RISC-V Performance Dashboard
Find performance results across CPU, Memory, Database, AI/ML, and more
Phoronix Test Suite
A comprehensive, cross-platform benchmarking framework covering CPUs, memory, storage, compilers, and more.
What these numbers are? Higher score indicates better performance (single-core and multi-core). NOTE: These numbers are averaged across all tested workloads (includes CoreMark, CacheBench, NPS, Compress-7zip, SciMark2, OpenSSL and BYTE), read more by clicking "How These numbers are calculated?" How these numbers are calculated?
Vendor | Board Name | SoC | Equally-weighted Composite Index |
---|---|---|---|
StarFive | VisionFive 1 | JH7100 | 108 |
StarFive | VisionFive 2 | JH7110 | 692 |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | 2114 |
Milk-V | Milk-V Pioneer Box | SG2042 | 18828 |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | 814 |
SPEC CPU 2017
Compute-intensive benchmark suite stressing CPU, memory hierarchy, and compiler/toolchain. SPEC CPU 2017 is organized into Integer and Floating-Point workloads, each reported as either throughput (SPECrate) or single-copy speed (SPECspeed).
Note : Only SPECrate 2017 intrate is listed, SPECspeed and SPECrate floating results will be added soon.
1. SPECrate 2017 Integer
SPECrate suites run multiple concurrent copies of each benchmark.
What these numbers are?
Higher scores indicate more throughput (work per unit of time).
How these numbers are calculated?
Vendor | Board Name | SoC | Score (Single Core / 1-copy) | Score (Multi Core / N-copy) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | 0.333 | 0.493 |
StarFive | VisionFive 2 | JH7110 | 0.692 | 2.165 |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | 3.475 |
Milk-V | Milk-V Pioneer Box | SG2042 | 1.130 | 50.05 |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | 0.6005 | 3.2 |
CoreMark
Industry-standard embedded CPU benchmark measuring basic integer operations and control flow.
What these numbers are? Iterations/second — higher is better (compare with same compiler/settings). How these numbers are calculated?
Vendor | Board Name | SoC | Iterations/sec (Single Core) | Iterations/sec (Multi Core) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | 3486.388 | 6399.6867 |
StarFive | VisionFive 2 | JH7110 | 5194.735 | 20347.9709 |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | 5633.3481 | 22571.4453 |
Milk-V | Milk-V Pioneer Box | SG2042 | 8860.0259 | 33748.6670 |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | 6753.6901 | 40747.4718 |
UnixBench
Classic UNIX system benchmark aggregating various CPU and system tests into an index.
What these numbers are? Index score — higher is better. Multi index uses multiple threads. How these numbers are calculated?
Vendor | Board Name | SoC | Index (Single Core) | Index (Multi Core) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | 159.5 | 278.5 |
StarFive | VisionFive 2 | JH7110 | 286.8 | 990.7 |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | 340.6 | 1927.4 |
Milk-V | Milk-V Pioneer Box | SG2042 | 375.9 | 11979.8 |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | 332 | 1905.9 |
Geekbench
Cross-platform CPU/GPU benchmark with synthetic workloads for quick comparisons.
What these numbers are? Score — higher is better (compare same version/config). How these numbers are calculated?
Vendor | Board Name | SoC | Score (Single) | Score (Multi) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | 36 | 59 |
StarFive | VisionFive 2 | JH7110 | 69 | 207 |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | 127 | 526 |
Milk-V | Milk-V Pioneer Box | SG2042 | 149 | 1384 |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | 131 | 569 |
Sysbench
Modular benchmark for CPU, memory, I/O and databases; here focused on CPU and transactional tests.
What these numbers are? Events/sec and Transactions/sec — higher are better. How these numbers are calculated?
Vendor | Board Name | SoC | CPU Events/sec (Single Core) | CPU Events/sec (Multi Core) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
HPL (Linpack)
Solves dense linear systems to measure floating-point throughput (GFLOPS).
What these numbers are? GFLOPS — higher is better; impacted by BLAS, memory bandwidth, and threading. How these numbers are calculated?
Vendor | Board Name | SoC | GFLOPS (Single) | GFLOPS (Multi) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
LLVM Test Suite
Compile and run tests to evaluate compiler performance, code-gen quality, and binary size.
What these numbers are? Compile time (s) and Binary size (MB) — lower are better (with correctness maintained). How these numbers are calculated?
Vendor | Board Name | SoC | Compile Time (s) | Binary Size (MB) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
STREAM
Measures sustainable memory bandwidth and computation rate for simple vector kernels.
What these numbers are? Bandwidth/Rate — higher are better. How these numbers are calculated?
Vendor | Board Name | SoC | Bandwidth (MB/s) | Compute Rate |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
lmbench
Micro-benchmarks for measuring OS and hardware latencies and bandwidths (memory, file, network).
What these numbers are? Latency — lower is better; Bandwidth — higher is better. How these numbers are calculated?
Vendor | Board Name | SoC | Latency (ns) | Bandwidth (MB/s) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
MLPerf Tiny
Inference benchmark suite for tiny/embedded devices running ML workloads.
What these numbers are? Inference/sec — higher is better; Accuracy (%) shows model correctness. How these numbers are calculated?
Vendor | Board Name | SoC | Inference/sec | Accuracy (%) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
MLPerf Inference
Datacenter/edge inference benchmark across standardized ML tasks and datasets.
What these numbers are? Samples/sec — higher is better; Latency (ms) — lower is better. How these numbers are calculated?
Vendor | Board Name | SoC | Samples/sec | Latency (ms) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
ONNX Runtime
High-performance inference engine for ONNX models with multiple execution providers.
What these numbers are? Ops/sec — higher is better; Memory (MB) shows runtime footprint — lower is better. How these numbers are calculated?
Vendor | Board Name | SoC | Ops/sec | Memory (MB) |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
pgbench
PostgreSQL benchmarking tool simulating TPC-B-like transactions and SQL workloads.
What these numbers are? Transactions/sec and Queries/sec — higher are better. How these numbers are calculated?
Vendor | Board Name | SoC | Transactions/sec | Queries/sec |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |
rvv-bench
RISC-V Vector (RVV) micro-benchmarks for evaluating vector operations and throughput.
What these numbers are? Vector Ops/sec / Throughput — higher are better (depends on vector length and ISA). How these numbers are calculated?
Vendor | Board Name | SoC | Vector Ops/sec | Throughput |
---|---|---|---|---|
StarFive | VisionFive 1 | JH7100 | N/A | N/A |
StarFive | VisionFive 2 | JH7110 | N/A | N/A |
Milk-V | Milk-V Jupiter | SpacemiT K1/M1 | N/A | N/A |
Milk-V | Milk-V Pioneer Box | SG2042 | N/A | N/A |
Banana Pi | Banana Pi BPI-F3 | SpacemiT K1 | N/A | N/A |