[posila][0.17.39] Full Screen Flickering at unreproducible intervals.

[eradicator]

After a few hours of play i sometimes get a fullscreen flickering of what looks like a bit of the sprite atlas (gif attached). I haven't been able to reliably reproduce this, but there's a few factors that seem to be fixed:

• It seems to only occurs after prolonged play. Could indicate system overheating, but i haven't seen problems with other programs.
• Seems localized to a fixed "zone" on the map. The "zone" is different each time i think, but not larger than a chunk (more like 16x16 tiles). When leaving the zone the flickering stops.
• The flickering only happens at certain zoom levels. Player zoom is write-only so i don't know the exact values. When zooming out of the "zoom zone" the effect temporarily stops.
• The effect completely stops after a while of being out of the chunk-"zone". I.e. reentering the "zone" won't re-trigger it. Though it can happen more than one time per play-session.
• There is no related log output.

Currently playing on a i7-3630QM's (mobile) integrated "HD Graphics 4000" graphics.

Code: Select all

   0.012 System info: [CPU:       Intel(R) Core(TM) i7-3630QM CPU @ 2.40GHz, 8 cores, RAM: 19387/32606 MB, page: 20185/48606 MB, virtual: 111/8388607 MB, extended virtual: 0 MB]
   0.013 Display options: [FullScreen: 0] [VSync: 0] [UIScale: custom (100.0%)] [Native DPI: 1] [Screen: 255] [Special: 00] [Lang: en]
   0.016 Available displays: 1
   0.016  [0]: \\.\DISPLAY1 - Intel(R) HD Graphics 4000 {0x8000005, [0,0], 1920x1200, 32bit, 60Hz}
   0.076 [Direct3D11] Display: 0, Output: 0, DisplayAdapter: 0, RenderingAdapter: 0; d3dcompiler_43.dll
   0.086 Initialised Direct3D[0]: Intel(R) HD Graphics 4000; id: 8086-0166; driver: nvumdshimx.dll 10.18.13.5560
   0.086   D3D Feature Level: 11.0, DXGI 1.2+, SwapChain: 1,discard,-,-,none
   0.087   BGR 565 Supported: Yes
   0.087   MaximumFrameLatency: 3, GPUThreadPriority: 0
   0.193 Desktop composition is active.
   0.193 Graphics settings preset: medium-with-low-vram
   0.194   Dedicated video memory size 96 MB
   0.194 Graphics options: [Graphics quality: normal] [Video memory usage: high] [Light scale: 25%] [DXT: low-quality]
   0.195                   [Max load threads: 32] [Max texture size: 4096] [Tex.Stream.: 1] [Rotation quality: low] [Color: 32bit]

[posila]

Thanks for the report.

The fullscreen flicker is huge parts of sprite atlases being drawn in between terrain and items on belts. The most visible is icon atlas, but there are also belt spritesheets visible underneath it. At the same time belts are not drawn.

Interestingly, before this happens, smaller corruption happens on frame 10, where belts are not being rendered for 2/3 of the screen and some junk is visible instead.

nvumdshimx.dll 10.18.13.5560 <--- you have also nVidia GPU, but play on Intel instead? Also the nVidia driver version is ancient, so I am going to guess the Intel driver is old as well. Is there any chance you could update them both?

Also I am confused about Windows version you might have ... you trimmed that from the log, so based on d3dcompiler_43.dll, DXGI 1.2+, I am guessing you have WIndows 7, but there shouldn't be BGR 565 support on DirectX on Windows 7 so is it OG Windows 8?

[eradicator]

Windows 8.0.I even had to downgrade the intel drivers once because the "latest version" had a serious bug for half a year, and then i stopped checking. I could try updating both nvidia+intel i guess, but i'll need time to run a full backup before. The laptop has had some cooling issues under heavy load lately, and not using the nvidia seems to help, which is why i've been using the built-in graphics.

In the meantime i've also seen the belts "shaking" sideways, i.e. being drawn 1 pixel to far to left, then to right. But it only happend once so i don't have a video.

[posila]

The laptop has had some cooling issues under heavy load lately, and not using the nvidia seems to help, which is why i've been using the built-in graphics.

Opening it up cleaning fans and heat sinks would help with cooling issues, I think. If it hasn't been cleaned its entire lifetime, it is probably all clogged up with dust.

Windows 8.0

Respect.

I could try updating both nvidia+intel i guess, but i'll need time to run a full backup before.

Maybe wait until next release, I have made some tiny changes that might help alleviate this issue, so I'd be interested to know if it actually helps or not.

[eradicator]

Nope, it's been semi-regularly vacuum cleaned. Though the construction allows no easy access to the fans directly. I suspect the heat-conductive paste has reached it's end-of-life after all the years. I bought some a few month ago but haven't found the time to apply it as it'll require deconstructing the whole laptop.

Also duh..yea, 8.0. I remember not liking *something* about 8.1, but not what. I really hate interface changing updates. Btw, the latest available intel driver is 15.33.45.4653 (2017-05-23). Not sure if you want to support outdated configurations like this.

[eradicator]

Maybe wait until next release, I have made some tiny changes that might help alleviate this issue, so I'd be interested to know if it actually helps or not.

Sure, i'll wait a bit then. While i don't know the *exact* conditions it does regularly reproduce itself ^^.

[posila]

Btw, the latest available intel driver is 15.33.45.4653 (2017-05-23). Not sure if you want to support outdated configurations like this.

Well, I'd prefer not to, but given expetations associated with 2D games I don't have much of a choice :/. However, if the issue happens only on small number of devices, I might not be able to solve it.

[eradicator]

Hehe, yea. Associations. I'm certainly pretty happy that i can continue to run my medium factory on the IGP and still have 60FPS! Certain other indie games don't even get there with the nvidia...

Luckily this isn't gamebreaking as it can be worked around by literally "walking away", so nobody'll blame you if can't find the cause :). If there's anything i can help (verbose logging?) just say the word.

[eradicator]

Maybe wait until next release, I have made some tiny changes that might help alleviate this issue, so I'd be interested to know if it actually helps or not.

Would that be 0.17.42 from today, or the "next" next release?

[Bilka]

Would that be 0.17.42 from today, or the "next" next release?

The one from today.

[eradicator]

Sorry to keep you waiting, haven't played much this week. I did stumble about a very similar looking bug in 0.17.42, on a different map though, so i can't say for certain if this is the original bug that looks different now, or if it's a different bug.

I did find a new few things though, most importantly i now have a savegame file that reproduces the bug 100% on my machine even after restarting the game (0.17.42, your release cycle is faster than me :p):

_flickerglitch.zip

(3.13 MiB) Downloaded 199 times

Steps to reproduce:

1. load the savegame. i confirmed that it works without any mods so hopefully you can ignore them.
2. do NOT MOVE (later trials revealed this might not be as critical as i thought)
3. zoom out as far as you can
4. the bug should now become visible, though you may have to wait a while (not longer than a minute for me, see below)

There's some additional factors that seem to (un-)trigger the glitch:

1. it seems to only reproduce if i resize the factorio window to an *inner window size* of 1838x690 (same as the video, i use an autohotkey based window manager). maximizing or fullscreening (1920x1200) factorio stops the flicker, but it comes right back when i downsize the window again.
2. set autosave interval to 5 minutes. eh? yea. when "waiting for it to happen" it seems to occur more often directly after an autosave. but when i tried to increase autosave frequency to 1min it stopped.
3. later i noticed it also happens while the menu is open and the game therefore paused.
4. it's also seems critical that the game runs *above* speed=1. the savegame is set to speed=10. though this makes recording a video of it much more difficult as my recorder can't keep up anymore. (also encoding to VP9 failed, so you only get a h264 video attachment)
5. the flicker is now *below* other entities while it was above them last time. the content of the flicker is also different.

flicker.jpg (380.31 KiB) Viewed 14528 times

filcker.mp4

(643.76 KiB) Downloaded 218 times

[eradicator]

Reproduces easily on the original map now too.
Looks like at least 2 atlas layers at the same time.
Also it seems that the "wrong aspect ratio" of the atlas "overlay" in the post above is directly related to the window shape.

fun_with_icons.png (2.6 MiB) Viewed 14506 times

[Twinsen]

See also: 71378, 71515

[eradicator]

@Twinsen: I'm guessing the savegame posted above didn't reproduce on your system?

[Twinsen]

It didn't. It's almost surely related to something hardware/driver specific.

[posila]

I am back. Thanks for additional info.
It is more reproducible now because as of 0.17.42 the rendering doesn't use buffers from previous frame if they still have unused space in them and discards buffers on the first usage in the frame. However, the fact this made it more reproducible is interesting as it means it doesn't happen entirely randomly, but rather some specific sequence of commands sent to GPU triggers it.

Since it started to appear more consistently, more people are reporting it, so far everyone had Sandy Bridge or Ivy Bridge iGPUs.

I tried to reproduce it on Intel HD Graphics 2500 on i5 3750S (Ivy Bridge), but with no luck. However, not all mods you use are on mod portal (your eradicator* mods aren't) and I have no idea what they do or don't contain, so something might have been drawn differently and that might be reason why I can't reproduce it. Can you try to reproduce it again and share your save with entire mod pack + new log from that run + config.ini?

[eradicator]

Can you try to reproduce it again and share your save with entire mod pack + new log from that run + config.ini?

As i already said in the post, the mods can be ignored. Just to be triple sure i just now disabled all the mods and loaded the save again, and it reproduces exactly as described in the post with the savegame (still 0.17.42). Have you followed the steps exactly (windows size, autosave interval, game speed, etc)? If i skip any one of those steps it won't reproduce for me anymore.

config.ini

(21.82 KiB) Downloaded 192 times

[eradicator]

Wasted some more time on it. But it stays elusive. This time i removed all mods before testing. The general idea is to spam trees and game.print(). And after that zooming slowly in and out.

Start a new map -> use command -> zoom in/out. If nothing appears start the next map, rinse repeat. On some maps it works instantly after the command, on some maps it doesn't work at all.

Here's another savegame. If i load it and zoom in out it reproduces all the time here. Works in fullscreen too.

The command code isn't cleaned up:

Code: Select all

/c
local p = game.player
local trees = {}
local names = {}
for _,v in pairs(game.entity_prototypes) do
  names[#names+1] = v.name
  if v.type == 'tree' then
    trees[#trees+1] = v.name
    end
  end
p.character.destructible = false
  
local garbage = string.rep(string.rep('garbage',50)..'\n',50)

function spawn (name,count,position,force)
  local count = count or 10
  local pos   = p.position
  if not (name and pos) then return end

  for i=1,count do
    pcall(function() p.surface.create_entity{
      name  = trees[math.random(#trees)],
      force = force or game.forces.enemy,
      position = p.surface.find_non_colliding_position(name,pos,32*2,0.5) or pos,
      } end)
    end
  end
spawn('tree-09-red',2000)
game.print(garbage)
--[[simulate zooming out after zooming in]]
p.zoom = 2 
p.zoom = 0.4

_flickerglitch4.zip

(2.33 MiB) Downloaded 180 times

config.ini

(21.82 KiB) Downloaded 194 times

factorio-current.log

(5.86 KiB) Downloaded 177 times

[posila]

Thanks for trying to provide a reproducible scenario, I'll try your new steps out.

Could you try if you are immediately able to reproduce it with save from 71515? (I mirrored it on my dropbox https://www.dropbox.com/s/c59kavyeek8i1 ... 2.zip?dl=0). You may need to change window client area size to 1360x768. I want to test if conditions that trigger it are consistent between different HW that has this problem.

[eradicator]

I can reproduce it on that save without moving. But i have to do a fair bit of zoom-in-out trials. And the effect is very faint and fades away after a while even without moving. It appears to be "in between" two zoom levels.

...After having written that i switched to fullscreen, and there it reproduces the same aggressive permanent flickering as on my own saves.

(And while i'm at it...the two savegames i uploaded still reproduce perfectly after upgrading .42 -> .45)

(Btw, i'll be in Prague next month if you haven't fixed it until then ;p)