After idling in any game for anywhere between half a minute → 2 minutes (max time observed), the clientside untracked memory will spike to 100+MB and will persist/increase at a slower but still significant rate.
I’m not confidently sure exactly when this started happening, just noticed it when my memory usage skyrocketed from mid-350’s to low-500’s without making significant changes to my game. I believed it was a personal error until I noticed it also occuring in an empty baseplate with no scripts running.
Here is a video of it happening. The memory leak occurs at around 00:55.
The strangest thing is that I don’t get it on my other games, which would eliminate the possibility of a faulty CoreScript/DefaultScript, and in DBS Test Servers (DBS Test Servers - Roblox) where I first noticed the issue, I played the game with ALL localscripts disabled and still got the memory leak.
Here is a duplicate of the game with ALL LocalScripts disabled. When I join (Players.CharacterAutoLoads is disabled by default.) I get the same memory leak issue.
Also, something I can checked it seems that none of the behaviour changes from Workspace change anything, same issue regardless, on default or not. It could be one of these can still be affecting something, but I would expect something to change when I enable all of them to their newer counterparts, but no.
When you join, it already is using a bit of memory, 5-7, but it goes up pretty quickly, my other baseplates from before don’t have this issue, they always stay at 0.
Ohh, can someone please confirm that this only happens on low quality levels? I see something odd where untracked memory jumps if you’re on low quality in the baseplate, but goes back down if you dial the quality all the way to 10.
However it does seem like when it has a instantaneous spark in memory usage, that it will still have a bit of leaked memory left, but even then, yes, having a higher quality setting does seem to decrease that number by a large amount.
Changing back to lower graphics settings seem to make that come back instantly to a higher number.
Yes, this is exactly what happens, though in some rare, rare outliers I will join and get an instant 30-40MB leak, then it will jump again to the usual 100-200MB.
Here is a recording that I sent earlier to mister @OuterspaceNemo a few hours ago; this place has zero scripts running on clientside and serverside.
I’m not sure if discourse gives me permissions to add you to a private thread, but I’m sure something can be done if you need more samples of this issue or I can post them in this thread. NEVERMIND I thought this was a separate public thread
Also my friend @PoppaPengo is getting the same issue in some of his games.
Upon debugging this further I believe this problem, or at least my reproduction of this problem, doesn’t represent a memory leak – it’s simply the case that “untracked” memory is computed completely incorrectly.
It would help to get two microprofile dumps to confirm - one before the problem happens (when untracked memory is low) and one after (when it’s high). But our computation of untracked memory, I believe, is simply broken.
The good news is that we plan to rework the dev. console memory reporting tool next year!
Pre-spike/when I first saw memory jump (since from my observations I think the Memory usage performance stat and untracked memory update at different rates, so this may be inaccurate)
These dumps sort of confirm my suspicion although they are kinda odd in a different way I’d also recommend checking if the memory consumption that is reported by the Task Manager changes at all. If it doesn’t then it’s definitely not a leak and just a bug in “untracked memory” calculation.
I’d also recommend checking if the memory consumption that is reported by the Task Manager changes at all. If it doesn’t then it’s definitely not a leak and just a bug in “untracked memory” calculation.
