[] MP Memory Leak

Discussion in 'Multiplayer' started by Zed, Jun 7, 2018.

  1. Zed

    Zed Well-Known Member

    Visual Record @ https://www.twitch.tv/videos/270400868 (errors begin issues @ 6hrs 39 minutes)

    In the continued pursuit of being able to run TerraTech Multiplayer Indefinitely another memory leak has been spotted. Specifically, there seems to be a cascading memory usage which is not cleared between multiplayer rounds that eventually leads to TerraTech using up too much memory to continue operation, even slowing down in-game time.

    rounds were 1hr each with the large map setting. Terratech was running 1920x1080 windowed @ fantastic.

    The first 5 or 6 rounds (hours) went well as memory continued to expand, however in round 7 the program would become unresponsive (micropauses with windows blue processing circle) but would not give the option to close the program nor crash.

    Memory usage progressed from 3.0 used / peak with 5 committed to 6 peak / used with 10 committed by the time the program was unable to complete the 7th round despite being 8 hours elapsed time.


    During the reloading screen of auto-rematch (between rounds) have unity perform a memory cleanup / streamline bringing the usage down to 4gb / 7gb committed (which would adhere to TerraTech's suggested computer memory of 8gb)

    Log (12mb uncompressed) and dxdiag attached below.

    Attached Files:

    Last edited: Jun 7, 2018
  2. Zed

    Zed Well-Known Member

    2nd attempt with the newest Geforce Drivers had a jittery Start, resulting in a crash as the first round attempted to roll over into its 1st automatic rematch.

    Attached Files:

  3. Zed

    Zed Well-Known Member

    Another crash occurred during the testing with the latest GeForce Drivers; (DxDiag should still be valid [811b]

    Attached Files:

  4. Zed

    Zed Well-Known Member

    Video Log : https://www.twitch.tv/videos/270551557 (This video also covers the two crash folders above)

    Similar to the original post, this time with an updated GPU driver as noted in the first crash. Same conditions of 1hr long rounds with large map crates on (though they will not spawn if host does not select a tech.)

    Similar result; slowdown began mid-way through the 5th hour long round (Time index 9hrs 10 minutes into the video) where ingame would lose seconds compared to the Stopwatch app as the memory usage would not clear coupled with the falling GPU usage which, when dropped to zero would prompt the loss of seconds in the match as compared to "real" time.

    Attached below is the log dealing with the final 5 rounds shown in the video.

    Attached Files:

  5. Zed

    Zed Well-Known Member

    @kae Along with the information above detailing the newest memory leak (0811 unstable) I've started a third round of testing which I will leave run.

    In this third round of testing I've changed a GPU setting from "optimal power" to "max performance" as depicted below to see if it will improve the outcome of the test. With any luck the testing will still be running most of working friday GMT.

  6. Zed

    Zed Well-Known Member

    https://www.twitch.tv/videos/270734119 Is the video log for the third Multiplayer Test after applying the above mentioned GPU setting change..

    Overall this test was more successful I feel, same general settings of 1hr matches, large map, crates, running @ fantastic 1920x1080 windowed mode. For most of the test the Machine merely ran while the host was AFK with peak memory usage topping 8gm with a commit of 11.5gb.

    That said, upon my return it was noticeably slow and there were times when GPU usage dropped to zero which would cause in-round time to slow to a crawl for minutes at a time only for the GPU usage to once again "perk up" permitting in-round time to flow normally for around 30-45 seconds before once again slowing to a crawl.

    To me, this would seem to point to an insufficiently aggressive clearing / caching mid-round, along with the above-noted insufficient clearing in-between rounds. The round did indeed complete however a bug reporter (first time) appeared in-between rounds.

    Output Log after 17 hours is roughly 31mb and is included as a .zip archive below.

    Attached Files:

    Aardvark123 likes this.
  7. WhitePaw2002

    WhitePaw2002 Modding Husky



    oh wait, that move has already been used...
    Zed likes this.
  8. kae


    Hey guys, I'll take a look at this, thanks for the report!
    Zed likes this.

Share This Page