Jump to content
Sign in to follow this  
David Roch

The BP=0 conclusions

Recommended Posts

Hi captains!Here are the conclusions of months of research leaded by ******* Altuve and reported in the longest thread in Avsim's history... :--)We, at Avsim, are appreciative of his fantastic work and devotion to the community and would like to express here our warmest thanks.This guide is not a "do this, do that" document. ******* explains every detail, concept or element, making them easy to understand to non-expert users.During the last month he also spent hours to educate people in this forum so there is a bunch of very smart guys that can take over and give him a well-earned rest. That's why we are locking this thread together with the "monster" BP=0 one and invite you to use the forum as usual for further information or questions about the BP=0 find.We ask everyone who copy this document elsewhere on the web to join a link to this thread and to give ******* the credit he deserves.Thank you in advance================================================DISCLAIMER:What you are about to read CAN PRODUCE UNFAVORABLE RESULTS depending on your particular setup and slider settings, so this is NOT a 'solution' to any problem, this is simply a hidden, undocumented tweak that can be applied to Flight Simulator X to obtain aproximately a 30% performance improvement.Before I begin, a little background.. I've used this tweak personally, for more than a year, second, I have dedicated considerable time 'understanding' the tweak, its purpose, and specially the 'reasons' why there is such a 'dramatic' performance increase. So please... DO NOT TRY if you have 'issues' or problems in your current setup, only try it *IF* you are absolutely certain that you have a fairly stable setup and that you UNDERSTAND what I'm explaining in the following lines:Lets first discuss 'how' applications, 'particularly' FSX works (I will keep it simple)When you 'fly' in FSX, each 'frame' its like a 'photo' it contains certain objects and textures that need to be 'drawn' for EACH 'photo'. When you play this 'photos' in rapid succesion, say, for example 25 'photos' per second, the perceived sensation is that you are 'moving' FPS = Frames per Second, each 'frame' is the 'photo' I'm talking about.In FSX, the 'rendering' or 'drawing' of this objects, occurs, on a per 'frame' basis.. the higher the frames per second, the higher the number of objects that need 'drawing' This 'drawing' works by having the 'CPU' instruct your 'video card' what to 'draw' on screen. Each 'instruction' the CPU sends to the video card is 'queued' in a 'buffer' FSX, then 'determines' the 'threshold' where that buffer is 'filled' to flush all the commands to 'another' buffer, this one is called the Command 'ring' buffer. This happens, so FSX can syncronize the CPU and the GPU in terms of instructions proccesed. If this 'buffer' the one created by the application, didn't exist, then you could 'under certain conditions' COMPLETELY CRASH FSX. The reason for this 'crash' is that FSX would have to send the intructions 'directly' to the 'Command Buffer', which is the one that the Video card reads to 'draw' the objects on screen. If the video card is not fast enough to process the INCREDIBLY high number of 'draw calls' sent by the CPU per each frame (photo), then STALLING occurs. Stalling, means that the CPU 'writes' into the command buffer FASTER than what the video card is capable of proccesing (the command buffer its a ring) its finite it can hold 'certain' number of instructions at a given time, if the video card doesn't read it fast enough it gets overwritten and video corruption occurs (you'll know this because spikes come out from the AUTOGEN)If you understand the above correctly, then you also must understand that 'application managed' buffers (aka BufferPools) 'cost' some CPU usage, and particularly in FSX this usage is MASSIVE! however, this is NOT a bad thing, its a 'safety' mechanism that the application uses so its almost imposible that it will 'fill' the command buffer with instructions. The 'Aplication managed' buffer, is like a 'middle man' it coordinates what goes into the command buffer and 'how' to keep things in sync so you don't batch an exessive ammount of objects for the video card to process.So, why does Microsoft includes an 'application managed' Buffer (also called Explicit Vertex Buffers)? they did so, so ANY ONE could use FSX regardless of setup and ensure 'up to a point' that this will perform reliably indepently of hardware. So, DONT BLAME MICROSOFT!! they did the right thing and this is NOT a bug.So, what happens when you tell FSX to 'bypass' this 'application managed' buffer and send all the 'draw' instructions DIRECTLY to the command buffer for the video card to process? (instructions are really sent to the D3D API, I'm just keeping the explanation simple for our sarcastic folks out there) well.. what happens is that FSX doesn't need the middleman anymore, what happens is that FSX doesn't need to control what to draw or when to draw it. Now, it simply pumps all the information as fast as it can to the Video Card, and by doing so you FREE a massive ammount of resources that are now allocated to your current flying session!! Now.. Do you understand why this tweak is SO powerfull, but at the same time, EXTREMELY dangerous for stability? It is because you need to have an INTIMATE knowledge of WHAT is going on in your computer!! you need to know 'exactly' how much is 'too' much for the GPU to process.. and this is, where things get interesting.How can you be 'sure' that your video card is in TOP shape and able to process ANY and EVEYTHING you throw at it? remember... this is NOT about 'raw' power... this is about the card ability to 'process' the commands sent by the CPU, FASTER than the CPU ability to 'send' those commands.. its like a 'race' and the GPU needs to be faster! so, this holds specially true for folks using overclocked i7's @ 4.2 and DONT BELIEVE your GPU % Usage meter! it only shows USER SPACE... it doesn't tell you how much the 'kernel' (system) is using to process commands in the queue.. AND don't compare FSX to Crysis please... those are games OPTIMIZED to make full usage of hardware programable shaders! they are perfectly balanced!well, you need to start forgeting EVERYTHING you have learned in the past 4 years.. specially things like:(AntiAliasing) but only when it is controlled by nHancer(Anisotripic Filtering) again, ONLY if controlled by nHancervSync set to ON (this is a KILLER)Multiple Monitor setups (even clone mode)The ENBSeries modthe above, KILL the card ability to process commands 'but not the ability to RENDER or DRAW the objects' so if you absolutely NEED the above, you can stop reading now ;) So, if you want to give this tweak a try, you need to start by using the application controlled AntiaAlias and Anisotropic filters, you need to also force vSync to OFF, and make sure you are running a single monitor setup. this will ENSURE, that the card is in top shape to process commands 'as quick' as it can... specially, if using nVidia cards (ATI's dont have this problem, more on that later)One side note: ATI's 5870 vs nVidias GTX 285 they are both EXCELENT cards... but they have completely different architectures, the ATI's have LOTS of tiny 'slow running' shader processors, the nVidias have fewer shader processors (a lot less) but they are TWICE as fast... so, how does this affects the cards ability to 'process' and 'render' draw commands?well.. its up to you to decide: the ATI's have 1600 small processors, they are called 'shader processors' the more processors you have, the better the card ability to 'multitask' and do parallel processing so, the ATI's are FANTASTIC reading the command buffer faster than ANY card on the planet (even the new nVidias GTX 480, also called by their codename Fermi) so with an ATI you could practically have FSX nirvana if FSX were to send all draw instructions to the command buffer bypassing the BufferPools.. however, THERE IS A CATCH. Since ATI's have more shader processors, they need to run 'slower' (they run at 700Mhz) same as the core clock. so, complex scenery, clouds, add-on aircraft will considerably LOWER your total average FPS. so, again, its up to you to decide... if you only fly default planes, and want EVERY SINGLE SLIDER MAXED out, vsync, nHancer, ENBSeries mode etc. then the ATI is for you, ITS IMPOSSIBLE TO CRASH IT! I could crash mine (I only owned it for like 2 days) no matter what I tried, but I lost 6FPS! to me thats unnaceptable.Now.. the nVidias, particularly the GTX 285, have exactly 240 Shader processors, they run at 1476Mhz (and some can be overclocked even higher) this card, is a monster... and even though the ATI's (in raw speed) are faster, the nVidias can render 'complex' scenes much quicker (specially things that relay on shaders such as clouds, high water settings, buildings etc.), so the CPU doesn't have to wait for a particular scene to be rendered. Remember, the FASTER the card 'renders' a scene, the faster the 'frames' will be processed and the CPU will keep producing them! so, as you see, there needs to be 'BALANCE'. remember, that any complex system will be limited by the speed of the slowest running member on the system. Now, back to video card comparisons: The downside to the nVidias?? THEY SUCK at reading instructions from the command buffer fast enough, so they can be stalled by the CPU IF you are running high frame rates and using complex autogen (which is what fills the command buffer quicker, specially after SP2 where there is massive object batching per frame) so when using nVidias LIMITING your framerate to 25-30FPS and lowering autogen is paramaunt. (including the steps I have already mentioned) like vsync OFF, single monitor setup, No ENBSeries and application controlled AA and AF. but don't worry.. The GTX 480's will change all that :) they have 480 shader processors :) still much less than the ATI's but enough to give you Flightsim nirvana and turn FSX into a whole new ball game. So, IT IS completely possible to achieve what everyone though was a dream. FULL MAXED sliders and fluidity (I don't include car traffic) MAX 2.0 Water (which is a killer) or bloom (the other killer) but you can still have pretty descent AI traffic and easily achieve 20-25 FPS under the most demanding conceivable situation, thats quite good.So... IF you understand the above information COMPLETELY and you are a competent guy that knows how to tweak your fsx.cfg file, then by all means, give this a go.. otherwise, don't even think about trying it, and please, don't private message me because I'll not respond. I'm providing all this information for the benefit of the community. The tweaks that you need to add to the fsx.cfg file are:[bufferPools]UsePools=0 (note that PoolSize is ignored if UsePools equals 0. Use 1 if you experience crashes )When you see 'toggle' values (1 or 0) it means ON/OFF - UsePools its an ON / OFF value (1 or 0)PoolSize its a 'size' value (in bytes) If you 'DISABLE' the pools, you will get increasedperformance, AND ALSO, instability if you don't 'balance' your components and sliders apropiately, in case of instability, then simply DO use pools by changing the value of UsePools to 1 and 'adjust' PoolSize. BE CAREFUL (and forget everything you have been told about this value)it does NOT use video memory, PERIOD. it uses SYSTEM memory, because its a special type of poolcalled Explicit Vertex buffer which DOESN'T GO INTO VIDEO MEMORY unless they have A VERY SPECIFIC FLAG(and they don't) more info here: http://msdn.microsoft.com/en-us/library/ff539490.aspx you can also do your own tests and see how 'increasing' PoolSize affects the size of the fsx.exe process proportionally.[GRAPHICS]HIGHMEMFIX=1 Fixes errors with texture addressing modes in WDDM1.0 and 1.1 when using a lot of video memoryThe HIGHMEMFIX=1 you see above, fixes a bug in the FSX engine on how it handles texture addressing modes (Wrap,Clamp) and initial render states on single pass shaders, it will completely prevent textures, buildings and entire cockpits from dissapearing! this 'bug' is triggered when there is a high video memory usage situation. so, enjoy :) this is my way of giving to a community that has given me so much over the years.---------------------------OPTIONAL[GRAPHICS]SHADER_CACHE_VERSION=1 // Increment this number EVERYTIME you change fsx.cfg (it simply rebuilds the shader cache)[DISPLAY]TextureMaxLoad=9 // Carefull. CAN induce stutters on LOW END systems! use multiples of 3 *ONLY* (3, 6, 9 etc) perfect for PhotoRealistic scenery and PNWThe formula on how to determine your 'optimal' TextureMaxLoad goes like this (credit to Steve Lacey here http://www.steve-lacey.com/blogarchives/20...blurries.shtml)If UPPER_FRAMERATE_LIMIT exists, then:MAX_TEXTURE_DATA = (TEXTURE_MAX_LOAD * (TextureMaxLoad * TEXTURE_BANDWIDTH_MULTIPLIER)) / UPPER_FRAMERATE_LIMITIf UPPER_FRAMERATE_LIMIT doesn't exist (UNLIMITED FRAMES) then:MAX_TEXTURE_DATA = TextureMaxLoad * TEXTURE_MAX_LOADAs you can see above, TEXTURE_BANDWIDTH_MULTIPLIER its just that, a multiplier. Evidently, changing it DOES change things, however,when running UNLIMITED frames, it is USELESS, so you are better of playing with TextureMaxLoad directly. be VERY CAREFULL, the 'resulting'value of this formula (MAX_TEXTURE_DATA) corresponds to the MAXIMUM number of bytes the TEXTURE MANAGER is allowed to 'upload' perframe. If the resulting value (in bytes) for MAX_TEXTURE_DATA is TOO HIGH you will spike your GPU! (and cause a stutter), so this is a 'test and see' value. Ideally, you'll see an impact on COMPLEX high resolution textures sceneries like PNW and/or PhotoReal scenery. It is important to mention that THIS value does NOT have an impact on CPU if you are running a Multiple core setup, due to the texture manager threads running in the cores responsible for texture loading and object batching. (which are the last 3 if you have 4 cores)Now, the following requires a careful explanation... please, READ. This 'assumes', you have an i7 With Hyperthreading OFF.. so, it goes like this;CORE0 CORE1 CORE2 CORE3CORE0 is responsible for: Fibers and main schedulerCORE1, CORE2 and CORE3 are responsible for the Texture Manager AND Object Batching (Autogen)'Fibers' are like small processes that do 'things' cooperatively with other 'processeses' (sorry, I know this is not I high tech explanation)I'm just trying to cater for everyone. In FSX. Fibers are responsible for the 'loading' of the terrain system. They need to communicate with themain scheduler (which also RUNS on CORE0) to 'coordinate' and cooperatively multitask on terrain rendering every time a 'frame' is rendered.Those two, the fibers AND the main scheduler are running on the SAME CORE... now, here's a trick. Fibers are STUCK to CORE0, you can not move themout of there, however, YOU CAN move the main scheduler to run on another core!! so, if you do this:[JOBSCHEDULER]AffinityMask=14you are telling FSX to use only CORE1, CORE2 and CORE3.. so, what about CORE0? it STILL runs the FIBERS! because they are not bound to the AffinityMasksetting. so, what you are doing is making things now much more efficent. So now, FSX will run like this:CORE0 CORE1 CORE2 CORE3CORE0 is responsible for: FibersCORE1 is responsible for: main schedulerCORE2 and CORE3 are responsible for the Texture Manager AND Object Batching (Autogen)thats great balance! (and a little increase in performance too!) thats why so many people say FIBER_FRAME_TIME_FRACTION affects their performancewhen they lower it! of course it does, we'll talk about that now :)Now, lets micro manage the 'fibers' and the terrain loading system. Consider this:[MAIN]FIBER_FRAME_TIME_FRACTION=0.25the above value means that 0.25 (which means 25%) percentage! is the % ammount of 'time' that the 'fibers' will 'cooperate' with the main schedulerto 'render' the terrain system. So, assume you run you sim LOCKED to 25 FPS... thats 25 Frames EACH second, correct? 1 second is the same as1000 miliseconds. So, if you divide 1000 miliseconds / 25 frames what do you have? you have Miliseconds PER frame. That number is 40 Milisecond.so, now we KNOW that EACH frame, takes aproximately 40 Miliseconds to render (if running frames locked at 25). Now, out of those 40 millisecondsPER frame, HOW MUCH time (or % of time) you want the fiber to cooperatively multitask with the main scheduler? well.. if you read re-read and analyzewhat Adam told Phil Taylor you'll see Adam hinted Taylor on 'what' the 'optimal' time is to do just that, Adam 'optimal' value required for the fiberto communicate with the thread is 10ms read here: (thats why they came up with the 0.33)http://blogs.msdn.com/ptaylor/archive/2007...-week-or-2.aspxSo, that value should be adjusted according to your frame rate. Of course, it REALLY doesn't have any IMPACT whatsoever UNLESS you are running a SINGLE core machine (so you can not offload the scheduler) or you run tileproxy or something VERY heavy. For cases like tile proxy or VERY complex terrain I assume 10ms is simply not enough so you need to give the fiber MORE TIMEotherwise, the blurries are going to be TERRIBLE!And talking about blurries, terrain and fiber time... :) we have: [TERRAIN]SWAP_WAIT_TIMEOUT=30 This variable is the number of frames that the terrain engine will wait for terrain textures to be loaded into video memory before forcing new tiles of terrain triangles to be rendered. This is because terrain triangles are bound to a specific texture resolution by the texture coordinates contained in each vertex. When the terrain engine wants to increase the texture resolution on a tile of terrain, for example when you fly closer to the tile, the texture coordinates must be updated to reference the higher resolution texture. However, the new tile can't be rendered successfully unless the new texture is available in video memory. The terrain engine will wait for some number of frames determined by SWAP_WAIT_TIMEOUT for the new texture to be loaded into video memory. If after SWAP_WAIT_TIMEOUT frames the texture still isn't available, the terrain engine starts using the new triangles anyway. This usually results in a tile of terrain that is either missing or textured strangely/incorrectly for a few frames until the texture finally shows up in video memory.(A warm thanks to Adam Szofran, a Microsoft ACES team member and well known community member for his clear definition :-))Now... lets talk about 'frame locking'You need to understand the following:When you 'fly' in FSX, each 'frame' its like a 'photo' it contains certain objects and textures that need to be 'drawn' for EACH 'photo'. When you play this 'photos' in rapid succesion, say, for example 25 'photos' per second, the perceived sensation is that you are 'moving' at 25 frames per second. FPS = Frames per Second, each 'frame' is the 'photo' I'm talking about.So, the above is VERY important to take into account, because in FSX, all operations! terrain rendering, object batching etc occur at the single 'frame' level... what does that mean to you? I means, the HIGHER the frames, the MORE work FSX needs to do.. so, say for example, that you run your SIM at 30 FPS and you like it.. but, then someone got into your head the idea that if its not 60 FPS then it is NOT real, well.. from 30 to 60 FSX will DOUBLE the ammount of things that it needs to do! so, don't be surprised when you start seening a christmas tree and strange spikes coming from the ground :).. I won't get into the 'FPS eye perceived thing' that has plaged our forums since day 1. run the sim at the FPS you like! I personally, got USED to run at 25, and let me tell you its perfectly fine to me.. because I got USED TO IT.. I hated it, but after a week, you simply 'adjust' your eyes to it.So, should you use the FPS limiter or run to UNLIMITED? well it depends on your flying style... if you run default planes or non CPU intensive planes or only 2-D cockpits, then your CPU is NOT going to be overworked.. in that case, simply go for the frame lock inside FSX.. period.Now, if you fly COMPLEX a/c specially using the VC you NEED to set it to unlimited! however... there is a problem ;) what happens with what I just told you about the sim working HARDER the higher the frames? what happens if frames start jumping from 30 to 60 then to 25 and again to 60? where is the 'balance' here?? what if you have UsePools=0? what do you expect will happen when you get a 'sudden' jump to 70 FPS? CRASH! ;) besides... the 'perceived' sensation (because is not real) when frames 'jump' is a 'stutter' so you need 'smoothness' consistency... and thats were the EXTERNAL frame rate limiter can help you ;) capiche?Now... use this as a REFERENCE... specially if using a HIGH SPEED CPU, and simply increase the Scenery Complexity and Autogen on at at time, and run your frames locked either in FSX or with the external limiter and setting the option to unlimited inside the sim. you want to find your 'sweet' spot, or the point where the GPU is not being overworked, I suggest, as 'reference' you use a GPU monitoring tool.. again, as reference, because what you see there is NOT 100% accurate.SPECIAL THANKS:I want to extend a public, sincere thanks to PMDG, because they were the only development group that took the initiative to 'organize' and actively work on ways to find a solution to the texture corruption problems. Without their involvement this could not have been possible. So, show some appreciation guys and buy something from them :) I based MUCH of my investigations on Ryan's work and theories.. so, please.. THEY DO DESERVE CREDIT TOO.Thanks to all the guys at Avsim, David Roch, Stephen, Mitch, David Alexander, J van E, Jjjallen, PingPong, DJJose and all those who supported me since DAY 1, trying stuff, reporting back and being extremely supportive, .. I REALLY need a break, so don't freak out if I take a long one! ;) (also excuse any mispellings, english is not my mother tongue)******* Altuve, April 2010(Contact) -------------------------------------------------------------------------------------------------------------------------- :( Added or modified information is in RED


Best regards,
David Roch

AMD Ryzen 5950X //  Asus ROG CROSSHAIR VIII EXTREME //  32Gb Corsair Vengeance DDR4 4000 MHz CL17 //  EVGA GeForce RTX 3090 FTW3 ULTRA GAMING //  2x SSD 1Tb Corsair MP600 PCI-E4 NVM //  Corsair 1600W PSU & Samsung Odyssey Arc 55" curved monitor
Thrustmaster Controllers: TCA Yoke Pack Boeing Edition + TCA Captain Pack Airbus Edition + Pendular Rudder.

 

Share this post


Link to post
Share on other sites

You can find ******* FSX.cfg hereAbout vertical synchronization read here


Best regards,
David Roch

AMD Ryzen 5950X //  Asus ROG CROSSHAIR VIII EXTREME //  32Gb Corsair Vengeance DDR4 4000 MHz CL17 //  EVGA GeForce RTX 3090 FTW3 ULTRA GAMING //  2x SSD 1Tb Corsair MP600 PCI-E4 NVM //  Corsair 1600W PSU & Samsung Odyssey Arc 55" curved monitor
Thrustmaster Controllers: TCA Yoke Pack Boeing Edition + TCA Captain Pack Airbus Edition + Pendular Rudder.

 

Share this post


Link to post
Share on other sites
Guest
This topic is now closed to further replies.
Sign in to follow this  

  • Tom Allensworth,
    Founder of AVSIM Online


  • Flight Simulation's Premier Resource!

    AVSIM is a free service to the flight simulation community. AVSIM is staffed completely by volunteers and all funds donated to AVSIM go directly back to supporting the community. Your donation here helps to pay our bandwidth costs, emergency funding, and other general costs that crop up from time to time. Thank you for your support!

    Click here for more information and to see all donations year to date.
×
×
  • Create New...