the settings today.

The impression i'm left with is many of the sliders are more of a "it works until it doesnt" nature (similar to vram) rather than an even relationship between quality and fps. when you start getting close to the limits of the gpu the fall off in fps is quick.

Which I rationalise as you have a gpu compute "budget", as long as you are inside that budget, stuff runs as it should, but as soon as you exceed it, stuff needs to get queued up and takes much longer to finish.

e.g. render distance - turn down world objects and render distance can quite happily be increased.

That said, I was mostly concentrating on keeping greater than 60fps in VR, which due to the lower resolution than you are using render distance is much less noticeable (so lower render distance and higher world objects looks better in VR compared to 2D).

Also, dont forget the clouds, if you are testing in clear weather the clouds arent using any gpu budget, which will cause a hit when you start to fly in weather if you are right on the limit without weather.

All that said, in 2D, with everything maxed except AA (which I might see if I can raise for VR), Im not seeing less than 45fps at 2k in all the conditions I tested (rtx3070) 

I also dont want to spend to much time fiddling with settings atm, because I expect their behaviour will all change in the nearish future.

EDIT: also, _don't_ forget to restart XP _any_ time you make a settings change before drawing conclusions, it looks like a lot of settings are not fully applying yet until you fully restart.

Thanks for the tips, very kind but I know all that - I'm an old XP hand - I was one of the first to experiment with the DataRefs (cloud shadow and ATMO tweaks) - I don't think it was understood what I meant.- it doesn't make any difference (unless everything is set to very low) - even that doesn't really change e.g. the CPU usage.
Even if I set the render settings via the DataRefs to 3.5 (I think is theoretically lower than setting "low") it gives FPS over big cities between 22-27 depending on city and objects.
Anything above that will drop the FPS to 15 - 18 - 22 - but only then does the whole thing look really good (and in combination with SimHeaven's scenery and Ortho4Xp and with the new XP12 lighting engine it looks pretty word not allowed good).
But there is still room for improvement - just wait and see. Of course you have to find a compromise between looks and performance - but as I said the system load is not there - if my (even less cores) were running at 100% and the graphics card load was at 95% I would know where the bottleneck is - but it's not - that's what I meant !
But it's also a beta and it will certainly be optimised further, it was just meant as a comment.
But as you rightly said - you can't commit to anything yet - it was only one of the first tests!
Thanks anyway for the tips!
BR !

EDIT: second test:

I think render distance is only calculated over one core and in this case Core 1 (if you start at zero).
Because if you reduce the render distance to e.g. zero via DataRefs, the CPU and GPU load increases.
If you increase this to high, (render distance for example), the CPU utilisation also goes down (strangely enough - on core 1) and the GPU utilisation also goes down (FPS also) - so  it´s clear - CPU limited. All other cores are at 9-15 % - so there is definitely a "worm" in there.

 

Edited September 18, 2022 by AUA425

[Murmur 3,965]

The devs confirmed that it's gonna be optimized for better performance, so let's see what they can do.

[mSparks 2,094]

23 minutes ago, AUA425 said:

it gives FPS over big cities between 22-27 depending on city and objects.
Anything above that will drop the FPS to 15 - 18 - 22 - but only then does the whole thing look really good

let me put it another way, EGCB, looking east over Manchester

lost maybe 5-10 fps during screenshot taking.

But if you are only seeing 20-30fps on a 3090, something else is off (duff windows perf maybe, underclocked cpu etc)

 

[mSparks 2,094]

While I am in no way, saying I'm not hyped for more perf improvements... Duuuuudde

[Bjoern 1,171]

Y'all AMD users must overcome your Intel mindset from ten years ago. As stated by AMD after the Zen 3 release, a CPU temperature of 80°C with short higher spikes is no reason to worry at all and actually an inherent feature of Zen 3 CPU to maximize its performance. And like any modern CPU, (boost) clock speed will simply be reduced to keep temperatures in check and there is an auto-shutoff if things get really out of hand.

However, since better cooling translates into longer phases at maximum boost clock, I keep my 5800X under a Noctua NH-D15S (with a second 120 mm fan at the front) constantly running at 100% RPM. That way, temperature spikes caused by sudden increases in frequency demand are more quickly taken care of than with a fan RPM curve. And I prefer the constant noise over the variable one.

 

Zen 2 and 3 CPUs offer the "Curve Optimizer" function in the mainboard BIOS, which is basically a minor voltage offset for each or all cores. With a bit of patience for stability testing and at the risk of instability, one can find a negative voltage offset that decreases core voltage and subsequently keep the CPU cooler. I got lucky and my 5800X runs just fine with -15 on all cores.

See here:

 

[mSparks 2,094]

sidenote: maybe there's a bug in the multiple monitor pipeline.

[AUA425 130]

3 hours ago, mSparks said:

sidenote: maybe there's a bug in the multiple monitor pipeline.

thanks for the screenshots by the way - yes that could be - if I reduce the resolution it looks completely different of course (or use only one monitor).
 

EDIT:

must correct myself - with the default scenery (had not completely "disabled" everything during the test - SimHeaven) the difference is about 10 FPS, so it remains quite smooth even in KLAX or elsewhere, even with LOD 1.5 = high - global forests enabled !!

 

 

 

 

 

 

 

Edited September 18, 2022 by AUA425
upload of the pictures must be changed !

[mSparks 2,094]

2 hours ago, AUA425 said:

thanks for the screenshots by the way - yes that could be

As much for me as you, in 6 months or so they'll be the evidence of any (or not) perf improvements, when "the usual suspects" start claiming there was no perf improvements.

does FSR help? I do have a multiple moniter setup, but setting up the viewports is a complete mare on linux

(no UI, gotta edit scripts), and since Im VR or die now Im not sure Ill get round to it any time soon.

3 hours ago, AUA425 said:

had not completely "disabled" everything during the test - SimHeaven

one of the things that definitely makes a large difference for me is the lvl16 cloudless ortho I use, because 2D autogen sucks up both cpu and gpu budget, whereas ortho is much less demanding generating shapes and switching multiple textures.

[jcomm 6,021]

Well, my "OCD" was preventing me from playing XP12 while listening to that frenetic spin of the GPU fans, ad feeling the case getting really hot with core temps getting above 80ºC 😕

Until last night I decided to give Riva Tuner a try, downloaded it, installed, watched a couple videos teaching how to do "undervolting", decided not ot go that way because it was bedtime and I was slippy, but used the simple sliders that can control max Temp and Power and pushed both full left, which sets max Temp at 65ºC 🙂

Loaded XP12, made a few tests and, surprisingly while the fans rotated at the level they used to while using MFS, and the Temps never got above 70ºC, the smoothness and FPS didn't change at all n XP12 with good sliders for nice displays and dense weather + SimHeaven Europe and America !!!

Voilà!!

 

[jcomm 6,021]

Well, my "OCD" is also some kind of "hyperactivity" because where I wrote in the previous post "Rivatuner" I wanted to write "MSI Afterburner" 😕  That was actually what I installed and used to fine tune the max Temp and Power settings for my 3060 Ti GPU...

Afterburner (msi.com)

In this link you can see bellow "FAN" on the right pan, two sliders, one for "POWER LIMIT" the other for "TEMP LIMIT", which can be clamped, and what I did was moving the TEMP slider full left, which left the power slider @ 50%. 

Last night I further used Nvidia Control Panel to set max FPS = 30, but I will also try Vsync 1/2.

Together these settings are providing extra smooth / stutter free experience AND no high Temp or crazy Fans !

[jcomm 6,021]

Well,

one of mSparks's posts called my attention. He was mentioning that some addons can really make good use of the remaining threads, so, I re-enabled SMT for my 5600x and went flying the FF A320 in XP12 beta 4...

I was suprised to find out that ALL of the 16 cores filled with tasks, and indeed one thing I really can't complain about in XP12 is performance and smoothness. In that aspect it runs as smooth as MFS, no doubt !