Many Minecraft Server Software claim to be more performant, customizable or stable. But, who really is the best of them all? I tested all the popular Minecraft server software on chunk generation, player counts, etc.
Table of Contents
Specifications of the Server:
Server 1: Orange Pi5
- CPU: RK3588 chipset (with 4xA76/2.4GHz cores and 4xA55/1.8GHz cores)
- Disk: SD Card of 160 MB/s
- RAM: 16GB LPDDR4
- Internet Speed: 1 Gbps download, 500 Mbps upload
Why the Orange Pi5?
The Orange Pi5 isn’t a powerful machine, making it an ideal comparison to Free Minecraft Server hosts.
Server 2: Custom Build Server for Minecraft
- CPU: 13th Gen Intel® Core™ i5-13400F
- Disk: Crucial P3 500GB (CT500P3SSD8)
- RAM: 32 GB DDR4 (unsure of clockspeed)
- Internet Speed: 1 GBps download, 500 Mbps upload
Minecraft Server
Your results may vary based on hardware and setup. These benchmarks are a reference, not hard science. Plugins
Spigot
Founded in 2012, Spigot was one of the leading and biggest names in the Minecraft server community. Despite its loss in popularity, it still runs around 21.000 servers, making it the second most-used software according to bStats.
Chunk Generation
On the Orange Pi 5, Spigot averaged 13 chunks per second and finished in 0:05:07, being the slowest. CPU usage sat around 250%, with RAM at 1.5 GB.
Server 2 had double the performance with an average of 26.72 CPS, finishing after 0:02:36. The resource management was fairly similar to the Orange Pi 5.
Player Performance
When there are 50 players online running randomly, the server has a difficult time. The CPU jumps from around 10% to 60–150%. The RAM stays stable, only adding 100 MB.
PaperMC
PaperMC is currently the most used Minecraft Server Software. It enhances performance, customizability and stability compared to Spigot. But, is the claim really true?
Chunk Generation
On the Pi, Chunk generation is pretty good with an average of 18.85 CPS. It was done generating after 0:03:30 with around 260% CPU and 1.7GB of RAM usage.
The second server performs even better by utilizing more CPU power. 500% CPU and around 1.7GB of RAM, causes an average of 99.75 CPS, finishing after 36s.
Player Performance
The player performance is pretty stable on PaperMC. The CPU goes from 10% to 100% and stays around that number. The RAM only adds 100 MB.
Purpur
Purpur is a fork of PaperMC aiming to add more configurability, thus creating more unique gameplay. It doesn’t improve performance, so I expect the same performance as PaperMC.
Chunk Generation
As expected, Purpur performs identical to PaperMC. A chunk generation speed of 19.61 CPS, finishing after 0:03:24 with exactly the same usage of system resources.
Comparison: PaperMC vs. Purpur.
The second server has exactly the same performance as PaperMC.
Player Performance
The number of players mostly taxed the CPU, jumping from 15% to 125%. The RAM increased by 20 MB.
A note about Purpur
Don’t get me wrong, Purpur with its customizability is awesome. But, it’s totally overkill for most servers. It also has a lot of weird configuration options. Who’s going to change the amount of bees that can hide in a beehive? Or changing the maximum growth height of kelp?
LeafMC
LeafMC is a new fork of Purpur that aims for better performance, a vanilla-like experience and stability. They even have their own benchmarks, so we’ll see if they are right.
Chunk Generation
The generation speed is 20.58 CPS, with a finish time of 0:03:14, making it the fastest of all the software. Their resource management was also superior with 250% CPU and 1.3GB of RAM. All these statistics show that LeafMC is 7.62% faster compared to their competitors.
It wouldn’t come as a surprise that LeafMC performs great on the second server. With the same utilization of server resources, it has an average of 105.52 CPS, finishing after 34s. I do want to note that the difference between PaperMC and LeafMC is smaller, 5.5% compared with to previous 7.62% on the Orange Pi 5.
Player Performance
LeafMC’s player performance is a little worse compared to PaperMC/Purpur. The CPU is stable around 100%, but the RAM jumps 200 MB higher.
A note about LeafMC
LeafMC is fairly new, so they don’t support a lot of old versions and there could be a lot of unexpected bugs. It is also uncertain whether development will continue, which is common with new software. You should be aware that LeafMC could be no longer supporting newer versions/fixing bugs if there isn’t enough support.
Conclusion
For maximum performance, customizability and a little risk, go with LeafMC.
If you want stability, decent performance and fast updates, go with PaperMC.
For the Spigot users, just switch. There’s no point in using software that lacks the modern functions PaperMC,… has.
Purpur is still a great option for hosts who want extra customizability and stable software support. But, I do think it’s not suitable for everyone.
Mods
Forge
Forge has been one of the most popular ways for running mods on servers. Even though there are many successors, does it perform well?
Chunk Generation
Chunk generation is fast with an average CPS of 21.62, finishing after 2:53. It does use a whole lot more resources compared to others with the use of 400% of the CPU and 1.7GB of RAM.
Weirdly enough, the second server doesn’t perform better than the plugin software. With an average of around 70, it lags behind modern plugin software. It finished after 52s.
Player Performance
Player performance on Forge is great. From 20% to 80% CPU utilization and around 100 MB of extra RAM.
A note about Forge
Forge is better at handling large modpacks, ensuring smoother gameplay. Do note that their updates take a longer time due to their extensive modding platform. If you don’t directly want the most recent version, Forge is a better option. Read more about it here.
Fabric
Fabric aims to be lightweight and a performant mod loader. It’s been in the game for several years, so did it develop well?
Chunk Generation
Fabric offers fast chunk generation on the Orange Pi 5 with an average of 24.46 CPS. It finished after 0:02:42. It did use a whole lot more resources; 400% of the CPU, 2.3GB of RAM.
Fabric’s performance on the second server isn’t great. An average of 67.21 CPS, finishing after 53s.
Player Performance
Weirdly enough, I couldn’t bot the server.
A note about Fabric
Fabric is more lightweight than Forge, but it comes at the cost of less possibilities for mod developers. Lots of mods work often worse together and offer less overhauling experiences.
NeoForge
NeoForge was created due to internal disagreements about the Forge project. Therefore, the entire Forge team, except one, forked Forge and made NeoForge.
Chunk Generation
NeoForge performs better than Forge with an average of 22.54 CPS, ending after 2:44 generating. The resource management is great too with a 350% usage of the CPU and 1.3GB of RAM.
On the second server, NeoForge performed the worst, finishing after 1:16. The instability of the CPU was also noticeable, jumping from 100 to 700% every second.
Player Performance
Weirdly enough, I couldn’t bot the server.
Quilt
Quilt is made by and for mod developers. They offer more options to build your mod in an easier way. For the hosters, it automatically downloads essential dependencies such as the Fabric API.
Chunk Generation
Quilt is the slowest of all the modded server software. With an average of 22.23 CPS and finishing after 0:02:57.
Here’s the weird part. On the second server, Quilt performed the best at generating chunks with an average of 77.39 finishing after 48s.
Player Performance
Weirdly enough, I couldn’t bot the server.
A note about Quilt
Quilt is still in beta, meaning there could be lots of bugs and changes.
Conclusion
It’s not easy to choose between server software for mods. Performance-wise, it really depends on the server’s hardware. I’d recommend benchmarking the software yourself, so you can make a wise decission. The choice between the software also depends on its mod support.
Therefore, recommending the use of modding software is really difficult.
Summary
Server 1:
| CPS | Generating Time | P.P CPU | P.P RAM | Configurability* | |
| Spigot | 13 | 0:05:07 | 58-150% | +100MB | 3 |
| PaperMC | 18.85 | 0:03:30 | 100% | +100MB | 4 |
| Purpur | 19.61 | 0:03:24 | 125% | +20MB | 5 |
| LeafMC | 20.58 | 0:03:14 | 100% | +250MB | 5 |
| Forge | 21.62 | 0:02:53 | 125% | +100MB | 1,5 |
| NeoForge | 22.54 | 0:02:44 | / | / | 1,5 |
| Fabric | 24.46 | 0:02:42 | / | / | 1 |
| Quilt | 22.23 | 0:02:57 | / | / | 1 |
Server 2:
| CPS | Generating Time | |
| Spigot | 26.72 | 0:02:36 |
| PaperMC | 99.75 | 0:00:36 |
| Purpur | 99.64 | 0:00:36 |
| LeafMC | 105.52 | 0:00:34 |
| Forge | ~70 | 0:00:52 |
| NeoForge | Too little information | 0:01:16 |
| Fabric | 67.21 | 0:00:53 |
| Quilt | 77.39 | 0:00:48 |
*Rated from 1 to 5: 1 means basic, 5 means everything can be configured (even kelp).
Issues with my research
While I try to provide the best and most relevant benchmarks, it doesn’t provide an overall picture.
Minecraft servers are more than generating chunks or having a lot of players online. Other factors like entities, redstone computation or even crafting have an impact on the performance. I’ve only tested on a few parts of the server software, not all of them.
As I’ve stated before, the chances of getting different results on different hardware is high.
It’s also important to note that the RAM usage is always different on start due to weird Java behavior. Therefore, there is a lot of nuance.
Please take my results as a reference.
Leave a Reply