NVML
A GPU management library provided by NVIDIA for monitoring and managing the status and performance metrics of NVIDIA GPUs.
Timeline (last 90 days)
Topology-aware scheduling reads hardware topology information (such as affinity data provided by NVML and the node topology manager) to preferentially place communication-intensive tasks on resource combinations with higher interconnect bandwidth.
Graphics card detection relies on NVIDIA's NVML interface to read VRAM capacity and compute capability level, based on which the resolution limit and whether to enable half precision (fp16) or VRAM optimization strategies are determined.
GPU telemetry is typically exposed by NVML or DCGM, covering power consumption, temperature, SM clock, memory bandwidth utilization, ECC error counts, NVLink link status and more.
NVML (NVIDIA Management Library) is the low-level monitoring and management interface provided by NVIDIA, on which nvidia-smi, dcgm-exporter, and various GPU monitoring agents are built.
nvidia-ml-py is the Python binding for NVML and can read temperature, SM/memory clocks, power draw and throttle reasons at sub-second granularity within a benchmarking loop.
nvidia-smi provides instantaneous card-level readings and cannot distinguish which process or request is consuming VRAM; per-tenant attribution requires instrumentation at each service layer.
All Facts (7)
Fleet Intelligence的实时遥测能力建立在NVML和DCGM等底层监控技术之上
85%UnverifiedTopology-aware scheduling reads hardware topology information (such as affinity data provided by NVML and the node topology manager) to preferentially place communication-intensive tasks on resource combinations with higher interconnect bandwidth.
50%UnverifiedGraphics card detection relies on NVIDIA's NVML interface to read VRAM capacity and compute capability level, based on which the resolution limit and whether to enable half precision (fp16) or VRAM optimization strategies are determined.
50%UnverifiedGPU telemetry is typically exposed by NVML or DCGM, covering power consumption, temperature, SM clock, memory bandwidth utilization, ECC error counts, NVLink link status and more.
50%Unverifiednvidia-ml-py is the Python binding for NVML and can read temperature, SM/memory clocks, power draw and throttle reasons at sub-second granularity within a benchmarking loop.
50%UnverifiedNVML (NVIDIA Management Library) is the low-level monitoring and management interface provided by NVIDIA, on which nvidia-smi, dcgm-exporter, and various GPU monitoring agents are built.
50%Unverifiednvidia-smi provides instantaneous card-level readings and cannot distinguish which process or request is consuming VRAM; per-tenant attribution requires instrumentation at each service layer.
50%