On-premises robot system requirements

Observe the recommended system requirements and configuration guidelines for running performance testing robots in an on-premises environment. These recommendations are based on UiPath internal benchmarks and field investigations.

Multiplexing

Performance testing robots rely on multiplexing, meaning multiple Virtual Users (VUs) can be executed in parallel on the same machine.

The number of VUs a single robot executes simultaneously is called multiplexing factor (e.g., 10 VUs in parallel equals to multiplexing factor 10). The multiplexing factor is different per automation type.

• API automations support the highest multiplexing factor due to being CPU/lightweight and I/O-bound. Typical ranges are 20–100+ VUs per robot, depending on API payload size.
• Web automations support multiplexing >1. Typical ranges are 2–10+ VUs per robot, depending on page complexity (images, frameworks), think/think time, and DOM activity.
• Desktop automations are restricted to 1 VU per robot because of Windows session constraints.

Effective multiplexing depends on:

• CPU architecture/performance and available RAM
• Workload type and complexity (API > Web >> Desktop for multiplexing potential)
• System overhead and background services

Tip: Use separate robot pools for API vs. Web workloads to tune multiplexing independently and avoid resource contention.

Virtual machines

• Use compute-optimized VMs with strong CPUs and SSD storage.
• Recommended baseline: 16 physical CPU cores, 32 GB RAM, SSD.
• Alternative (lower-efficiency) setups: 4 cores / 8 GB RAM or 8 cores / 16 GB RAM.
• Keep virtual machines clean of unnecessary background services.

Software and operating systems

• Windows Server 2022 or newer
• Google Chrome installed (latest stable) for Web Automations

API Automations do not require a browser, but the above OS requirement still applies.

Resource ratios

• CPU: 0.5–1.8 VUs per physical core (measured on 16 physical cores)
• RAM: 0.5–2 VUs per GB RAM (measured on 32 GB)
• Count physical cores only; hyper-threading provides little benefit under sustained load

These baseline ratios apply most directly to Web workloads. API workloads often achieve higher VU density per core/GB due to lower rendering overhead. Validate your workloads via dry run and full execution before finalizing sizing.