I did some testing using the integrated frametime counters inside minecraft accessible via F3 + ALT

The device specification are as follows: Core i7-8565U and Nvidia MX230, the gpu driver version is the latest up to date: 466.47, windows version is 21h1.

Testing methodology

The test was run by using a client created via the launcher based on minecraft 1.16.5 with the default java (8) and another one created in snapshot 21w20a with the default java (16.0.1). Both clients received 8 GB of allocated RAM on a system total of 16GB DDR4 running at 2400mhz CL17 running in dual channel. No applications where running in the background while running the test.

To perform the test I created a world in 1.16.5 and made a minecart railtrack to be able to standardise the logging. I let the game load all the chuncks in the track at least once after I finished building the track and then I closed the game after putting myself in the minecart in the orientation to run the benchnark and then I closed the world, I run the optimise world and made a backup of it in order to be able to re-use it for further snapshots under the same conditions. I extracted another copy and optimised it in 21w20a to convert it. At this point I started benchmarking: I waited precisely 10 seconds after being spawned in the minecarft for the fps to stabilise and I begun logging and started moving with the minecart.

Furthermore I created the same world in 21w20a with the seed (no datapack) and I made the same track and I compared the results to world in 21w20a but converted from 1.16.5, both tests on the newly generated and converted world gave the same results so they are functionally the same and I will report on the converted world for simplicity also for further snapshots in which I will use a converted from 1.16.5 world.

The settings where smooth lighting to maximum, max fps to unlimited, vsync off, render distance 16, biome blend off, particles to all, entity render distance 150 per cent, fullscreen on at 1080p, clouds on and graphics was set to fancy. Via the gamerules I locked the daylight cycle, time was set to noon, weather was set to clear and mob spawning was not allowed. No mobs or items where in the benchmark rendered area because I used carefully the kill command to get rid of everything which was spawned before I type the gamerule to disable mob spawning.

Further update in comparison to the day after I created the issue: I did the same testing also by using the iGpu (intel HUD 620) and more or less the same performance degradation in percentage is experienced like what happens with the Nvdia dGpu. I have no plans to record the iGPU runs because I use the iGPU to record the screen and so this would impact the iGPU performance run, this latter does not happen when I record the dGPU since as I said it is the iGPU to do the recording job.

For further clarification I link the private youtube video I made about the 2 of the test runs I recorded.

NOTE: the clips only contain precisely the moment the cart starts moving and they end 2 seconds after going till the end of the benchmark run. The 2 benchmarks in the video have the same amount of frames and are exactly the same lenght.)

Here is the link to the private video: [*https://youtu.be/QFO4NHPXb9M*]

I am reporting this because Open GL 3.2 and Java 16 where introduced to modernise the game engine and to give developers new development tools. I know that Mojang has an eye of regard about how the updates can impact performance and I also know that the changes and bugfixes made with each update can change the performance and sometimes make the game run less well but only on the surface because the bug or thing which was changed was acting in an unintended and wrong way beofore the fix. Therefore personally I am fine with the appearent impact if the appearent boost we had before was due to something not acting properly, however I still wanted this to be brought to the attention of the developers just to make sure everything is acting as they intented in the current snapshots.

I hope this feedback can be useful.