clpeak vs fio
clpeak
github.comfio
github.comSide by side
| Figure | clpeak | fio |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | — | — |
| Windows | Yes | Yes |
| macOS | Yes | Yes |
| Linux | Yes | Yes |
| iOS | Yes | — |
| Android | Yes | — |
| Self-hosted | — | Yes |
| Plans published | 1 | 1 |
| Spec score · PC Benchmark Software | 8.4 (#11) | 8.4 (#12) |
| Spec score · PC benchmark tools | 8.0 (#33) | 8.4 (#17) |
| CLI or automation | Yes | Yes |
| Comparison database | No | No |
| Distribution | Releases provide platform zip files containing the CLI and app, backend variants, and a macOS app DMG; Linux is also available through Snap and Android through Google Play. | The repository identifies GPL-2.0 as the project license. |
| Operating systems | Windows; macOS; Linux; iOS; Android | Linux,Solaris,AIX,HP-UX,macOS,NetBSD,OpenBSD,Windows,FreeBSD,DragonFly |
| Platforms | The project describes its app as running on Windows, macOS, Linux and Android. | The project lists Windows installers for tagged releases and packages for Linux distributions, Solaris, and BSDs; its release notes also mention macOS. |
| Purpose | clpeak measures peak compute throughput, memory bandwidth and latency on CPUs, GPUs and NPUs. | fio simulates configurable I/O workloads for performance testing and reproducing bugs. |
| Result export formats | JSON | normal text,terse (semicolon-delimited),JSON,JSON+,bandwidth/latency logs |
| Support | The project directs users to open a GitHub issue and attach verbose run diagnostics when reporting bugs. | The maintainers direct bug reports to the fio mailing list or GitHub rather than to them by direct email. |
| Workload areas | compute throughput; memory bandwidth; matrix throughput; kernel launch latency; AI inference | Storage I/O: read,write,trim,mixed workloads; bandwidth,IOPS,and latency |
Plans
In their own words
clpeak
clpeak is free GPL-3.0 software that benchmarks peak compute throughput, memory bandwidth, and latency on CPUs, GPUs, and NPUs. It runs the same tests through available device APIs so results can be compared between machines. Coverage includes compute, matrix engines, vendor libraries, bandwidth, latency, AI workloads, and CPU-specific operations. GPU backends include CUDA, ROCm/HIP, Metal, oneAPI/SYCL, Vulkan, and OpenCL; AI runtimes include ONNX Runtime, Core ML, and LiteRT. The app detects devices, streams measurements, and keeps renameable run history that can be exported as JSON. Its CLI selects devices, backends, and test categories, describes measurements, saves JSON runs, and compares saved runs. It runs on Windows, macOS, Linux, iOS, and Android. Releases include platform archives, a macOS DMG, Snap distribution, and Google Play availability.
fio
fio is a free storage I/O workload simulator for performance testing and reproducing bugs. Its job files can define multiple threads, processes and files, with sequential, random or mixed reads and writes using buffered, direct or raw I/O. Tune block and I/O size, I/O engine, queue depth, target files or devices, and job count. Documented engines include Linux io_uring and libaio, POSIX asynchronous I/O, and synchronous read/write options. Separate engines support HTTP GET and PUT requests to WebDAV or S3 endpoints and access to Ceph RADOS and RBD. Verification options can check written data after a workload, including repeated burn-in tests. Results are available in normal text, terse, JSON and JSON+ formats, plus bandwidth and latency logs; terse output is intended for scripts, tables or graphs. fio supports Linux, Solaris, AIX, HP-UX, macOS, NetBSD, OpenBSD, Windows, FreeBSD and DragonFly. The HTTP engine supports direct I/O only at depth 1, with scaling through multiple jobs.