⚠️ AI-generated project. The code, documentation, and wiki in this repository were written largely by an AI agent. It works on the author's hardware, but it has not been independently audited. Read it before you run it, and use it at your own risk.

NativeHardwareMonitor

A native C++ Rainmeter plugin that reads real-time hardware metrics — GPU, CPU, Memory, Network, Storage, and Ping — directly from vendor APIs (NVML, NVAPI, ADLX, ADL2) and Windows APIs (WMI, PDH, Win32). No third-party monitoring tool, no kernel driver, no service — a single DLL.

NHM\overview.ini above ships in the package and opens on install — every value in it comes from this plugin, and it is one self-contained file to read or copy from. The ten per-category skins beside it carry every metric the plugin exposes, with Debug=2 set, as a reference.

Requirements

• Windows 10/11, x64 (Rainmeter and this plugin are both 64-bit)
• Rainmeter installed
• NVIDIA GPU: NVML/NVAPI ship with the driver — nothing extra
• AMD GPU: ADLX/ADL2 ship with the driver — nothing extra
• No kernel driver, no service, no elevated install

Install

1. Get NativeHardwareMonitor.dll (download a release, or build — see the wiki)
2. Copy it into %APPDATA%\Rainmeter\Plugins\
3. Restart Rainmeter
4. Reference it from any measure with Plugin=NativeHardwareMonitor

Upgrading from an earlier version

Run 2.1.8 or later. Anything older carries one of two faults at plugin unload, and 2.1.4 and 2.1.5 each fixed the one the other introduced.

Before 2.1.4: Rainmeter froze and never came back. Any unload — a skin refresh that removed the last NativeHardwareMonitor measure, !Quit, or an installer replacing the DLL — could leave Rainmeter unresponsive with the CPU idle and nothing more in the log. The unload ran AMD's ADLXTerminate() from DllMain while Windows held the loader lock, and it waits there on a driver thread that cannot finish until that lock is released. The same teardown could instead fault inside nvml.dll.

2.1.4 only: Rainmeter crashed instead. Fixing the freeze meant never joining the polling threads at unload, but nothing else stopped them either, so FreeLibrary unmapped the code they were still running — c0000005 in NativeHardwareMonitor.dll_unloaded. 2.1.5 pins the module so the unload is a no-op, and stops every poller when the last measure goes.

Close Rainmeter before replacing the DLL. Nothing can overwrite a plugin that a loaded skin still has mapped: an installer writes the skins and the @Vault copy, leaves the old DLL alone, and the plugin still running is the previous one — which, on an affected build, is also the one that breaks on the way out. See the changelog.

Quick example

[MeasureGpuTemp]
Measure=Plugin
Plugin=NativeHardwareMonitor
Category=GPU
Metric=Temperature

[MeasureCpuLoad]
Measure=Plugin
Plugin=NativeHardwareMonitor
Category=CPU
Metric=Usage

A value of -1 means the metric is unsupported by your hardware/SDK; -2 means a plugin error. Everything else is a real reading.

One measure, every metric

A measure with no Category= is a hub: read any metric off it as a section variable, instead of declaring a measure per value.

[NHM]
Measure=Plugin
Plugin=NativeHardwareMonitor

[MeterStats]
Meter=String
Text=[NHM:GPU(name)] — [NHM:GPU(temp)]C — [NHM:CPU(usage)]% — [NHM:GPU(count)] GPUs
DynamicVariables=1

One function per category — GPU, CPU, Memory, Network, Storage, Ping, Battery, System, Motherboard, Display — each taking (metric, [device], [decimals]). The metric name accepts the same aliases as Metric=. Options need DynamicVariables=1 to re-resolve; actions do not.

Numbers print with up to three decimals, trailing zeros trimmed, so whole numbers and byte counts stay whole. NumOfDecimals cannot change that — a section variable is already text by the time the meter sees it — so pass the decimals instead: [NHM:CPU(usage,,0)] rounds to a whole number (an omitted middle argument still means device 0).

The same thing from Lua, with no measure per value:

local usage = tonumber(SKIN:ReplaceVariables('[NHM:GPU("usage")]'))
local name  = SKIN:ReplaceVariables('[NHM:GPU("name")]')

package/Skins/NHM/sectionvariables.ini reads every metric through both paths side by side, and lua.ini does the same from a script.

Documentation

Full docs live in the Wiki:

• Installation — download/build, deploy the DLL, requirements
• Configuring Measures — every measure option in detail
• Metrics Reference — every Category/Metric, its unit, device indexing, per-vendor support
• How It Works — architecture, the update/polling model, GPU fallback
• Quirks & Troubleshooting — value semantics, what needs a driver, gotchas

Building

Open NativeHardwareMonitor.slnx in Visual Studio 2022 and build the x64 configuration, or build from the command line with MSBuild. See the Installation page for details.

Why no kernel driver

CPU temperature and true per-core voltage require SMU/MSR access through a kernel driver. This project deliberately ships no driver and no service — those metrics are out of scope, and anything they'd expose returns -1. Everything here is readable from user space.

License

GNU General Public License v2 or later — see LICENSE.

This is not a free choice: the Rainmeter plugin SDK header this plugin includes (RainmeterAPI.h) is GPL v2-or-later, and Rainmeter itself is GPLv2. A plugin that includes that header and links Rainmeter is a derivative work, so it must be distributed under the GPL. Any fork or derivative of this plugin inherits the same requirement.