Board index FlightGear Support Graphics

Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Graphics issues like: bad framerates, weird colors, OpenGL errors etc. Bad graphics ar usually the result of bad graphics cards or drivers.
Forum rules
In order to help you, we need to know a lot of information. Make sure to include answers to at least the following questions in your initial post.

- what OS (Windows Xp/Vista, Mac etc.) are you running?
- what FlightGear version do you use?
- what graphics card do you have?
- does the problem occur with any aircraft, at any airport?
- is there any output printed to the console (black window)?
- copy&paste your commandline (tick the "Show commandline box on the last page of FGRun or the "Others" section on the Mac launcher).
- please upload a screenshot of the problem.

If you experience FlightGear crashes, please report a bug using the issue tracker (can be also used for feature requests).
To run FlightGear on old computers with bad OpenGL support, please take a look at this wiki article. If you are seeing corrupted/broken textures, please see this article.

Note: If you did not get a reponse, even after 7 days, you may want to check out the FlightGear mailing lists to ask your question there.

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby V12 » Wed Nov 20, 2019 8:42 am

Yesterday I tested P3D v4.3 on PC very similar with my new machine (AMD R7 3700X). And it is absolutly amazing, how good multithread working. No jittering, no stuttering, fps do not drop under 30 on max settings (with default and V-Sync OFF not under 60). CPU almost 100%, GPU almost 100% (fps not throttled down). Later I checked some discussions about P3D technology and reason is simple - this app can spread processes efffective on all CPU cores, for example, FDM runs on #1, addons things (gauges, etc.) on #2, weather system #3, scenery precomputation and texture generation #4 etc.

Dear developers, You should consider to rebuild FG's kernel in this way, because adding additional stuff like OSM objects into new scenery will fps push down more and more, even on CPUs with highest singlecore power.
Fly high, fly fast - fly Concorde !
V12
 
Posts: 2757
Joined: Thu Jan 12, 2017 5:27 pm
Location: LZIB
Callsign: BAWV12

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Thorsten » Wed Nov 20, 2019 8:51 am

and reason is simple - this app can spread processes efffective on all CPU cores, for example, FDM runs on #1, addons things (gauges, etc.) on #2, weather system #3, scenery precomputation and texture generation #4 etc.


That's all nice PR, but in reality doesn't help with framerate:

Image

Whether you distribute the few percent that are not rendering across several CPUs or not isn't really much of an issue (in fact, in a somewhat unorganized design that is FG, you're prone to lose performance when synchronizing the different threads). The whole real problem is to organize rendering.
Thorsten
 
Posts: 12490
Joined: Mon Nov 02, 2009 9:33 am

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby V12 » Wed Nov 20, 2019 10:32 am

On the Your graphs PR, in the reality clearly visible difference - no jittering, stuttering or lagging with 200 ms frame pauses...
Fly high, fly fast - fly Concorde !
V12
 
Posts: 2757
Joined: Thu Jan 12, 2017 5:27 pm
Location: LZIB
Callsign: BAWV12

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Thorsten » Wed Nov 20, 2019 10:38 am

Yes, so let's do an exercise in epistemics:

You observe differences - how do you know the differences are related to the thing you claim causes them? Could not other factors (differently organized rendering pipeline, less FDM realism, Windows-specific optimizations,...) cause the same observation?
Thorsten
 
Posts: 12490
Joined: Mon Nov 02, 2009 9:33 am

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby V12 » Wed Nov 20, 2019 11:27 am

I can see :
- not very advanced graphics in FGFS
- much better visual appearance in FSX, P3D
- no stuttering, higher fps in P3D
- CPU loaded only on 6-7% on 8 core CPU with HT and GPU is only on 30% load in FGFS and very low fps in some cases (EDDF, Paris with OSM etc)
- full CPU / GPU load higher fps in P3D with far better visual appearance than in FG

Summed - In FSX partialy, in P3D v4 very good usage of the CPU and GPU resources.
Fly high, fly fast - fly Concorde !
V12
 
Posts: 2757
Joined: Thu Jan 12, 2017 5:27 pm
Location: LZIB
Callsign: BAWV12

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Thorsten » Wed Nov 20, 2019 1:03 pm

You didn't answer the question - where is the evidence for multithreading and not any of the other possible factors being the decisive thing?

- not very advanced graphics in FGFS
- much better visual appearance in FSX, P3D


That's mostly texture quality / artwork which has zip relationship with execution speed - in actual reality, our rendering which does not rely on pre-computed textures for e.g. seasons but can change dynamically is quite a bit more powerful (and involved).

See, that's the problems with your posts about the topic in a nutshell - you're not even making an effort to understand what actually goes on, you're just mixing it all together.

X-Plane used to fill all CPUs by computing real FDMs for AI traffic (not sure whether they still do it) - it was completely useless for the flight experience itself, but made all users happy because they saw all their CPUs fully used. I don't think I have any problems coming up with a few threaded Nasal loops which fill all of your CPUs to the brim - so then you would conclude that FG suddenly managed to use your computer really well?
Thorsten
 
Posts: 12490
Joined: Mon Nov 02, 2009 9:33 am

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby V12 » Wed Nov 20, 2019 2:08 pm

Check discussions about Affinity mask in FSX or P3D and You will see that 13 years old FSX is multithreaded but poorly optimized, P3D v3 is better but still with 4 GB limit, XP and P3D v4 are much better optimized and not limited with 4 GB. When I move AI FDM computation from main thread to another running parallel with main thread, main loop is shorter and may run with higher frequency and ofcourse faster.
Fly high, fly fast - fly Concorde !
V12
 
Posts: 2757
Joined: Thu Jan 12, 2017 5:27 pm
Location: LZIB
Callsign: BAWV12

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Thorsten » Wed Nov 20, 2019 3:03 pm

I don't want to check any discussion - I want the evidence that prompted your conclusion that multi-threading is the relevant factor. I'm quite certain if I install any 13 year old software - multi-threaded or not - it'll be quite fast on modern hardware - much faster than modern software. Because it uses 13 year old technology to do its thing and had to cope on 13 year old hardware.

So there's still nothing about multi-threading in what you argue, quite the opposite.

When I move AI FDM computation from main thread to another running parallel with main thread, main loop is shorter and may run with higher frequency and ofcourse faster.


Yes - have you ever written code for multiple CPUs? Because I have.

What you describe is the theory. In practice though, you need extra infrastructure to synchronize the two instances properly (if you don't, you get crashes) - which means you need to make processes wait for the sync point and you need to use the sync infrastructure rather than just using variables as you need them.

So it may happen that you make the two bits of code run in parallel much slower because they have to wait for each other to reach the next sync point whenever you want to exchange a variable.

Basically code has to be designed up-front to run in parallel and the problem has to be suitable in the first place - otherwise you don't gain anything. If you would re-write 20 years of FG with a strict design idea in mind, yeah - you'd get somewhat faster code (not much, because rendering is still the main chunk). But there's no workforce to do that, we're lacking a few dozen man =-years here. If you try it with the existing code which is grown rather than designed - good luck.
Thorsten
 
Posts: 12490
Joined: Mon Nov 02, 2009 9:33 am

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Richard » Wed Nov 20, 2019 8:14 pm

As I've previously mentioned I've done quite a lot of research in this area and there is some extra performance to be had from using multiple cpu cores; however my diagram (measured using 2019.2, F-15) is very similar to that which Thorsten postead earlier

Image

I'm using the F-15 here because I know that it is pretty well optimised for both Nasal and rendering performance; however it is still heavier than a lot of aircraft at an average of 3.7ms.

When I'm using my experimental multi-threaded rendering there is a small increase in performance which is almost equivalent to the 3.7ms of "the rest" - that usually works out to about an extra 10fps (on my system, with the F-15)

Image
Richard
 
Posts: 810
Joined: Sun Nov 02, 2014 11:17 pm
Version: Git
OS: Win10

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby GinGin » Wed Nov 20, 2019 8:26 pm

Interesing and FACTUAL datas Richard, thanks .
How efficient are the multi thread lines that we can add in the launcher as properties ?


X-Plane used to fill all CPUs by computing real FDMs for AI traffic (not sure whether they still do it) - it was completely useless for the flight experience itself, but made all users happy because they saw all their CPUs fully used




Very interesting. I wouldn’t have expect that
Last edited by GinGin on Wed Nov 20, 2019 8:30 pm, edited 1 time in total.
GinGin
 
Posts: 1580
Joined: Wed Jul 05, 2017 11:41 am
Location: Paris
Callsign: Gingin

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby Thorsten » Wed Nov 20, 2019 8:28 pm

that usually works out to about an extra 10fps (on my system, with the F-15)


An extra 10 fps on top of what baseline?
Thorsten
 
Posts: 12490
Joined: Mon Nov 02, 2009 9:33 am

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby GinGin » Wed Nov 20, 2019 8:33 pm

Thorsten wrote in Wed Nov 20, 2019 8:28 pm:
that usually works out to about an extra 10fps (on my system, with the F-15)


An extra 10 fps on top of what baseline?


On the picture , it seems to,be 70 with single and 80 with multi
Not massively significative
GinGin
 
Posts: 1580
Joined: Wed Jul 05, 2017 11:41 am
Location: Paris
Callsign: Gingin

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby WoodSTokk » Wed Nov 20, 2019 9:22 pm

Hmm, interesting. Thats about 15% with F-15.
Can you check a small less developed aircraft to see the impact on that?
WoodSTokk
 
Posts: 1077
Joined: Tue Oct 17, 2017 3:30 pm
Location: Milky Way/Sol/Earth/Europe
Callsign: SX-W57
IRC name: WoodSTokk
Version: 2020.4.0
OS: Debian Bullseye

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby wkitty42 » Wed Nov 20, 2019 9:51 pm

GinGin wrote in Wed Nov 20, 2019 8:33 pm:On the picture , it seems to,be 70 with single and 80 with multi
Not massively significative

that depends... on my system with its lowly NVIDIA GeForce GT730 GPU, 1-FPS is significant... i can fly at 8-10FPS but it is noticeably jittery... 15-20FPS is OK... 20+FPS and we're having a good time... the thing is that it also depends, in my case, on the weather (clouds being rendered is a big one) and where i am (paris is a killer, boston livable, ksfo generally nice) also plays a large part... in the few years (5 or so) that i've been here (with the same machine setup) i've had to gradually reduce some of my settings to try to have decent flights... these things are also one reason why i generally fly the UFO ;)

here's another example: my system is 4Ghz AMD FX8350 8-core 16Gig RAM with the afore mentioned GT730... i've updated to the latest SpaceShuttle development branch... some weeks ago, i was able to load the craft and get into space... today with a "Nightly/RelWithDebInfo" build of FG's latest code with the compositor enabled, i've waited for over an hour for the shuttle to load but it ain't happening... possibly because the sim cannot find some shaders, effects and fragments... maybe it is looping over and over trying to load them? i dunno... all i have is this...

Code: Select all
   66.74 [INFO]:input      aircraft = SpaceShuttle-launch
   66.74 [INFO]:general    found aircraft in dir: /home/wkitty42/myflightgear/Aircraft/SpaceShuttle
   66.76 [POPU]:aircraft   Aircraft does not specify a minimum FG version: please add one at /sim/minimum-fg-version
   66.76 [INFO]:general    Found language resource for: en_US
[...]
   70.41 [INFO]:view       Splash screen progress binding-subsystems
   88.91 [ALRT]:input      Could not locate shader: Compositor/Shaders/secondary_lights.frag
   88.91 [ALRT]:input      Error building technique: couldn't find shader Compositor/Shaders/secondary_lights.frag
   89.06 [ALRT]:input      Could not locate shader: Compositor/Shaders/secondary_lights.frag
   89.06 [ALRT]:input      Error building technique: couldn't find shader Compositor/Shaders/secondary_lights.frag
[...]
 6783.41 [ALRT]:input      Could not locate shader: Compositor/Shaders/secondary_lights.frag
 6783.41 [ALRT]:input      Error building technique: couldn't find shader Compositor/Shaders/secondary_lights.frag
 6790.80 [ALRT]:input      Could not locate shader: Compositor/Shaders/secondary_lights.frag
 6790.80 [ALRT]:input      Error building technique: couldn't find shader Compositor/Shaders/secondary_lights.frag
 6798.24 [ALRT]:input      Could not locate shader: Compositor/Shaders/secondary_lights.frag
 6798.24 [ALRT]:input      Error building technique: couldn't find shader Compositor/Shaders/secondary_lights.frag


so taking 70.41 and subtracting it from 6798.24 and dividing that by 60 gives me 112 minutes that i've been waiting... yes, one CPU is pegged so it is processing something... no clue as to what or where in the init process of the shuttle it is... good thing i've been busy doing other things while waiting... so at this point, my only option is to send fgfs the TERM signal to get out of it... i'm probably going to disable the compositor, rebuild and try again... i started this all because i simply wanted to see how fast the shuttle loaded compared to the developer's stated 3 minutes on they gaming laptop... sorry, guys... i still can't keep the t[h]orstens straight :(
"You get more air close to the ground," said Angalo. "I read that in a book. You get lots of air low down, and not much when you go up."
"Why not?" said Gurder.
"Dunno. It's frightened of heights, I guess."
User avatar
wkitty42
 
Posts: 9146
Joined: Fri Feb 20, 2015 4:46 pm
Location: central NC, USA
Callsign: wk42
Version: git next
OS: Kubuntu 20.04

Re: Threadripper 1920 (12/24 threads)+Vega56 = 10fps in W10?

Postby wkitty42 » Thu Nov 21, 2019 3:29 am

i wanted to touch back on my post above... i recompiled my FG without the compositor and ran the above again... this time it was fine... compiled in Nightly/RelWithDebInfo, the shuttle loaded in 3 minutes, 3.47 seconds if the time stamps in the log are accurate... i recompiled again in Dev/RelWithDebInfo and loaded in 2 minutes, 50.42 seconds... again, if the time stamps are accurate... i've done three or four launches with no problems... external view at launch predictably knocks my frame rate to the floor but as we ascend, it gets a lot better... i know i saw 75+ FPS at some point... i've got a little more to figure out the steps of but a trial recording of one of the launches was great... once i get the steps down for proper launch and proper opening of the payload bay doors with antenna usage, i'll have to try to live stream it and see how that goes ;) :mrgreen:
"You get more air close to the ground," said Angalo. "I read that in a book. You get lots of air low down, and not much when you go up."
"Why not?" said Gurder.
"Dunno. It's frightened of heights, I guess."
User avatar
wkitty42
 
Posts: 9146
Joined: Fri Feb 20, 2015 4:46 pm
Location: central NC, USA
Callsign: wk42
Version: git next
OS: Kubuntu 20.04

PreviousNext

Return to Graphics

Who is online

Users browsing this forum: No registered users and 2 guests