Benchmarking Milk-V Megrez
By Akif Ejaz | Nov 14, 2025

This article will have little different format as our previous benchmarking articles. I this article we'll run Phoronix Test Suite (PTS), CoreMark®, GeekbenchUnixBench on Milk-V Megrez board and compare it with our existing RISC-V SBCs and results mentioned cloud-v.co/riscv-comparison

This article is effort of UET Lahore, ITU Lahore students who participated in RISC-V Benchmarking Hackathon at 10xEngineers from 1st Oct - 30 Oct 2025.

We’ve already discussed in detail in our previous blogs about the complete setup process for running above mentioned benchmarks, so I'll jump right into results and analysis part. 

Please review our previous blogs and cloud-v.co/riscv-comparison for complete setup process details.

1. System & Environment Details

Table below mentions the general details about Milk-V Megrez board, please read full specification details on milkv.io/megrez | docs/megrez

Vendor Board Name OS Compiler (GCC) SOC CPU Cores RAM GB RISCV Extensions
Milk-V Megrez Debian/RockOS gcc/g++ 14.2.0 ESWIN EIC7700X P550 @1.8GHz 4 32 RV64GBCH


2. Phoronix Test Suite (PTS)

Phoronix Test Suite (PTS) is an open-source, cross-platform benchmarking platform that automates everything from test installation through execution and result aggregation to help you compare system performance reliably.

2.2 Setup and Installation

Download the latest release from github/phoronix-test-suite/releases and follow the README.md for more details if you want. or clone akifejaz/riscv-benchmarks for automated setup.

2.3 Running CPU Benchmarks

There are more than 650 tests available by default but in this article, we will focus only on selected CPU benchmarks as listed below.

Benchmark/Test Name Description
CoreMark Measures the performance of core CPU operations (integer, control, memory).
CacheBench Evaluates memory hierarchy performance including L1/L2 cache and RAM throughput.
NPB (NAS) Benchmarks parallel computation performance using NASA kernels (OpenMP/MPI).
Compress-7zip Tests CPU and memory performance by compressing files using the 7-Zip algorithm.
SciMark2 Measures cryptographic operations like AES, SHA, and RSA performance.
OpenSSL Measures the performance of core CPU operations (integer, control, memory).
BYTE Unix Bench General-purpose system benchmark testing CPU, I/O, and system call performance.
FFTW Tests performance of Fast Fourier Transforms using the FFTW library.


2.4 Benchmark Results & Performance Comparison

2.4.1 BYTE Unix Benchmark


2.4.2 CacheBench

2.4.3 SciMark

2.4.4 7-Zip Compression

2.4.5 Coremark

2.4.6 OpenSSL



Note: not all results have been displays to keep the article size short, if you want to see all results, please visit the Openbenchmarking.org

3. CoreMark

CoreMark is an industry-standard benchmark from Embedded Microprocessor Benchmark Consortium (EEMBC) designed to measure a processor core’s integer performance using list processing, matrix manipulation, state-machine and CRC workloads, producing a single number for easy comparison.

Clone the CoreMark repo. from github/coremark and follow the README.md for more details. You can also clone akifejaz/riscv-benchmarks for automated setup. 

3.1 Running CPU benchmark (Single Core)

CoreMark is a small, portable CPU benchmark from EEMBC that exercises a representative mix of integer workloads list processing, simple matrix operations, a small state-machine, and CRC calculations to characterize the core performance of embedded CPUs. You tell CoreMark how many iterations to perform (call that 𝑁). The program executes the workload loop 𝑁 times and measures how long that takes, then reports a throughput metric typically shown as Iterations/Sec (the number of iterations completed per second).

To get better results we'll run the N Iterations for n times and then take mean of each Iteration. (See more details here)

The other numbers for other boards are comming from riscv-compare

4. Unix Bench

BYTE UnixBench often just called UnixBench, is a classic benchmarking suite for Unix-like systems, updated over the years from the original Byte Magazine benchmark. It runs a variety of tests (like Dhrystone, Whetstone, file copy, pipe throughput, process creation, system call overhead, and more) to generate an overall index score reflecting system performance.

In order to setup, clone the UnixBench repository from github/byte-unixbench and read the USAGE for more details. You can also clone akifejaz/riscv-benchmarks for automated setup.

4.1 Running CPU benchmark

Running the UnixBench is really easy, see complete details here

4.1.1 Single Core

4.1.2 Multi Core

5. GeekBench 6

Geekbench is widely used by hardware reviewers and manufacturers to compare device performance for tasks like gaming, image processing, and machine-learning workloads. This benchmark simulates real-life usage scenarios, resulting in scores that distinguish between single-core and multi-core performance.

Download the Geekbench 6.4.0 zip file from here and setup manually or you can also clone akifejaz/riscv-benchmarks for automated setup.

5.1 Running CPU Benchmark

To run the cpu benchmark you can use the below command. 

./geekbench6 --cpu 2>&1 | benchmark.log

After successful run, it should automatically upload the results to geekbench website and provides such links as below.

https://browser.geekbench.com/v6/cpu/14333692

6. Analysis and Comparison 

The Milk-V Megrez sits in the mid-range for overall throughput, but as a 4-core RISC-V board it delivers class-leading per-core performance, consistently outperforming VisionFive 1/2 and often matching or slightly beating other competitors in single-core CoreMark, UnixBench, and Geekbench. It shines in floating-point and memory-bound workloads, topping the SciMark2 composite, leading most FP kernels, and showing strong CacheBench write and read/modify/write bandwidth thanks to its out-of-order P550-class cores, sizeable caches, and fast DDR4. 

Its weaknesses show up in raw integer and heavily parallel workloads: CoreMark and UnixBench multi-core scale to only around 1.8× instead of the ideal 4×, and it falls behind 8–64-core boards like Jupiter, BPI-F3, and Pioneer Box in Dhrystone, 7-Zip, and OpenSSL/SHA256. Overall, Megrez is best suited as a high-IPC, FP-focused, memory-rich 4-core platform for scientific computing, development, and general Linux workloads, rather than for maximum multi-threaded compression, cryptography, or high-density server tasks.

Contributors and Co-Authors
Muhammad Waleed AkramAbdul MuizMuhammad YousafAli Tahir from UET Lahore.
and 
Muhammad Miqdad AhmadSyeda Maham BatoolMaheen Zahid from ITU Lahore



Akif Ejaz 17 November, 2025
Share this post
Archive

Benchmarking RISCV SBCs with SPEC 2017 (Single/Multi Core)
By Akif Ejaz | Aug 18, 2025