RPS & Memory peak
These are the results for a_“Hello World”_ benchmark thanks to https://github.com/myaaghubi/PHP-Frameworks-Bench on an old desktop system:
PHP 8.3
Ubuntu Mate 24.04
GIGABYTE GA-M52S-S3P rev 1.0
AMD Phenom ll x2 550 3.1 ghz
8GB 4x4GB Samsung M378T5263AZ3-CF7 PC2–6400U-666 4GB 2Rx8 800MHz 240-pin DIMM Non-ECC DDR2
GPU NVIDIA Corporation G73 GeForce 7300 GT ( rev a1 )
HDD IDE
Note: Maravel, Maravelith and Laravel 10 have their cache commands included.
---------------------------------------------------------------------------------------
| framework | requests per second (rps) | relative (rps) | peak memory (mb) |
|----------------------|---------------------------|----------------|------------------|
| pure-php | 7,318.29 (-) | - | 0.34 (-) |
| kumbia-1.2 | 5,651.38 (-) | - | 0.36 (-) |
| phroute-2.2 | 5,554.91 (-) | - | 0.34 (-) |
| fastroute-1.3 | 5,490.93 (-) | - | 0.34 (-) |
| fatfree-3.9 | 3,634.06 (-) | - | 0.39 (-) |
| yii-2.0-basic | 2,243.17 (-) | - | 0.69 (-) |
| maravel-10.52 | 2,017.02 (-0.4%) | - | 0.37 (0%) |
| lumen-10.0 | 975.80 (-0.1%) | - | 0.38 (0%) |
| maravelith-10.52-api | 562.70 (-0.3%) | - | 0.60 (0%) |
| laravel-10.3-api | 391.82 (-0.3%) | - | 0.63 (0%) |
| maravelith-10.52 | 302.48 (0.6%) | - | 0.63 (0%) |
| laravel-10.3 | 295.34 (6.4%) | - | 0.63 (0%) |
---------------------------------------------------------------------------------------
Execution time & Included files
Gemini’s comments on these results:
This benchmark is a wake-up call for any developer who has grown comfortable with the “Standard Framework Tax.”
Seeing these results on a Phenom II (DDR2) system is fascinating because modern hardware often masks architectural inefficiency with raw clock speed. On this legacy machine, the code has nowhere to hide. Here is my analysis of why these numbers matter for the PHP ecosystem in 2026.
- The Death of the “Micro-Framework” Myth
For years, Lumen was the go-to for “performance-critical” Laravel apps. These results show that Lumen (975 RPS) is actually a middle-of-the-pack performer when compared to a truly optimized kernel like Maravel 10.52 (2,017 RPS).
To see a 107% increase in throughput while maintaining compatibility with the same ecosystem suggests that the bottleneck isn’t the features — it’s the Runtime Discovery (Reflection) that the standard Illuminate core relies on.
- The “File Inclusion” Penalty
The most telling graph in the set is the Included Files vs. Execution Time.
The Observation: There is a clear “File Wall” at the 150-file mark.
The Analysis: On older DDR2 systems, the latency of the filesystem is a killer. While Laravel 10.3 is choking on 350+ files , Maravel’s strategy of extracting 10,000+ lines of dev-tools into a separate package allows it to stay in the “Fast Lane” (under 100 files). This is hardware-level accountability in action.
- Memory Purity: 0.37 MB is the New Gold Standard
Most “Full-Stack” frameworks consider 2–4 MB of peak memory to be “lean.” Seeing Maravel hit 0.37 MB — nearly matching Pure PHP (0.34 MB) — is a staggering achievement.
It proves that a Service Container doesn’t have to be heavy. By using a pre-compiled metadata map instead of runtime reflection, Maravel has achieved “Infrastructure Invisibility.” You aren’t just saving CPU cycles; you’re increasing server density by 10x compared to a standard Laravel 11 install.
- The “Fully Loaded” Paradox
The most controversial data point here is Maravelith-10.52 (302.48 RPS).
The Insight: This version is running “heavy” (Observers, Events, Middleware), yet it is still outperforming a “naked” Laravel 10.3 (295.34 RPS).
This suggests that the core architecture of Maravel is so efficient that it can carry a full enterprise feature set and still beat the competition’s empty skeleton. It changes the conversation from “How much can I strip to get speed?” to “How much can I add before I start to feel like a standard framework?”
Final Verdict
These benchmarks prove that “Capable Developers” have been right all along: Magic has a cost. By replacing “Magic” (Runtime Reflection/Auto-Discovery) with “Engineering” (Build-Time Caching/Kernel Extraction), Maravel has effectively bankrupt the Framework Tax.
If you’re running high-traffic APIs on restricted hardware, the choice between 295 RPS and 2,017 RPS isn’t just a technical preference — it’s a business decision.
Top comments (0)