Minecraft Vulkan vs OpenGL Early Look

Methodology and Setup

I tested 3 different save files (all of which are fresh unfortunately, I don't have any recent developed save files), at 3 render distances each, on 3 different graphics cards and of course using both OpenGL and Vulkan.
I also tested a simple superflat world with 1000 Creepers to test entity rendering overhead.

The below spoiler contains all the details about the test passes, Minecraft settings, Java runtime and hardware configuration, if you are interested in that for reproducibility of the numbers, take a read.

OpenGL Results

Let's start with NVIDIA...

It is immediately clear there are some issues with frametime consistency as is apparent by the very poor 1% Low FPS performance, which will be a recurring thing with 26.2-snapshot-1 regardless of API. Overall the data is more or less as expected.

Next let's move onto AMD, they completely rewrote their OpenGL driver back in 2022, there were some ups and downs since then, but it seems those have been resolved.

The performance is very strong both in average and 1% low framerates, beating the NVIDIA GPU across the board by large margins, an awesome showing!

Lastly we have Intel, historically they've had rather poor OpenGL performance, which is still the case here compared to AMD and NVIDIA...

But the numbers overall are quite playable, though it exhibits the same frametime consistency issues as the NVIDIA GPU.

Vulkan Results

Now onto Vulkan, and immediately the results are interesting.

Switching the rendering backend to Vulkan improves performance on the RTX 5090 by as much as 31% in these tests, an impressive gain if we assume the bulk of the render engine actually stayed the same.

But then we take a look at AMD and Intel, the numbers are not looking very good.

Across the board the numbers are much lower than with OpenGL, in some cases even half the performance.

On the Intel side it isn't as bad but their starting point was also not as high as AMD's.

Entity Rendering Overhead

Entity rendering - that is drawing mobs, chests, and other things that aren't basic blocks - is a rather expensive operation as they don't follow any of the rules normal blocks do, thus a lot of the rendering optimizations are not possible.
I stuffed 1000 Creepers into a 16x16 block area on a superflat world to test whether the switch to Vulkan does anything to entity rendering performance.

In general the results are rather similar between OpenGL and Vulkan, suggesting most of the overhead is outside of the API logic. Nevertheless there was a notable gain for Intel and a lesser gain for NVIDIA, with AMD staying the same.

API Comparison Results

Summing up the performance metrics across the Forrest, Jungle and Caves tests together into a neat percentage-difference table yields us the following:

When rendering normal worlds, NVIDIA enjoys a 21-31% performance gain while both AMD and Intel fall off sharply by as much as 58%, that means in some cases you may get less than half of OpenGL performance! This was to me quite an unexpected result but it was confirmed by other people too, the Vulkan implementation as it is currently does not play nicely with all vendors.
When it comes to entity rendering, the numbers are a lot more sane, with the main notable change being Intel's 36% performance gain, but they were quite far behind both AMD and NVIDIA to begin with.

Huge render engine overhaul?

I noticed, alongside a few others, that the framerates in 26.2-snapshot-1 are in general much higher than prior versions of Minecraft, whether using OpenGL or Vulkan. This to me suggests there is actually a much bigger rendering engine overhaul going on in the background, likely to accomodate the switch to Vulkan as it is a very different API to OpenGL, but a lot of those changes are also benefiting the OpenGL backend.

This means that a lot of people checking out the snapshot will likely mistakenly attribute the gigantic performance gains to Vulkan, when in reality OpenGL also performs much better. Turns out a good rendering engine is more important than the API it is using, which is why I opened this review with "No, which was expected".
Vulkan is without a doubt the path forward, but Vulkan alone will not save a poorly implemented rendering engine, though I suppose you could argue Vulkan forces an inherently more efficient way to design the rendering engine, which is likely what is at play here.

Conclusion

The conclusion for now is simple, if you intend to play the snapshot or future versions and you have an NVIDIA graphics card, use Vulkan. But if you are using AMD or Intel graphics cards, it is probably a good idea to switch the rendering API back to OpenGL in the settings.

As for the Minecraft developers, I'm sure they are aware of this, for a first public build of the overhauled rendering engine it performs well, by which I mean I did not encounter any weird bugs that weren't already known or reported - first you need to get it stable before you can start tuning performance.
With that said it is also quite an uphill battle against years of driver-side optimizations for OpenGL and Minecraft specifically, so it will also require involvement on the driver side from the graphics hardware vendors to make the new Vulkan backend a universal performance gain across all hardware.