gpu_telemetry 0.2.0

High-level GPU telemetry background logger
Documentation

GPU Telemetry

Constantly logs basic GPU information in a background thread. Metrics are entirely about memory for now. Its easy to implement.

Backends

Different backends support different things:

Windows Only Windows & Linux
Vulkan None Heap size and usage/budget (if available)
DX usage/budget and reservation None

Other backends or operating systems are unsupported. For more support, you can use custom telemetry backends.

Usage

This crate is pretty easy to use and versatile. An example usage is shown below:

use std::time::Duration;
use gpu_telemetry::monitor::{GpuMonitor, telemetry::Telemetry};

fn monitor_gpu_stuff() {
    let monitor: GpuMonitor<Telemetry> = GpuMonitor::start(Duration::from_millis(50)).unwrap();
    // do gpu stuff...
    let _ = monitor.stop().unwrap();
}

And you can specify your own logger instead of the default Telemetry one:

use std::time::Duration;
use gpu_telemetry::monitor::GpuMonitor;

fn monitor_gpu_stuff() {
    let monitor: GpuMonitor<TheBestestLogger> = GpuMonitor::start(Duration::from_millis(50)).unwrap();
    // do gpu stuff...
    let _ = monitor.stop().unwrap();
}

Just make sure TheBestestLogger implements the GpuLogger trait (and Send + 'static) and note that GpuMonitor::stop will return that logger.

You can also specify what backend you want to use:

use std::time::Duration;
use gpu_telemetry::monitor::GpuMonitor;

fn monitor_gpu_stuff() {
    let monitor =
        GpuMonitor::<Telemetry, crate::sys::VulkanGpuQuery>::start(Duration::from_millis(50))
            .unwrap();(Duration::from_millis(50)).unwrap();
    // do gpu stuff...
    let _ = monitor.stop().unwrap();
}

You can even specify both. Between the logger and the query, GpuMonitor doesn't offer much more than a clean and universal API.