Frequently Asked Questions

Answers to the most common questions about gpulse.

What GPUs does gpulse support?
gpulse supports NVIDIA (via NVML, drivers 450.0+), Apple Silicon (via Metal and IOKit, M1/M2/M3/M4), AMD (via ROCm 5.0+), and Intel (via Level Zero, Arc and Data Center GPU Max series). GPU vendor is auto-detected at startup.
Does gpulse work on Linux?
gpulse currently supports macOS (both Apple Silicon and Intel). Linux support with NVIDIA and AMD GPUs is actively in development and will be available in a future release.
How does OOM prediction work?
gpulse uses three detection algorithms: LinearTrendDetector (least-squares regression on memory usage), SpikeDetector (sudden large memory increases in a sliding window), and CompositeDetector (runs both and returns highest-confidence result). When a leak is detected, it calculates estimated time to OOM.
What data does gpulse collect?
None. gpulse monitors your local GPU hardware and does not send any data anywhere. There is no telemetry, no analytics, and no phone-home behavior. All processing happens on your machine.
Is gpulse open source?
The free tier is free to use, but the source code is not publicly available. gpulse is distributed as a compiled binary to fund continued development through Pro and Enterprise tiers.
How is gpulse different from nvidia-smi?
nvidia-smi is a snapshot tool. gpulse is a real-time monitoring dashboard with continuous updates, 7 view modes, memory leak detection with OOM prediction, multi-vendor support (Apple Silicon, AMD, Intel), 15 color themes, and process management from within the UI.
Can I monitor multiple GPUs?
Yes. gpulse automatically detects all GPUs in your system. Grid view tiles all GPUs, Compare view shows them side by side, and you can press 1-8 to jump to a specific GPU.
Can I try gpulse without a GPU?
Yes. Demo mode simulates GPUs with realistic workload data: gpulse dashboard --demo nvidia:4. This simulates 4 NVIDIA GPUs with active workloads, temperature fluctuations, and memory leak patterns.
Does gpulse affect GPU performance?
gpulse has negligible overhead. It reads metrics through vendor-provided libraries (NVML, Metal, ROCm, Level Zero). Typical usage: less than 1% CPU on a single core, ~20 MB memory, zero GPU impact.
Can I export metrics to Prometheus or Grafana?
Yes. gpulse includes a built-in Prometheus metrics endpoint at localhost:9090/metrics. This exposes all GPU metrics in Prometheus format, ready for Grafana dashboards.
Can I monitor GPUs on remote machines?
Fleet monitoring for multi-machine setups is coming in the Pro tier. It will support monitoring up to 20 machines via SSH with a unified dashboard.

What GPUs are supported?

gpulse supports four GPU vendors out of the box:

  • NVIDIA — via NVML (NVIDIA Management Library). Requires NVIDIA drivers 450.0 or later.
  • Apple Silicon — via Metal and IOKit. Works on M1, M2, M3, and M4 chips.
  • AMD — via ROCm. Requires ROCm 5.0 or later on supported Linux distributions.
  • Intel — via Level Zero. Supports Intel Arc and Data Center GPU Max series.

gpulse auto-detects your GPU vendor at startup. No configuration is needed to select the right driver interface.

Does gpulse work on Linux?

gpulse currently supports macOS (both Apple Silicon and Intel). Linux support with NVIDIA and AMD GPUs is actively in development and will be available in a future release. Sign up for the newsletter to be notified.

How does OOM prediction work?

gpulse uses three detection algorithms running in real time:

  1. LinearTrendDetector — performs least-squares linear regression on memory usage over time. If the slope is positive and R² exceeds the threshold, it flags a leak and calculates time to OOM.
  2. SpikeDetector — detects sudden large increases in memory within a sliding time window.
  3. CompositeDetector — runs both detectors and returns the highest-confidence result.

See Leak Detection for a deep dive into the algorithms.

What data does gpulse collect?

None. gpulse monitors your local GPU hardware and does not send any data anywhere. There is no telemetry, no analytics, and no phone-home behavior. All processing happens on your machine. Your GPU metrics, process names, and system information never leave your device.

Is gpulse open source?

The free tier is free to use, but the source code is not publicly available. gpulse is distributed as a compiled binary. This allows us to offer a high-quality product while maintaining the ability to fund continued development through Pro and Enterprise tiers.

How is gpulse different from nvidia-smi?

nvidia-smi is a snapshot tool that shows current state. gpulse is a real-time monitoring dashboard that:

  • Updates continuously without re-running a command
  • Provides 7 specialized view modes for different tasks
  • Runs memory leak detection algorithms with OOM time prediction
  • Supports Apple Silicon, AMD, and Intel (not just NVIDIA)
  • Includes 15 color themes with accessibility options
  • Offers process management (filter, sort, kill) from within the UI

Can I monitor multiple GPUs?

Yes. gpulse automatically detects all GPUs in your system and shows them simultaneously. Grid view tiles all GPUs, Compare view shows them side by side, and you can press 1-8 to jump to a specific GPU.

Can I try gpulse without a GPU?

Yes. Demo mode simulates GPUs with realistic workload data: 

gpulse dashboard --demo nvidia:4

This simulates 4 NVIDIA GPUs with active workloads, temperature fluctuations, and even memory leak patterns so you can explore every view mode.

Does gpulse affect GPU performance?

gpulse is designed to have negligible overhead. It reads GPU metrics through vendor-provided libraries (NVML, Metal, ROCm, Level Zero) that are specifically designed for monitoring. Typical resource usage:

  • CPU: <1% on a single core
  • Memory: ~20 MB resident
  • GPU impact: none (reads metrics only, does not allocate GPU memory)

Can I export metrics to Prometheus / Grafana?

Yes. gpulse includes a built-in Prometheus metrics endpoint that you can scrape at a configurable address (default localhost:9090/metrics). This exposes all GPU metrics in Prometheus format, ready for Grafana dashboards.

How do I report bugs or request features?

Email us at info@gpulse.ai. We read every message and typically respond within 24 hours. Include your gpulse version (gpulse --version), OS version, and GPU model when reporting bugs.

Can I monitor GPUs on remote machines?

Fleet monitoring for multi-machine setups is coming in the Pro tier. It will support monitoring up to 20 machines via SSH with a unified dashboard. Join the waitlist to be notified when it launches.