The Police State upholders at the Roblox Developer Forum are not happy with my findings, going so far as to remain adamant and declare that my information is somehow off-topic and not related to development discussion. I have since, in response decided to host my findings here. The original post remains below:

Greetings, developers!

I've observed and studied significant performance differences between Rōblox engine versions during recent optimization work, and I request community collaboration and investigation in order to:

1. Document measurable regression/improvements in the engine

2. Identify strategies for dealing with performance constraints introduced by recent engine updates

3. Discuss long term solutions

Before we begin, lets lay out how I will be referring to each version of the Rōblox engine.
2021E - This refers to v463 of the Rōblox engine, compiled in early 2021, thus granting it the name "2021E".
2025M - This refers to v679 of the Rōblox engine, compiled in mid 2025, thus granting it the name "2025M".
Quite simple, right?

Benchmark Concept:

• Same game (Shrine map shown below)


• Identical graphics settings (FutureIsBright lighting)

Results:

Visual Evidence:
2021E Client (Consistent 100-200 MB):

2025M Client (Same Scene, 1140 MB):

Why, in the same game, does the 2025M client take up 1000 megabytes more than the 2021E client, despite almost looking the exact same?

Additional Observations:

• Baseplate game on the 2025M client exceed 1300MB RAM, while the Shrine used <=200MB on the 2021E client. What causes this?
  

The 2021E client, in comparison to the 2025M client, includes:

Windows 7 and 8 support, as it should be

^ Potential XP and 2000 support with third party community patches and extensions

No React-Lua topbar + recent unremovable "bloat" (Party, Self View/Camera, Voice Chat, Captures, Music) (With the exception of the 2019 React-Lua PlayerList update)

No TextChatService, instead pure Lua, editable legacy chat

Less telemetry (less resource overhead)

Lack of Hyperion (Also, less telemetry and less resource overhead)

(My work) - Remote third party asset loading capabilities, inviting audio files with however length you need it, unrestricted meshes and textures

Lack of occlusion culling

^ Certain devices report performance LOSS from this update, and I have seen negligible improvements, most likely because my games are as optimized as they can be.

Some Questions:

0. Do you save your games locally on your computer, or do you prefer to keep them on Rōblox's servers?


1. Can others replicate these observations?


2. Could Long-Term-Support (LTS) branches of Rōblox Engine help preserve projects?


3. Assume that interns behind controversial updates such as the recent topbar update, desktop app, playerlist and menu were to roll back their changes, and instead of chained nests of unoptimized React-Lua code, coded it directly using pure Lua and pure Lua alone. This would in theory create UI structure and general execution expectations of what we see in the pre-2019 CoreGui packages. Would this have significant performance boosts?


4. Rōblox is infamous for randomly breaking games, regardless of how hard they try and put great investment into backwards compatibility:
   
   
   If LTS is not an option, then how can we prevent situations where our projects and masterpieces can be completely destroyed within seconds because of decisions made by corporate higherups we can't control? If we can't, what are our options, outside of just using another engine?


5. What is the consensus of developing games using self hosted & copylefted Rōblox private servers as an alternative to moving engines?


6. If using a Rōblox private server is not an option, even if self hosted, how are Rōblox developers meant to deal with the treacherous and hostile landscape of other development engines, where most if not all of which do not provide any or proper, native support for Lua, let alone Rōblox's in-house Luau programming language?

Let's approach this from an alternative perspective other than "old rōblox good because it wuz my heckin childhood", please.

Share data, direct observations, and straight forward comparisons.

If you are interested in conducting benchmarks for yourself, please head to the Rōblox Freedom Distribution so you will be able to test with the 2021E client and the 2018M client if need be.

The Police State of the Development Forum has struck my topic down once again!
I find it quite interesting how the same moderator sent me an extremely identical message earlier but didn't lock my topic or make it invisible.
I appealed the last topic and 'Eric' refused to let me know why my topic was marked as off topic.
What makes this topic not fit for Development Discussion? I even outlined how it's meant to initiate discussion in regards to the engine's decline in performance and optimization, but I don't think this is an accident, rather they don't like people pointing negative things about the engine?

Hmmm...

He "took it down" 1-2 hours before taking it down again. Interesting

This information is particularly interesting to read. After human brain processing, I can conclude that the employee was likely, in the event of non-compliance, threatened to be consumed by the carnivorous worm overlords.

There was an error in the previous message. The previous message should state that they are threatened by the company in some ambiguous fashion. There are no worm pverlords on our planet Earth. Only us humans.

Let's be honest the second they went public it went downhill

Also this feels like something that would be in a video on yet another Roblox controversy. Just taking down a post and saying it was off-topic.

They also have a motive to do this. Investors

If they see this they're pulling out and the stock's gonna crash

I mean just look at this image here Roblox is not in charge anymore just the investors and some other humanoid creatures

Also I know nothing about investing and stocks but I am guessing stocks are the investor's money