Process Spotlight — top by RAM
CPU Temp
—
—
GPU Temp
—
—
System RAM
Shared memory pool visible to the CPU and all containers. 32 GB available to the OS — the other 96 GB is reserved as VRAM.
—
%
— / — GB
VRAM
GPU memory carved from the unified pool (96 GB). Consumed by models loaded via Ollama or llama-server. Drops to near-zero when models unload.
—
%
— / — GB
CPU %
Average CPU utilisation across all 16 cores. High load with moderate CPU% usually means waiting on I/O or GPU rather than compute.
—
%
16 cores
Swap
Disk-backed overflow. When RAM fills up, inactive pages spill here. High swap use slows things down — disk is ~100× slower than RAM.
—
%
— / — GB
GTT
Graphics Translation Table — system RAM dynamically mapped into GPU virtual address space. Flexible complement to VRAM; reclaimed by the OS when idle.
—
%
— / — GB
Disk
Root filesystem (/) usage. Docker images and volumes live here. Run 'docker system prune' to reclaim space from unused images.
—
%
— / — GB