Release v0.4.2

Removed CPU telemetry (background temperature and clock speed sampling during benchmark runs). It relied on undocumented, OS-specific interfaces (raw AppleSMC keys on macOS, WMI ACPI thermal zones on Windows, /sys/class/thermal and /sys/class/hwmon on Linux) that proved unstable and inconsistent across operating systems and hardware, and live per-core clock speed was never obtainable on macOS at all through any public API. Rather than ship telemetry data that can silently jump between physical sensors mid-run or simply not exist on a given platform, this release disables the feature entirely: fossbench.net no longer accepts or displays it, and the client no longer collects it. The one-time background system-activity check (CPU/memory/process sampling before a run) is unaffected and still available via --no-system-check.

Full Changelog: v0.4.0...v0.4.2

Release v0.4.1-rc1-hotfix1

Hotfix: CPU telemetry (temperature and clock speed) recorded during benchmark runs was not working on Windows or macOS, showing empty temperature data points and a constant clock speed. Windows temperature now reads the ACPI thermal zone over WMI, and clock speed reads live per-core frequency via CallNtPowerInformation instead of a static registry value. macOS temperature now reads the SMC directly across Intel and Apple Silicon (M1-M5) Macs. macOS clock speed now reports as unavailable rather than the fixed nominal frequency it previously showed, since no live-frequency API exists on macOS.

Full Changelog: v0.4.1-rc1...v0.4.1-rc1-hotfix1

v0.4.1-rc1

FossBench v0.4.1-rc1

This release candidate adds CPU telemetry recording to benchmark runs: average clock speed and CPU temperature are sampled twice a second throughout the run and uploaded alongside your results, so results pages can show thermal/clock behavior over the course of a run.

This is a release candidate — the telemetry feature is ready for testing. Please report anything that looks off (missing sensor data, obviously wrong readings, platform-specific gaps) before this is promoted to a stable v0.4.1.

Note: an earlier iteration of this feature also recorded per-workload "phase" markers alongside the telemetry samples. That was found to distort benchmark scores under sustained load (run-to-run swings of up to 4x on the same hardware) and has been removed — telemetry sampling itself does not have this problem at the current 2 samples/sec rate.

Full Changelog: v0.4.0...v0.4.1-rc1

Release v0.4.0

FossBench v0.4.0

This release overhauls how FossBench measures and reports CPU performance. Results now include two benchmark suites:

• RAW measures the architecture-specific assembly kernels.
• REAL measures optimized portable C code, giving a better picture of typical compiled application performance.

Measured runs are now averaged instead of taking the fastest result. Integer and memory bandwidth workloads also prevent unwanted compiler
auto-vectorization, which previously allowed compiler versions and optimization choices to distort scores. Because the workload and scoring
behavior changed, v0.4.0 results should not be compared directly with results from earlier versions.

The minimum runtime for each test has been reduced from 2 seconds to 0.4 seconds, making complete benchmark runs much faster while retaining
three measured samples per workload.

Hardware reporting now includes L1, L2, and L3 cache sizes. Physical core detection has been improved on Linux, macOS, and Windows, including
systems with SMT and PowerPC or ARM topology layouts that do not report core information through /proc/cpuinfo.

The 32-bit x86 build now targets the original i386 instruction set with portable scalar kernels instead of requiring SSE2 or Pentium 4
instructions. Windows i386 builds remain compatible with Windows XP.

Uploads use the new fossbench-cpu-v2 workload format and include both RAW and REAL results, architecture details, cache sizes, and expanded
system information.

Release v0.3.0

Fixed x86/Linux builds on systems without OpenSSL development headers. Result uploads now use plain HTTP on every platform, removing the OpenSSL build and runtime dependency.

Full Changelog: v0.2.2...v0.3.0

Release v0.2.2

Fixed x86/Linux builds on systems without OpenSSL development headers. Result uploads now use plain HTTP on every platform, removing the OpenSSL build and runtime dependency.

Full Changelog: v0.2.1...v0.2.2