ircubic / lib_gpu

GPU access library

lib_gpu v0.0.1

This is a library for getting information from and overclocking (NVIDIA) GPUs. It's partially based on reverse engineering the private interfaces of the NVIDIA nvapi.dll library to allow for functionality that the public interfaces don't expose.

It has two main interfaces, one C++ based and one that is a simplified set of C functions for linking with other languages via FFI solutions.

The library is currently in a slightly rough state due to having prioritized functionality over readability and documentation. The readability and documentation will be improved in later versions.

The library is released under the terms of the MIT license.

I make no guarantees for the health of your system when using this library, and it may in fact lead to any of the following: your system catching on fire, your GPU exploding, a portal to the unknown dimensions opening and releasing hellspawn from your computer, or the end of humanity as we know it.

Use with caution.

Usage examples

Note about overclocking

Overclocking is done in delta values, meaning that when you see the value 140 below, that means +140 over the default value, not that you're setting the actual value to 140. Negative values are also allowed, and the limits are specified by the overclock profile returned by the relevant API functions.

Simplified interface

The simplified interface is available from nvidia_simple_api.h and must first be initialized before you can use the other methods:

bool success = init_simple_api();

Then you can query values:

unsigned int gpu_count = get_gpu_count();
char name[NVIDIA_SHORT_STRING_SIZE];
bool success = get_name(0, name);
struct GpuClocks clocks = get_clocks(0);
if (success && clocks.coreClock) {
  printf("GPU %s has the following core frequency: %.1f MHz", name, clocks.coreClock);
}

Or set some overclock values:

bool success = overclock(140, GPU_OVERCLOCK_SETTING_AREA_CORE) &&
 overclock(300, GPU_OVERCLOCK_SETTING_AREA_MEMORY) &&
 overclock(87.5, GPU_OVERCLOCK_SETTING_AREA_OVERVOLT);

C++ interface

The C++ interface is available from lib_gpu.h and starts with the NvidiaApi class and allows you to query values. An important thing to note is that values are only updated when you call gpu->poll(), so if you need up-to-date values, you have to remember to poll:

auto api = NvidiaApi()
auto gpu = api->getGPU(0);
gpu->poll();
auto clocks = gpu->getClocks();
std::cout << "Current core frequency: " << clocks->coreClock << "MHz" << std::endl;
auto usage = gpu->getUsage();
std::cout << "Current GPU usage: " << usage->gpuUsage << "%" << std::endl;

You can also overclock:

GpuOverclockDefinitionMap overclockMap;
overclockMap[GPU_OVERCLOCK_SETTING_AREA_CORE] = 140;
overclockMap[GPU_OVERCLOCK_SETTING_AREA_MEMORY] = 300;
overclockMap[GPU_OVERCLOCK_SETTING_AREA_OVERVOLT] = 87.5;
bool success = gpu->setOverclock(overclockMap);