before i get confused people failing to reproduce turn off entity shadows.
after huge amount of testing i’m now fairly certain that the results i got yesterday about being able to deal with Y spikes were just noise/jvm warmup related :( .
though for no chest in front total blockEntities % gives consistent ~ -0.3% for no chest in front for silverfish while zombie gives ~+1.0% so the case at least one of the two spikes have no chest in front is solved.
with chest in front however silverfish gives ~+0.85% and zombie gives ~+1.0% but it gets polluted by noise way too easily and i don’t see a way to prevent that. also sometimes even that difference is gone and i can’t figure out why.
also chest in front does not act the same as having entity shadows on which the latter was somewhat easier and more consistent to differentiate.
i don’t think there is anything else in game left to be discovered and i’ve basically gone through all the JVM arguments and JDKs themselves. there is a chance some more targeted cpu fryer can work but eh
well it is at least worth it if you get a double pure spike i guess
i’m honestly not that disappointed since it preserves the task of dealing with zombie spawner spikes for my stronghold ai project is this what you feel when you place bets on both sides
they are unfortunately getting removed in ranked though i’m sad
either way there may still be an analysis wrote for archiving why it is unlikely to find more paths but it will probably not be written by me (it will be written by DeepSeek v4 pro)
zombie → 1 chest behind: 47.46/5.16
silverfish → 2 chests behind:
48.79/5.03
zombie → 1 chest behind, with silverfish entity:
45.69/4.56, delta = -1.77/-0.60
silverfish → 2 chests behind, with silverfish entity:
42.21/4.64, delta = -6.58/-0.39
for now i’m too sleepy honestly did not expect to find this at all this was meant to be a distraction project to waste my time i’ll write a detailed analysis tomorrow surely maybe hopefully
on my hardware it consistently cuts the matching spawner type ~4% more than the other spawner type so like probably significant enough you can remember the numbers and do it every time but it takes ~5s to get a silverfish probably also you have to not die to the silverfish so probably still only worth if you get a double spike stronghold in which case you still have to remember the numbers (i actually got one that it would work on for the sixth stronghold i tested and the spikes were perfect for me to differentiate them except i’m too stupid to remember how to do my own strategy and don’t remember the numbers myself)
also only works if spawner is the first block entity to be rendered so it won’t work on chest before spawner spike (Y spike) and also just a reminder big orange has higher priority than Y spike and that silverfish/zombie entity has to be the only entity in view cone but it has no false positive case so worst case you don’t get false information
oh also you have to turn off entity shadow and turn off hitboxes and not hover and not do chest preemptive (i’m sure you can come up the logical conclusion with this last one) you can break close chests in chest corridors to force doing this but i don’t know if it’s worth it
either way i recommend not doing this yet until further information is disclosed but the experiment people can hop on for this one
failed paths:
- external memory bandwidth burner: i was only able to get ~2% bigger increment for silverfish compared to zombie but that just so happens to negate the effect of zombie spawners normally having ~1% bigger spike and also it fries my computer too much
- mysterious JVM argument that specifically makes zombie spawners give bigger spikes: makes a significant ~6% difference but almost certainly illegal so not disclosing any further information
- another mysterious JVM argument that makes JIT much lazier to compile hot paths of the code (so you can warming up silverfish spawners and not zombie spawners to get different spikes, later i found was brought up by chloe in march) : makes a solid ~5% difference but you have to choose from a) restarting instance every time you finish a game and warmup everything again or b) have hour long instance warmup sessions and still have significant performance loss and still have to not stare at zombie spawners for long or you have to warmup again.
- turn up difficulty before loading stronghold chunks: chance of zombie spawner zombie having weapon increase from 1% to 5%. never worth it.