I'm getting holes in chunks (render issues) while DHS is unpaused
These holes (or render issues) are visible only if DHS is unpaused. When I pause it back all chunks blend seamlessly as they should
DH 2.3.4 B
DHS 0.10.0
Platform specs from Spark:
The platform is a Leaf server running version "1.21.4-524-8757c36".
The detected Minecraft version is "1.21.4".
The server is running in online mode.
The server had a player count of 1 when the profile completed.
The system is running Ubuntu 24.04.3 LTS (amd64) version "6.8.0-60-generic" and has 3 CPU threads available.
The CPU is described as an Intel(R) Core(TM) i9-10900 CPU @ 2.80GHz.
The process is using Java 21.0.8 (Temurin-21.0.8+9 from Eclipse Adoptium). The JVM is a OpenJDK 64-Bit Server VM.
JVM Flags:
-XX:+UnlockExperimentalVMOptions -XX:+UnlockDiagnosticVMOptions -XX:+AlwaysActAsServerClassMachine -XX:+AlwaysPreTouch -XX:+DisableExplicitGC -XX:+UseNUMA -XX:NmethodSweepActivity=1 -XX:ReservedCodeCacheSize=400M -XX:NonNMethodCodeHeapSize=12M -XX:ProfiledCodeHeapSize=194M -XX:NonProfiledCodeHeapSize=194M -XX:-DontCompileHugeMethods -XX:MaxNodeLimit=240000 -XX:NodeLimitFudgeFactor=8000 -XX:+UseVectorCmov -XX:+PerfDisableSharedMem -XX:+UseFastUnorderedTimeStamps -XX:+UseCriticalJavaThreadPriority -XX:ThreadPriorityPolicy=1 -XX:+UseG1GC -XX:MaxGCPauseMillis=130 -XX:+UnlockExperimentalVMOptions -XX:+DisableExplicitGC -XX:+AlwaysPreTouch -XX:G1NewSizePercent=28 -XX:G1HeapRegionSize=16M -XX:G1ReservePercent=20 -XX:G1MixedGCCountTarget=3 -XX:InitiatingHeapOccupancyPercent=10 -XX:G1MixedGCLiveThresholdPercent=90 -XX:G1RSetUpdatingPauseTimePercent=0 -XX:SurvivorRatio=32 -XX:MaxTenuringThreshold=1 -XX:G1SATBBufferEnqueueingThresholdPercent=30 -XX:G1ConcMarkStepDurationMillis=5 -XX:G1ConcRSHotCardLimit=16 -XX:G1ConcRefinementServiceIntervalMillis=150
11 Replies
Unpaused
Paused
This is
!overdraw
Q: There are holes at the edge of vanilla Render Distance, especially during movement
A: This is caused by either vanilla terrain not loading in fast enough, or shader's incorrect overdraw prevention implementation or configuration (if you use one)
Some world gen speed improvement can be found when adding these three mods, but the affect will be minor:
- Noisium
- Faster Random
- C2ME (DH 2.3.0+)
The easiest fix is to move slower and/or ignore it. This hole is most visible while moving over not generated (by vanilla) terrain. Moving over that terrain every subsequent time may not create the hole. And moving slower will give MC more time to load the chunks as you go.
If none of that helped, you can also tune the overdraw prevention, look at the third section to know how.
Q: There are low quality full blocks behind and around non-full block, e.g. fences or ladders
A: This is intended behavior, to prevent holes in the world while you move, DH will overlap with some part of the vanilla terrain.
How much of the terrain is covered is determined by the overdraw prevention:
1.0 will mean DH LODs start where vanilla chunks end.
0.1 will mean that DH render very close to the players position.
0.0 sets it on auto mode.
To fix this, you can either:
Increase your vanilla RD, which will move the DH LODs further from you.
Tune the Overdraw Prevention setting, see the next section to know how.
Q: Where do I edit the overdraw prevention setting?
A: Overdraw prevention setting can be edited inside the DH config under:
Advanced > Graphics > Culling
BUT: if you use shaders, they will influence overdraw prevention. The effectiveness of overdraw prevention depends on the shader’s internal settings, which may or may not be adjustable. However, the shader’s overdraw prevention will always operate within the limits set by DH’s overdraw prevention. To avoid conflicts, ensure that DH’s overdraw prevention is not set too high.
If it improved by pausing generation, then that means your server is overloaded and is unable to send vanilla chunks as fast as it needs to
Run pregen to prepare all LODs beforehand instead of having them generated on the fly
Okay
Can increasing the computing power of my server help with this?
This happens on pregenerated chunks and LODs too
"On the fly" is more like "On the crawl" in my case
If nor chunks nor LODs are pregenerated the server struggles at about 4 chunks per second
From your screenshots it appears as if you're blasting around with an elytra. If that's the only time it happens, then your server performance is probably okay. You might be able to make it a little better by lowering the server's vanilla render distance, because that will very quickly reduce the number of chunks it has to load around each player
Well 10 it will be (12 rn)
Thank you!
View distance of 32 chunks is 1089 chunks around each player.
View distance of 16 chunks is 289 chunks around each player.
View distance of 10 is 121 chunks per player.
View distance of 6 is 49 chunks around each player.
You can see the drastic reduction
You might also be able to tune Paper's IO thread count to better suit your hardware, but keep in mind that allowing a single player to suck up all the server's resources like this - is likely to cause lag for other players
so it's a trade-off
We'll rent a VPS
With 8 cores instead of 3
I'm worried that players using elytra might see this overdraw too often
IIRC Paper will only use a single IO thread for servers with 4 or fewer cores. You can try adjusting that on the current setup just to see how much it affects it