> For the complete documentation index, see [llms.txt](https://docs.cloudeka.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cloudeka.ai/deka-gpu/deka-gpu-autoscaling/keda-autoscalling/example-autoscaling-vllm-with-keda-based-on-gpu-kv-cache-usage/metric-gpu-kv-cache-usage.md).

# Metric: GPU KV Cache Usage

Rule of Thumb for GPU KV Cache Utilization:

{% hint style="success" %}
According to [NVIDIA NIM Operator documentation](https://docs.nvidia.com/nim-operator/latest/service.html), a threshold of 50% is recommended as a default for scaling triggers.
{% endhint %}

<table><thead><tr><th width="146.6666259765625">Utilixation Range</th><th>Description</th></tr></thead><tbody><tr><td>0% - 30%</td><td>Low usage - resources are under-utilized</td></tr><tr><td>30% - 70%</td><td>Medium usage - healthy range under normal load</td></tr><tr><td>70% - 90%+</td><td>High usage - risk of cache eviction or inference slowdowns</td></tr></tbody></table>