Accelerator benchmark
Checks the worker's accelerators and measures its speed.
How it works
Lists every accelerator and driver the worker finds, then measures CPU throughput and memory bandwidth. Put it first in a flow to check a new worker.
Outputs
| Output | Gives | What it is |
|---|---|---|
Resultsmetrics | Metrics json | benchmark.json: throughput and memory of each accelerator |
Live telemetrytelemetry | Live telemetry | What the block sends while it runs, for views such as Chart or 3D view. |
Settings
| Setting | Default | What it does |
|---|---|---|
Duration (seconds)duration |
20 | From 5 to 3600 |
Precisionprecision |
fp32 | One of: fp32, fp16, bf16 |
Where it runs
On one of your computers with the EsaSim worker: NVIDIA GPU, AMD GPU, Intel GPU, huawei, Tenstorrent, Any worker.
In its code, settings are ctx.params["key"], inputs ctx.inputs["id"] and outputs ctx.output("id", "name.ext"); see Custom blocks.
More in Simulation
3D sceneSays which 3D model each object uses and which trajectory it follows, for the 3D view and the Render block.Fleet-size sweepHow many robots does the warehouse need? Hundreds of fleet-size × order-rate scenarios with inbound and outbound flow, real speeds, times, batteries and incidents, simulated at once as one batch on the CPU, an NVIDIA GPU or Tenstorrent cards.Landing sweepSimulates drone landings on a moving pad under wind, in parallel.RenderRenders a 3D scene on the worker into frames (pictures) and a video, and streams the frames while it works.Soft body modelTurns segmented anatomy into a soft-tissue model the 3D view simulates in real time: organs that sag, sway and push each other under gravity, held by a fixed skeleton and contained by the skin (metaball method of Gianluca De Novi, PhD).Warehouse fleetA warehouse of 3,600 m² run live: deliveries stored on the shelves and orders picked by up to 60 mobile robots, with real speeds, times, batteries and incidents, on the CPU, an NVIDIA GPU or Tenstorrent cards.