BenchmarksHorizon against Java Edition 26.3
Same machine, same world, same flight.
We run Horizon and vanilla Minecraft: Java Edition 26.3 one after the other on the same computer: a fresh world from the same seed, the same settings, and the same four flights over a village, a forest, the ocean and a mountain. Every figure comes from the same code for both games. Where Java does better, the tables say so.
Measured by us
The project's own machines, at render distance 12 (vanilla's default) and 32. VSync off, no frame cap, a 1280x720 window, simulation distance 12, everything else at vanilla's defaults. Each figure is the median over the runs; the frame figures are averaged over the four flights.
Measured by us
Mac mini (M4 Pro)
- CPU
- Apple M4 Pro
- GPU
- Apple M4 Pro
- Horizon
- v2026.10.22.04 · metal
- Java
- 26.3 (vanilla) · opengl
- Measured
- 2026-10-22 · 3 runs of each · 1280x720 window
On macOS this Horizon build presents its frames in step with the 144 Hz display even with VSync off, so its average frame rate here is capped at 144: compare the 1% lows and frame times instead. The Mac mini was also running other jobs during the runs (CI, simulators; 1-minute load average 20 to 56 on 14 threads), which costs both games, vanilla more at render distance 32.
| Render distance 12 | Render distance 32 |
|---|
| Horizon | Java | Horizon against Java | Horizon | Java | Horizon against Java |
|---|
| Average FPS | 144 (cap) | 123 | at least 1.2x faster | 144 (cap) | 43 | at least 3.3x faster |
|---|
| 1% low FPSthe slowest 1% of frames, as FPS | 97 | 29 | 3.3x faster | 109 | 8.0 | 14x faster |
|---|
| Frame time, 95th percentile | 7.7 ms | 17 ms | 2.2x faster | 7.7 ms | 88 ms | 11x faster |
|---|
| Start to in-worldprocess start to the spawn area loaded and drawn | 11 s | 22 s | 2.0x faster | 16 s | 66 s | 4.0x faster |
|---|
| Chunk load after a teleportthe four scenes' average | 9.4 s | 10 s | 1.1x faster | 16 s | 51 s | 3.1x faster |
|---|
| Memory while playing | 1.2 GB | 2.2 GB | 1.9x less | 1.8 GB | 2.9 GB | 1.5x less |
|---|
| Memory, peak | 1.3 GB | 2.4 GB | 1.8x less | 2.2 GB | 2.9 GB | 1.3x less |
|---|
Horizon is ahead on every figure here.
Measured by us
ThinkPad (i7-7500U, HD Graphics 620)
- CPU
- Intel Core i7-7500U CPU @ 2.70GHz
- GPU
- Intel HD Graphics 620 (KBL GT2)
- Horizon
- v2026.10.22.04 · vulkan
- Java
- 26.3 (vanilla) · opengl
- Measured
- 2026-10-22 · 2 runs of each · 1280x720 window
A 2016 laptop: 2 cores / 4 threads and Intel integrated graphics, Linux with Mesa 26.1.8 (Horizon on Vulkan, vanilla on OpenGL). It has no screen attached, so both games drew to a GPU compositor (sway on the Intel GPU) in a 1280x720 window. Horizon ran two flights per scene, vanilla two separate games of one. Render distance 32 is not shown: in 5 minutes neither game finished loading the spawn, and given 25, Horizon kept only about 900 to 1,400 chunk columns resident on this machine instead of the 3,200 a full view holds, so the two games would not have drawn the same world.
| Render distance 12 |
|---|
| Horizon | Java | Horizon against Java |
|---|
| Average FPS | 209 | 126 | 1.7x faster |
|---|
| 1% low FPSthe slowest 1% of frames, as FPS | 39 | 29 | 1.3x faster |
|---|
| Frame time, 95th percentile | 14 ms | 17 ms | 1.2x faster |
|---|
| Start to in-worldprocess start to the spawn area loaded and drawn | 51 s | 68 s | 1.3x faster |
|---|
| Chunk load after a teleportthe four scenes' average | 60 s | 47 s | 1.3x slower |
|---|
| Memory while playing | 797 MB | 2.0 GB | 2.6x less |
|---|
| Memory, peak | 986 MB | 2.1 GB | 2.2x less |
|---|
Where Java is ahead: chunk load after a teleport at 12; average FPS over the mountain at 12 (205 against 111).
Measured by us
Desktop (RTX 4080, i7-13700KF)
- CPU
- 13th Gen Intel Core i7-13700KF
- GPU
- NVIDIA GeForce RTX 4080
- Horizon
- v2026.10.22.13 · vulkan
- Java
- 26.3 (vanilla) · vulkan
- Measured
- 2026-10-23 · 3 runs of each · 1280x720 window
Linux (NixOS) with the NVIDIA 595 driver. The desktop is in daily use, so both games drew to a separate GPU compositor (a headless sway rendering on the RTX 4080) in a 1280x720 window, never on its screen: Horizon through Wayland, vanilla through Xwayland, both on Vulkan. Other jobs were running during the runs (1-minute load average 18 to 25 on 24 threads), which costs both games, chunk loading most.
| Render distance 12 | Render distance 32 |
|---|
| Horizon | Java | Horizon against Java | Horizon | Java | Horizon against Java |
|---|
| Average FPS | 2,993 | 2,600 | 1.2x faster | 672 | 849 | 1.3x slower |
|---|
| 1% low FPSthe slowest 1% of frames, as FPS | 311 | 312 | about the same | 89 | 112 | 1.3x slower |
|---|
| Frame time, 95th percentile | 0.5 ms | 0.5 ms | about the same | 3.6 ms | 3.1 ms | 1.2x slower |
|---|
| Start to in-worldprocess start to the spawn area loaded and drawn | 18 s | 30 s | 1.7x faster | 9.0 s | 59 s | 6.5x faster |
|---|
| Chunk load after a teleportthe four scenes' average | 1.8 s | 11 s | 6.0x faster | 4.7 s | 32 s | 6.8x faster |
|---|
| Memory while playing | 1.2 GB | 2.8 GB | 2.4x less | 1.6 GB | 3.3 GB | 2.1x less |
|---|
| Memory, peak | 1.4 GB | 3.0 GB | 2.2x less | 1.9 GB | 3.4 GB | 1.8x less |
|---|
Where Java is ahead: average fps at 32; 1% low fps at 32; frame time, 95th percentile at 32; average FPS over the mountain at 12 (2,824 against 2,673); average FPS over the village at 32 (881 against 686); average FPS over the forest at 32 (1,072 against 590).
Your hardware
Pick a graphics card, a processor and, if you know it, your memory. If we or at least three players have measured that pair, you see what was measured; otherwise an estimate from the nearest measured machine, as a range.
The picker needs JavaScript; the tables above and the method below work without it.
From players
Players who allow benchmark uploads in the launcher's data collection settings share the report of horizon-client --benchmark standard. Results are grouped by graphics card, processor and render distance, and a group shows only once at least 3 different installs are in it, so no one can be picked out. Only medians are published, with how the group's installs split by memory, graphics memory, drive and system in coarse classes (each class only with at least 3 installs too): no install ID, date, driver, exact memory or screen.
No groups yet, as of 2026-10-23. The page asks the telemetry server for newer ones.
No player results yet. A GPU and CPU pair shows here once at least 3 different installs have shared a benchmark with it.
Players' hardware
Every play session players with performance statistics on sent in the last 30 days, split by memory, graphics memory, drive, system and integrated or separate graphics. A class shows only with at least 3 different installs. Frame times are the sessions' average median and 99th percentile, at every render distance and setting players chose, so classes differ by more than their hardware.
Live from the telemetry server.
How the estimates work
An estimate starts from the measured machine closest to the one you picked, then scales its figures with a small table of specifications we wrote ourselves (no benchmark database):
- GPU factor: the geometric mean of the two graphics cards' ratio of FP32 shader throughput (TFLOPS) and of memory bandwidth (GB/s), as their makers state them.
- CPU factor: the ratio of the two processors' single-thread scores: top boost clock times a rough factor per chip generation (Zen 4 = 1).
- Frame rate: when the measured run split each frame into CPU and GPU time, the slower side is scaled by its factor; otherwise by the geometric mean of both. Whatever the split, the real figure lies between the measured one scaled by the smaller and by the larger factor; we widen that by 15% for the roughness of the table, and show it as the range.
- Load times are CPU work (generating, lighting and meshing chunks), scaled by the CPU factor. Memory is not scaled: it follows the world and the settings, not the speed.
- Memory size: when players' results say how much memory and graphics memory their machines have, a measured group whose players have neither yours counts as further away, and the estimate says what they had. With your memory picked, it also says when the game's peak use is a large share of it.
Estimates ignore drivers, the graphics API, thermal limits and the operating system, which can matter as much as the chip. Treat them as a rough idea, and run horizon-client --benchmark standard on your own machine for the real figure.
The specification table
| Graphics card | FP32 TFLOPS | Bandwidth, GB/s | Memory, GB |
|---|
| NVIDIA GeForce RTX 4090 | 82.6 | 1008 | 24 |
|---|
| NVIDIA GeForce RTX 4080 | 48.7 | 717 | 16 |
|---|
| NVIDIA GeForce RTX 4070 Ti | 40.1 | 504 | 12 |
|---|
| NVIDIA GeForce RTX 4070 | 29.1 | 504 | 12 |
|---|
| NVIDIA GeForce RTX 4060 Ti | 22.1 | 288 | 8 |
|---|
| NVIDIA GeForce RTX 4060 | 15.1 | 272 | 8 |
|---|
| NVIDIA GeForce RTX 3080 | 29.8 | 760 | 10 |
|---|
| NVIDIA GeForce RTX 3070 | 20.3 | 448 | 8 |
|---|
| NVIDIA GeForce RTX 3060 | 12.7 | 360 | 12 |
|---|
| NVIDIA GeForce RTX 3050 | 9.1 | 224 | 8 |
|---|
| NVIDIA GeForce RTX 2060 | 6.5 | 336 | 6 |
|---|
| NVIDIA GeForce GTX 1660 SUPER | 5.0 | 336 | 6 |
|---|
| NVIDIA GeForce GTX 1060 6GB | 4.4 | 192 | 6 |
|---|
| NVIDIA GeForce GTX 1050 Ti | 2.1 | 112 | 4 |
|---|
| AMD Radeon RX 7900 XTX | 61.4 | 960 | 24 |
|---|
| AMD Radeon RX 7800 XT | 37.3 | 624 | 16 |
|---|
| AMD Radeon RX 7600 | 21.8 | 288 | 8 |
|---|
| AMD Radeon RX 6700 XT | 13.2 | 384 | 12 |
|---|
| AMD Radeon RX 6600 | 8.9 | 224 | 8 |
|---|
| AMD Radeon RX 580 | 6.2 | 256 | 8 |
|---|
| AMD Radeon 780M (integrated) | 8.6 | 89 | shared |
|---|
| Steam Deck (AMD Van Gogh) | 1.6 | 88 | shared |
|---|
| Intel Arc A770 | 19.7 | 560 | 16 |
|---|
| Intel Iris Xe Graphics (96 EU) | 2.1 | 68 | shared |
|---|
| Intel HD Graphics 620 | 0.4 | 34 | shared |
|---|
| Apple M1 | 2.6 | 68 | shared |
|---|
| Apple M2 | 3.6 | 100 | shared |
|---|
| Apple M3 | 4.1 | 100 | shared |
|---|
| Apple M4 | 4.3 | 120 | shared |
|---|
| Apple M1 Pro (16-core GPU) | 5.2 | 200 | shared |
|---|
| Apple M4 Pro (20-core GPU) | 8.5 | 273 | shared |
|---|
| Apple M4 Max (40-core GPU) | 17.0 | 546 | shared |
|---|
| Processor | Boost, GHz | Generation factor | Single-thread score |
|---|
| Intel Core i7-7500U | 3.50 | 0.62 | 2.17 |
|---|
| Intel Core i5-7200U | 3.10 | 0.62 | 1.92 |
|---|
| Intel Core i5-8250U | 3.40 | 0.62 | 2.11 |
|---|
| Intel Core i5-9400F | 4.10 | 0.62 | 2.54 |
|---|
| Intel Core i7-9700K | 4.90 | 0.62 | 3.04 |
|---|
| Intel Core i5-10400 | 4.30 | 0.62 | 2.67 |
|---|
| Intel Core i5-1135G7 | 4.20 | 0.78 | 3.28 |
|---|
| Intel Core i5-12400 | 4.40 | 0.95 | 4.18 |
|---|
| Intel Core i5-13600K | 5.10 | 0.95 | 4.84 |
|---|
| Intel Core i7-13700K / KF | 5.40 | 0.95 | 5.13 |
|---|
| Intel Core i9-14900K | 6.00 | 0.95 | 5.70 |
|---|
| AMD Ryzen 5 3600 | 4.20 | 0.72 | 3.02 |
|---|
| AMD Ryzen 5 5600X | 4.60 | 0.88 | 4.05 |
|---|
| AMD Ryzen 7 5800X3D | 4.50 | 0.88 | 3.96 |
|---|
| AMD Ryzen 5 7600X | 5.30 | 1.00 | 5.30 |
|---|
| AMD Ryzen 7 7800X3D | 5.00 | 1.00 | 5.00 |
|---|
| AMD Ryzen 9 7950X | 5.70 | 1.00 | 5.70 |
|---|
| AMD Ryzen 7 9700X | 5.50 | 1.12 | 6.16 |
|---|
| AMD Ryzen 7 7840U / HS | 5.10 | 1.00 | 5.10 |
|---|
| Steam Deck (AMD Zen 2, 4 cores) | 3.50 | 0.72 | 2.52 |
|---|
| Apple M1 | 3.20 | 1.25 | 4.00 |
|---|
| Apple M2 | 3.50 | 1.28 | 4.48 |
|---|
| Apple M3 | 4.05 | 1.38 | 5.59 |
|---|
| Apple M4 | 4.40 | 1.50 | 6.60 |
|---|
| Apple M4 Pro | 4.50 | 1.50 | 6.75 |
|---|
How we measure
tools/perf_bench.py starts each game from a fresh world (seed 1234, creative), waits until every chunk around the spawn is loaded and drawn, then for each scene teleports, waits for the chunks, holds for 2 seconds and flies a straight 192-block path in 12 seconds. Every frame is recorded. Vanilla runs as players get it (your own client, the official launcher's Java options, Java 25), driven by a small development-only mod that only calls the game's own code; nothing of Minecraft is shipped. The full method and its caveats are in docs/benchmark.md, and every raw report is in the repository.
Measure your own computer with horizon-client --benchmark standard --render-distance 12: it writes a report to the benchmarks folder of the game folder.
Not an official Minecraft product. Not approved by or associated with Mojang or Microsoft.