Video Card Bugs

A forum for chatting about in-development game features.
User avatar
Arioch
Posts: 1403
Joined: Sun Feb 01, 2015 12:56 am
Location: San Jose, California
Contact:

Re: NVIDIA GeForce 7950 - OpenGL Driver

Postby Arioch » Mon Apr 20, 2015 9:19 pm

JohnnyW00t wrote:Another note - should both of the A's in STARS and SHADOW have the same dot in the middle?


No, that's intentional.

preston87
Posts: 4
Joined: Sun Apr 19, 2015 5:05 am

Re: Video Card Bugs

Postby preston87 » Fri Apr 24, 2015 5:15 pm

video card NVIDIA GeForse 840M, after the defeat, the game is off, not showing any errors. Long thought each end of the stroke

Xellzul
Posts: 6
Joined: Thu Jul 09, 2015 4:23 pm
Location: Czech Republic

Re: Video Card Bugs

Postby Xellzul » Tue Jul 28, 2015 1:04 am

After some gameplay i got

Code: Select all

  glTexImage2D(GL_TEXTURE_RECTANGLE_NV, 0, GL_RGBA, w, h, 0, GL_RGBA, GL_UNSIGNED_BYTE, data), GL_OUT_OF_MEMORY.
file: ..\ui\glx_addons.cpp, line: 499
error and the game crashed, because i didn't get dump request i manually created dump from task manager so i don't know if its the same as you want or as the dump in newer version i tested later where i got crash too but failed to make dump because of full disk :/ I will try to get the dump later
https://mega.nz/#!tt4DEJoZ!8D2Dbr0bkWAN ... xYDA60u410 (275MB, 1.8GB uncompressed)

on my windows 10 10240 64bit
game version : 13327
GPU: Intel HD Graphics 3000 on intel i5 2500k

EDIT: logs are included, but there is copy https://mega.nz/#!U4gBlYjI!t0K0BM5yix7u ... fcgrKAzz1w

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Tue Jul 28, 2015 1:55 am

Xellzul wrote:After some gameplay i got

Code: Select all

  glTexImage2D(GL_TEXTURE_RECTANGLE_NV, 0, GL_RGBA, w, h, 0, GL_RGBA, GL_UNSIGNED_BYTE, data), GL_OUT_OF_MEMORY.
file: ..\ui\glx_addons.cpp, line: 499
error and the game crashed, because i didn't get dump request i manually created dump from task manager so i don't know if its the same as you want or as the dump in newer version i tested later where i got crash too but failed to make dump because of full disk :/ I will try to get the dump later
https://mega.nz/#!tt4DEJoZ!8D2Dbr0bkWAN ... xYDA60u410 (275MB, 1.8GB uncompressed)

on my windows 10 10240 64bit
game version : 13327
GPU: Intel HD Graphics 3000 on intel i5 2500k

EDIT: logs are included, but there is copy https://mega.nz/#!U4gBlYjI!t0K0BM5yix7u ... fcgrKAzz1w


Many thanks for this report Xellzul. From the stack trace, it looks like the crash happened while generating updated planet textures. Planet texture maintenance has been our main source of crashes from the start of the beta, so, this is at least a familiar category of problems.

This one does indeed show signs of looking like a memory leak -- and perhaps one linked somehow to the intel HD graphics driver. There's no reason SiS should be using more than 300-400mb worth of RAM -- a 1.8gb dump file suggests, just by size alone, that new textures are being allocated unnecessarily.

Fortunately, you didn't trigger the "crash on launch" bug that I've seen reported on a handful of other Intel machines, which suggests that that bug is either rare or fixed.

About how long were you playing before the crash?

From your logs, it looks like you may have played through a number of tactical battles before it happened -- is that correct?

Xellzul
Posts: 6
Joined: Thu Jul 09, 2015 4:23 pm
Location: Czech Republic

Re: Video Card Bugs

Postby Xellzul » Tue Jul 28, 2015 4:20 am

Yes i was through several combats, and i have no idea how long i have been playing (could be 30mins?) but you said 400MB ram is enough? :D
Starting program: 560mb RAM
Starting new game: 731
Going to Main Menu: 720
Loading old savegame : ~1200MB
Doing some stuff here but NO COMBAT: up to 1300MB
going to main menu: 1160MB
starting new game: 1174MB
//There i made DUMP but went back to testing
loading back the old savegame and doing stuff: 1330MB
going into battle: ?? dont know :/
going out of battle to star map: 884MB
focusing few revealed star systems: 920mb
going back in time and doing battle again: 970mb, up to ~1100mb
pressing autocombat button in battle: 770MB
closing battle via menu: 918MB
focusing some star systems: 1000
edit: meanwhile while i created dump file the game jumped back to 1150mb
//created second dump here

Another thing i noticed there is awful performance while focusing star system that is revealed - rotating stars.. the FPS drops alot and the main menu performance is bad also (at least animations of focusing has slow response time)
this is tested on 13333

Edit: i didnt upload the dumps because they are 2,75GB in sum, you got the first one and i got slow upload, but if you want i will upload them

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Tue Jul 28, 2015 5:01 am

Xellzul wrote:Yes i was through several combats, and i have no idea how long i have been playing (could be 30mins?) but you said 400MB ram is enough? :D
Starting program: 560mb RAM
...
focusing some star systems: 1000
edit: meanwhile while i created dump file the game jumped back to 1150mb
//created second dump here


Looks like I need to get my hands on an HD Graphics machine, and see how much of this I can replicate. The numbers might almost make sense -- I am using a fair amount of video memory (maybe as much as 400mb+, now that I'm checking) along with up to 200mb in system memory -- so assuming the driver is pooling both together in it's memory reports, a footprint around 600mb wouldn't be too shocking. You're hitting twice that pretty quickly though -- and that's the likely source of these crashes. The question I need to try and settle is whether or not the high numbers are happening because of a true memory leak (probably due to a quirk in the driver, as it's not happening on other machines), or if the HD driver is just very inefficient, and using a lot more memory than it really needs (that's not a true "bug" per say, but it would still be something I'd need to program around).

Xellzul wrote:Another thing i noticed there is awful performance while focusing star system that is revealed - rotating stars.. the FPS drops alot and the main menu performance is bad also (at least animations of focusing has slow response time)


This could certainly be explained by the very high memory use. Particularly if the system is paging some of the texture memory to disk -- each time you viewed a new high-res asset (like a new planet texture), there'd be a huge delay from the paging overhead.

The HD 3000 also has older shader language support. If what you're seeing is a constant FPS frame hit (rather than a one-time thing, as new assets appear), then we're probably looking at a case where the shader code isn't well supported by the driver.

Xellzul wrote:Edit: i didnt upload the dumps because they are 2,75GB in sum, you got the first one and i got slow upload, but if you want i will upload them


The one you already uploaded, along with the memory numbers you've just given me, should be sufficient. I'll need to do some more focused experiments myself to try and track down just what's going on -- but this makes it clear that there's a serious issue, at least in the May 27-2015 version of the Intel HD 3000 driver.

If anyone can check for these sorts of very high memory footprints on other Intel HD drivers (in particular, the HD 4000 series driver, which is a bit more common), I'd appreciate it :)

Xellzul
Posts: 6
Joined: Thu Jul 09, 2015 4:23 pm
Location: Czech Republic

Re: Video Card Bugs

Postby Xellzul » Tue Jul 28, 2015 12:34 pm

well, on my intel hd 4600 laptop (i7-4702MQ)
first time i started the game it had like 350MB and managed just by starting new games, returing to menu and some combat to get over 800MB ram usage
BTW: GPU-Z claims ~4-5W (70% load) GPU power in star map, ~12-13W(95-100% GPU load) in combat and 13W (100% load) in race selection screen claiming 100%load
ACTION | SYSTEM MEMORY | GPU MEMORY (Dynamic)
program is not running | ----MB | 136MB
started program | 370 | 403
Race selection | 402 | 436
selecting all races | 463 | 666
clicking continue | 465 | 639
starting 7 players and 85stars game | 580MB | 750MB
waiting (up to 45sec) | 580MB | 430MB
focusing star system, unfocusing it and waiting | 570MB | 344MB
exploring two more star systens 720 | 535
waiting for a while 650mb | 206mb
reatreating 680mb | 337mb
going into main menu 690mb | 410mb
waiting a minute 678mb | 286mb
ending program --- | 136mb 15% load under half GPU core clock and 0.5W power usage

starting battle 650mb | 490mb

edit:
I will do more intel HD 3000 testing when i return home but for now i switched to gt 740m on my laptop, starting the game 80mb system memory and 340mb dedicated memory of gpu up to 100mb and 552mb dedicated after some exploring (3stars)

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Tue Jul 28, 2015 6:57 pm

Xellzul wrote:well, on my intel hd 4600 laptop (i7-4702MQ)
first time i started the game it had like 350MB and managed just by starting new games, returing to menu and some combat to get over 800MB ram usage
BTW: GPU-Z claims ~4-5W (70% load) GPU power in star map, ~12-13W(95-100% GPU load) in combat and 13W (100% load) in race selection screen
waiting a minute 678mb | 286mb
ending program --- | 136mb 15% load under half GPU core clock and 0.5W power usage


Well, that's certainly suspicious. It does indeed look very much like a driver-specific memory leak. Many thanks for pulling in the GPU-Z data as well -- that's another interesting piece of evidence.

I have a hunch about where this leak might be -- different vendors have different levels of support for multithreading, and what we could be seeing here is a failure to handle the case of textures loaded in one thread, but freed in a different one.

If you download the latest 'dev' build (currently r13335), you'll find a new option in the top left-hand corner of the Options Pane, which says "single_thread_textures". Try turning that on, then restarting the game. After a restart, you should notice that anytime you click on "next turn", the cursor changes briefly to a spinning "busy" icon.

I'd be interested to know whether or not the memory use on the Intel HD machines changes when the game has started with "single_thread_textures" enabled.

Thanks for all your help :)

Xellzul wrote:I will do more intel HD 3000 testing when i return home but for now i switched to gt 740m on my laptop, starting the game 80mb system memory and 340mb dedicated memory of gpu up to 100mb and 552mb dedicated after some exploring (3stars)


I'm fairly confident you won't see these sorts of memory issues on an NVIDIA card -- I've been doing most of my development with NVIDIA drivers -- and so the game's performance there has been pretty carefully tested. I may be underestimating the max texture memory footprint though -- my guess of "400mb" could be a bit low.

Xellzul
Posts: 6
Joined: Thu Jul 09, 2015 4:23 pm
Location: Czech Republic

Re: Video Card Bugs

Postby Xellzul » Tue Jul 28, 2015 9:42 pm

Another crash on my HD3000 after hour of gameplay, gl out of memory

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Thu Jul 30, 2015 8:32 pm

Xellzul wrote:Another crash on my HD3000 after hour of gameplay, gl out of memory


Hrm, yeah. Looks like the texture multithreading thing is a dead end. There's some gossip online that suggests that there's a memory leak in some intel drivers -- one which may be related to certain texture loading strategies -- but, none of the simple patches I've seen mentioned online are working for me. Hopefully, I can get this bug figured out next time I have physical access to an Intel HD machine.

Poklok
Posts: 6
Joined: Fri Aug 07, 2015 6:03 pm

Re: Video Card Bugs

Postby Poklok » Fri Aug 07, 2015 6:13 pm

Hi,

there's a... weird graphical bug when i play the game, which i think is probably related to my laptop's graphics card - which is ATI Mobility Radeon HD 5650. Basically the square around selected ship do not render properly - instead of green, the borders look like static. I'm also getting, previously described in this thread, Star Display glitch - particulary on blue giants.

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Fri Aug 07, 2015 6:46 pm

Poklok wrote:Hi,

there's a... weird graphical bug when i play the game, which i think is probably related to my laptop's graphics card - which is ATI Mobility Radeon HD 5650. Basically the square around selected ship do not render properly - instead of green, the borders look like static. I'm also getting, previously described in this thread, Star Display glitch - particulary on blue giants.


Thanks for the report. Can you try updating your graphics drivers? This is one case where I think AMD might have put out a patch.

Poklok
Posts: 6
Joined: Fri Aug 07, 2015 6:03 pm

Re: Video Card Bugs

Postby Poklok » Fri Aug 07, 2015 10:07 pm

Worked like a charm, thank you :)

halftea
Posts: 60
Joined: Tue Feb 10, 2015 1:16 am

Re: Video Card Bugs

Postby halftea » Sun Aug 09, 2015 7:10 pm

Had a video crash with a rendering error. Likely caused by having a bunch of video intensive chrome tabs open in the background. Sending crash dump for review.

User avatar
sven
Site Admin
Posts: 1620
Joined: Sat Jan 31, 2015 10:24 pm
Location: British Columbia, Canada
Contact:

Re: Video Card Bugs

Postby sven » Tue Aug 11, 2015 2:42 pm

halftea wrote:Had a video crash with a rendering error. Likely caused by having a bunch of video intensive chrome tabs open in the background. Sending crash dump for review.


Thanks for that upload. My hunch is that this is another case of the Intel HD 4000/3000 memory leak. On a relatively new laptop like yours, it looks like the leak isn't likely to lead to crashes -- unless you're playing for very long periods of time, or you have other memory-intensive programs competing for resources (as was probably the case here).

This is a bug that certainly needs to be fixed -- but, it also looks like one that's not seriously affecting any testers right now. So, is can probably drift down the priority list a bit.


Return to “Testing”

Who is online

Users browsing this forum: No registered users and 10 guests