GPU telemetry is runtime data emitted by a graphics processor or accelerator, such as utilization, memory use, temperature, power draw, clock rates, error counters, and queue activity. In AI and high-performance computing, these signals describe what the hardware is doing while workloads run, not just what the operating system sees.
It matters in cyber security because GPU behavior can expose suspicious or abnormal activity that standard endpoint logs miss. Sudden changes in utilization, unexpected memory pressure, unusual process scheduling, or fabric errors may point to misconfiguration, resource abuse, cryptomining, data exfiltration, or other misuse. Defenders use GPU telemetry to baseline normal workload patterns, correlate events across cloud and application logs, and detect anomalies in accelerator-heavy environments. Attackers may try to hide inside that layer by blending with legitimate compute demand, so visibility at the GPU level helps close an important blind spot.