Minecraft CPU Benchmarks: More Cache More Better!
So when I first wrote about AMD's tremendous improvements to their OpenGL driver stack, I couldn't help but think it would be interesting to see how Minecraft scales across various CPUs too... So I tested it!
A little backstory of why I think this is an interesting read - Minecraft is, let's say, interestingly written game, by that I mean it has quite interesting bottlenecks it runs into and CPU/GPU features it likes - which will become apparent with the actual numbers.
Tested Minecraft Versions and Methodology
I tested 5 different flavors of Minecraft to try to hit as many points of interest as possible, it isn't all perfect, on latest versions or with all the popular mods, but I tried!
- Minecraft 1.19.2 "Vanilla" (this is the base game with no mods)
- Minecraft 1.19.2 with the Sodium mod (this is a performance optimization mod made by CaffeineMC)
- Minecraft 1.19.2 with the Sodium mod and Sildur's Vibrant Shaders (shaders are "mods" that add extra visual effects such as shadows, global illumination, reflections and in some cases even full raytracing)
- Minecraft 1.17.1 "Vanilla" (this is an older version of the base game with no mods)
- Custom 1.6.4 modpack (this was to simulate a rather extreme case of modded versions since this particular modpack runs fairly poorly regardless of hardware)
The versions were tested with 3 different save files ("worlds") to simulate various stages of progress in the game.
All FPS numbers presented are an average of 3 runs (for a grand total of some 220+ test passes).
World and Test Pass Details
"New World"
This is to simulate an early game save file with no buildings, it consists of a surface run that transitions into a cave and ends in an abandoned mineshaft. This pass was used for all tested 1.19.2 versions.
"Late-Game Base"
This is to simulate a late-game unmodded experience in a base with lots of chests, mob farms and item sorting systems around. This pass was used in 1.17.1 as upgrading the world to 1.19.2 would ruin the mob farms.
Modded "Mid-Game Base"
This is to simulate, in my opinion, an average mid-game base in a heavy technical modpack - lots of automation everywhere, machines, power generation, storage systems, etc. but not quite end-game.
Software Configuration
All systems used their own clean Windows/Linux install to eliminate
potential issues with using the same OS install on different platforms.
All in-game settings linked at the end of the post.
Operating System and Software Versions
- Operating System: Microsoft Windows 10 Pro, version 22H2
- Radeon Driver: AMD Software Adrenalin Edition 22.10.3
- GeForce Driver: GeForce Game Ready Driver 526.86
- Frametime capture: OCAT 1.6.2
- Minecraft Launcher: MultiMC 5, version 0.6.16-3431
- Java Runtime (1.19.2, 1.17.1): Java 17.0.5
- Java Runtime (Modded 1.6.4): Java 1.8.0 update 341
- Ubuntu 22.10
- GNOME 43.0 running on Wayland
- Kernel 5.19.0-23-generic
- Mesa 22.2.1 with LLVM 15.0.2 and DRM 3.47
- Frametime capture performed using GALLIUM_HUD
- Minecraft Launcher: MultiMC 5
- Java Runtime (1.19.2, 1.17.1): OpenJDK 17.0.5+8-2ubuntu1
- Java Runtime (Modded 1.6.4): OpenJDK 8u342-b07-1
Java and Minecraft Settings
- Minimum Memory: 512MiB
- Maximum Memory: 2048MiB (6144MiB for Modded 1.6.4)
- Launch arguments: -XX:+UnlockExperimentalVMOptions -XX:+UseG1GC -XX:G1NewSizePercent=20 -XX:G1ReservePercent=20 -XX:MaxGCPauseMillis=50 -XX:G1HeapRegionSize=32M
- Render Distance: 20 chunks
- Simulation Distance: 12 chunks
- Max Framerate: Unlimited
- VSync: OFF
- All other settings were left at default
- Fabric Loader 0.14.10
- Sodium 0.4.4 build 18
- Render Distance: 20 chunks
- Simulation Distance: 12 chunks
- Max Framerate: Unlimited
- VSync: OFF
- All other settings were left at default
- Fabric Loader 0.14.10
- Sodium 0.4.4 build 18
- Iris 1.4.0
- Sildur's Vibrant Shaders v1.41 High
- Render Distance: 20 chunks
- Simulation Distance: 12 chunks
- Max Framerate: Unlimited
- VSync: OFF
- All other settings were left at default
- Render Distance: 20 chunks
- Max Framerate: Unlimited
- VSync: OFF
- All other settings were left at default
- Forge 9.11.1.965
- Full list of mods can be found at the end of the post
- Render Distance: Far
- Performance: Max FPS
- VSync: OFF
- Advanced OpenGL: OFF
- All other settings were left at default
Hardware Configuration
As I don't really have a big collection of CPUs - I am not a reviewer - I tested whatever I had laying around, again, not an ideal case (I would have liked to add recent Intel CPUs too) but it does show some interesting data regardless.
All systems were tested both with an AMD Radeon RX 6600 and NVIDIA GeForce GTX 1660 graphics cards.
All systems were tested at 1440p. Shaders were also tested at 480p due to severe GPU bottlenecks.
System Specs
- Intel Core i5 4460
- Asus Z97-P Motherboard
- 16GB Dual-Channel Dual-Rank DDR3-1600 CL9 Memory
- AMD FX-8350
- ASRock 970 Extreme3 Motherboard
- 16GB Dual-Channel Dual-Rank DDR3-1600 CL9 Memory
- AMD Ryzen 5 2600
- Asus PRIME B450 PLUS Motherboard
- 64GB Dual-Channel Dual-Rank DDR4-3200 CL22 ECC Memory (no I seriously don't own smaller capacity DDR4)
- AMD Ryzen Threadripper 3960X with only 1 CCD and Dual-Channel enabled
- GIGABYTE TRX40 AORUS XTREME Motherboard
- 64GB Dual-Channel Dual-Rank DDR4-3200 CL22 ECC Memory
- AMD Ryzen 7 5800X3D
- Asus PRIME B450 PLUS Motherboard
- 64GB Dual-Channel Dual-Rank DDR4-3200 CL22 ECC Memory
Minecraft 1.19.2 "New World" Performance
Without further ado, let's dive into the results, starting with the unmodded "vanilla" 1.19.2 game.
Minecraft 1.19.2 "New World" - 1440p
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
AMD Radeon RX 6600 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 | |||
NVIDIA GeForce GTX 1660 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 |
Let's start from the bottom - the FX-8350 is actually keeping pace with
the i5 4460 which is generally a much faster CPU in games, it even manages
better 1% and 0.1% Lows (higher Lows means the game is less choppy/stuttery).
While Minecraft may seem like a single-threaded game - and for all intents
and purposes it is a single-threaded game given just how much it relies on
the performance of one individual thread - it definitely does like to have
more cores around, especially for tasks like loading new parts of the world,
which is where the FX can pull ahead.
Of course that all seems like a slideshow when compared to the newer Ryzen
CPUs, while the Zen+ based Ryzen 5 2600 is not an overly crazy increase
over the FX or Core i5, the simulated Zen 2 6-Core certainly is! Managing
an insane 80% FPS increase over the previous generation - this can largely
Ryzen 7 5800X3D, with its massive 96MB L3 cache, you may think it isn't
pulling as far ahead as expected, and indeed it isn't since the Radeon
RX 6600 used for testing is not enough to unleash the true power of this
CPU - yes! an RX 6600 in Minecraft! And same holds for the GTX 1660.
More on that later though, for now, notice the 1% Low FPS for the 5800X3D!
Over a 2.5X increase on the simulated Ryzen 5 3600X - this is where the
3D V-Cache is pulling its weight, not in average framerates, but rather
in framerate stability.
Sodium 1.19.2 "New World" Performance
Now let's move to the Sodium optimization mod.
Minecraft Sodium 1.19.2 "New World" - 1440p
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
AMD Radeon RX 6600 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 | |||
NVIDIA GeForce GTX 1660 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 |
As can be seen - the mod does what it says - FPS is much improved, in
fact it just so happens to bring the FPS up by a CPU tier, at least the
ones tested here. But again, the Ryzen 7 5800X3D is GPU limited.
For that reason I decided to re-test both Vanilla and Sodium 1.19.2 with
the 5800X3D at 480p, only with the Radeon RX 6600 since it was more of an afterthought, but here it comes...
Minecraft 1.19.2 - 1440p and 480p
GPU: AMD Radeon RX 6600
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | Average FPS 480p | |
---|---|---|---|---|
Vanilla 1.19.2 "New World" | ||||
AMD Ryzen 7 5800X3D | ||||
AMD Ryzen 5 3600X (Simulated) | ||||
AMD Ryzen 5 2600 | ||||
Intel Core i5 4460 | ||||
AMD FX-8350 | ||||
Sodium 1.19.2 "New World" | ||||
AMD Ryzen 7 5800X3D | ||||
AMD Ryzen 5 3600X (Simulated) | ||||
AMD Ryzen 5 2600 | ||||
Intel Core i5 4460 | ||||
AMD FX-8350 |
Do keep in mind that all CPUs would see a slight performance uplift at 480p, even though they are nowhere near GPU bound, so the results are not exactly comparable. Regardless, the 5800X3D is finally able to show its true potential, and it seems that Sodium likes the extra cache capacity more than the base game - seeing a 41% increase compared to Vanilla's 11% increase against 1440p - interesting!
Minecraft 1.17.1 "Late-Game Base" Performance
Now let's move onto the truly CPU-heavy tests, the 1.17.1 "Late-Game Base" and the 1.6.4 Modded "Mid-Game Base" test passes. Maybe you were wondering why this comparison is even relevant, seeing as even the FX-8350 was good for over 200fps on average in the previous graphs, but keep in mind that was a simple scene...
Minecraft 1.17.1 "Late-Game Base" - 1440p
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
AMD Radeon RX 6600 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 | |||
NVIDIA GeForce GTX 1660 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 |
Now with a big mob farm and large storage systems around, even the simulated Ryzen 5 3600X is quite stuttery, definitely playable, but it is again a far cry from the Ryzen 7 5800X3D, which is averaging 97% higher average performance and 88% to 117% higher 1% Low FPS!
Modded Minecraft 1.6.4 "Mid-Game Base" Performance
And it is a similar story with the older modded version of the game, albeit not as extreme towards the Ryzen 7 5800X3D. Though you can see the single-threaded advantage of the i5 4460 over the FX-8350 show as it is now clearly pulling ahead in all metrics.
Modded Minecraft 1.6.4 "Mid-Game Base" - 1440p
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
AMD Radeon RX 6600 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 | |||
NVIDIA GeForce GTX 1660 | |||
AMD Ryzen 7 5800X3D | |||
AMD Ryzen 5 3600X (Simulated) | |||
AMD Ryzen 5 2600 | |||
Intel Core i5 4460 | |||
AMD FX-8350 |
One thing I do want to point out here are the 0.1% Lows between the Radeon
and GeForce GPUs though, the Radeon GPU had periodic lag spikes of 100-500ms
when moving around the base and loading/unloading parts of the world. A manual
reload also took about 6-8 seconds instead of being practically instant like
on the GeForce GPU. This appears to be a driver issue with all older versions
of Minecraft, I have already reported this to AMD and hope it will eventually be fixed.
This is no longer correct as it has been fixed, see updated information below.
Dec. 1, 2022 Update!
As expected, this was very quickly fixed after I had reported it, specifically with the 22.11.2 driver.
It also turns out the issue affected any Minecraft versions older than release 1.8,
so anything up to and including 1.7.10. Not-so-coincidentally this lines up
with a large render engine rewrite that happened in Minecraft 1.8.
The large lag spikes are gone and as a result the 0.1% Low FPS is much higher,
in fact even higher than NVIDIA's. This also means average FPS got boosted
simply by there being more total frames rendered as the game is not frozen
for hundreds of milliseconds at times.
As I no longer have the whole battery of systems set up, I was only able to test the AMD Ryzen 7 5800X3D system.
Modded Minecraft 1.6.4 "Mid-Game Base" - 1440p
CPU: AMD Ryzen 7 5800X3D
RAM: Dual-Channel Dual-Rank DDR4-3200 CL22
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
Radeon RX 6600 (Driver 22.11.2) | |||
Radeon RX 6600 (Driver 22.10.3) | |||
GeForce GTX 1660 (Driver 526.86) |
As can be seen, the 0.1% Low FPS saw a huge 3.5x improvement over the 22.10.3 drivers.
The 1% Low and average FPS saw smaller gains, but that still means AMD now has a rather
nice lead on NVIDIA in this test.
In case you are curious, this is what the frametime graphs for these two
AMD driver versions look like.
Minecraft 1.19.2 "New World" Performance with Shaders
Lastly let's have a look at shader performance...
Minecraft Shaders - Sildur's Vibrant Shaders - 1440p (and 480p)
AMD Ryzen Systems: Dual-Channel Dual-Rank DDR4-3200 CL22
Intel Core and AMD FX Systems: Dual-Channel Dual-Rank DDR3-1600 CL9
0.1% Low FPS | 1% Low FPS | Average FPS | Average FPS 480p | |
---|---|---|---|---|
AMD Radeon RX 6600 | ||||
AMD Ryzen 7 5800X3D | ||||
AMD Ryzen 5 3600X (Simulated) | ||||
AMD Ryzen 5 2600 | ||||
Intel Core i5 4460 | ||||
AMD FX-8350 | ||||
NVIDIA GeForce GTX 1660 | ||||
AMD Ryzen 7 5800X3D | ||||
AMD Ryzen 5 3600X (Simulated) | ||||
AMD Ryzen 5 2600 | ||||
Intel Core i5 4460 | ||||
AMD FX-8350 |
...almost all of which is GPU bound as expected, but I knew this would happen so I tested all shader versions at both 1440p and 480p from the very start. The 480p FPS is shown as an orange extension of the 1440p bar graphs. The Ryzen 7 5800X3D is actually still GPU bound even at 480p, but bumping it down to 144p seemed like way too much of a joke... 480p is already absurd, but I don't have super powerful GPUs to test with.
Memory Ranks
I also tested the effect of quad-rank vs dual-rank, as expected there is typically a performance increase, though it appears that NVIDIA's OpenGL driver cares much less about it than AMD's.
Minecraft Performance Scaling with Memory Ranks - 1440p
CPU: AMD Ryzen 5 3600X (Simulated on Threadripper 3960X)
Dual-Channel DDR4-3200 CL22
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
Vanilla 1.19.2 "New World" | |||
Quad-Rank, RX 6600 | |||
Dual-Rank, RX 6600 | |||
Quad-Rank, GTX 1660 | |||
Dual-Rank, GTX 1660 | |||
Sodium 1.19.2 "New World" | |||
Quad-Rank, RX 6600 | |||
Dual-Rank, RX 6600 | |||
Quad-Rank, GTX 1660 | |||
Dual-Rank, GTX 1660 | |||
Vanilla 1.17.1 "Late-Game Base" | |||
Quad-Rank, RX 6600 | |||
Dual-Rank, RX 6600 | |||
Quad-Rank, GTX 1660 | |||
Dual-Rank, GTX 1660 | |||
Modded 1.6.4 "Mid-Game Base" | |||
Quad-Rank, RX 6600 | |||
Dual-Rank, RX 6600 | |||
Quad-Rank, GTX 1660 | |||
Dual-Rank, GTX 1660 |
Then there is the case of Modded 1.6.4 regressing with quad-rank, I'm not sure I can explain that, but it was reproducible... Overall I'd say stick to the typical "more ranks = better" though...
Radeon Driver Comparison
Of course I can't put out a review about Minecraft performance without mentioning AMD's OpenGL driver. So I took the testing done here and also added the old Adrenalin 22.5.1 driver - which does not include the spicy new OpenGL driver - and I also compared it against the Linux driver stack, which has seen high praise during the time of AMD's "bad" Windows driver.
Minecraft 1.19.2 Radeon Driver Comparison - 1440p
CPU: AMD Ryzen 7 5800X3D
GPU: AMD Radeon RX 6600
RAM: Dual-Channel Dual-Rank DDR4-3200 CL22
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
Vanilla 1.19.2 "New World" | |||
Ubuntu 22.10 Mesa 22.2.1 | |||
Windows 10 AMD Software 22.10.3 | |||
Windows 10 AMD Software 22.5.1 | |||
Sodium 1.19.2 "New World" | |||
Ubuntu 22.10 Mesa 22.2.1 | |||
Windows 10 AMD Software 22.10.3 | |||
Windows 10 AMD Software 22.5.1 |
As expected, the 22.5.1 driver lags far behind both the Linux "Mesa" stack and the 22.10.3 driver. But what may surprise some is that the new Windows driver actually outperforms the Linux driver! Even when GPU bound with Sodium, the Windows driver produces better 1% and 0.1% Lows.
In heavier scenes such as 1.17.1 and modded 1.6.4 the differences are quite small...
CPU-Intensive Minecraft Radeon Driver Comparison - 1440p
CPU: AMD Ryzen 7 5800X3D
GPU: AMD Radeon RX 6600
RAM: Dual-Channel Dual-Rank DDR4-3200 CL22
0.1% Low FPS | 1% Low FPS | Average FPS | |
---|---|---|---|
Vanilla 1.17.1 "Late-Game Base" | |||
Ubuntu 22.10 Mesa 22.2.1 | |||
Windows 10 AMD Software 22.10.3 | |||
Windows 10 AMD Software 22.5.1 | |||
Modded 1.6.4 "Mid-Game Base" | |||
Ubuntu 22.10 Mesa 22.2.1 | |||
Windows 10 AMD Software 22.10.3 | |||
Windows 10 AMD Software 22.5.1 |
...with the Windows driver winning in the more modern 1.17.1 version of the
game and losing in the older modded 1.6.4 version, though still producing
better 1% Lows. Interestingly enough, all drivers - old 22.5.1, new 22.10.3
and even the Linux "Mesa" stack - have awful 0.1% Lows here compared to NVIDIA.
So this may be a more fundamental issue that AMD has to address...
EDIT Dec. 1, 2022: As noted in the updated Modded 1.6.4 section,
the poor 0.1% Low FPS performance has actually been fixed with the Windows driver
22.11.2, which means that in Minecraft versions 1.17.10 and older, the new Windows
driver has a clear lead on both the Linux driver and the old Windows driver.
Lastly a look at shader performance.
Minecraft 1.19.2 Shaders Radeon Driver Comparison - 1440p and 480p
CPU: AMD Ryzen 7 5800X3D
GPU: AMD Radeon RX 6600
RAM: Dual-Channel Dual-Rank DDR4-3200 CL22
0.1% Low FPS | 1% Low FPS | Average FPS | Average FPS 480p | |
---|---|---|---|---|
Sildur's Vibrant Shaders High | ||||
Ubuntu 22.10 Mesa 22.2.1 | ||||
Windows 10 AMD Software 22.10.3 | ||||
Windows 10 AMD Software 22.5.1 |
And here we can see Mesa pulling ahead in GPU-bound averages at 1440p, but falling behind slightly in Lows and 480p. I suspect it sees a slight lead at 1440p because it may be producing slightly more optimal shader kernels? Not entirely sure.
Closing Words
I did not expect the Ryzen 7 5800X3D to have such dominant 1% and 0.1% Lows, the high average FPS is very nice too, but Minecraft genuinely never felt this smooth - and it "only" took a decade of hardware progress while the developers manage to make it perform worse every release... Anyways, as I was unable to include any recent Intel CPUs, I will not draw any CPU recommendations from this - take it simply as a case study into how Minecraft performs on various chips of random selection.
However, one purchasing recommendation I will draw from this is on the GPU front, and it will heavily depend on what Minecraft versions you play. Radeon is simply better for recent versions of the game (I suspect the cut-off is 1.8 when the renderer switched to VBOs, but I have not verified this), but if you play older versions of the game, NVIDIA may be the better choice due to not having to deal with awful lag spikes every so often. Which actually goes somewhat against my original conclusion from the Radeon 22.5.1 vs 22.7.1 review a couple months back, but that is simply because I have more data to go off of now, that original review did not include such an old version of Minecraft.
All in all, the Ryzen 7 5800X3D is very impressive, old CPUs are starting to fall behind by quite a margin even in a seemingly-light game like Minecraft and AMD did incredible things to their OpenGL driver but they still have room to improve - Fun!
And with that I thank you for taking the time to read this and hope you find it entertaining if not useful!
- Full In-Game Settings
- Modded 1.6.4 Mod List
- Graphs Download (Includes some graphs that were not shown here)