How we will measure

This page is published before any result, so that the method cannot be bent to the numbers. Status: the stand is being assembled; no results yet.

Status

September 2026: methodology drafted, stand components being ordered, automation scripts in progress. The comparison table on the front page uses vendor documentation and community reports until the first full run. When results appear, they will carry a date, exact versions and a link to raw data.

The stand

Host: desktop PC with an RTX 4070 or better (NVENC with AV1), 144–240 Hz monitor.
Clients: a laptop with an integrated GPU (worst-case decoder) and a desktop with a discrete GPU.
Network: gigabit switch for LAN; a Linux box with tc netem to emulate WAN: RTT 20/40/80 ms, loss 0.5/1/2 %, jitter.
Input-to-photon meter: photodiode on the client screen plus a microcontroller that fires the click and timestamps the light change (an NVIDIA LDAT or equivalent); a 240 fps camera as the secondary method.

What is measured

Metric	How
Input-to-photon latency	30 clicks per data point, p50 and p95, monitor response time subtracted
Frame time p99 and dropped frames	from the client's frame log; for RDP FIDO from the built-in trace
Picture quality	VMAF and PSNR of the captured stream against the host's original at equal bitrate; bitrate needed for equal VMAF
Behaviour under loss	same scenes on the 0.5/1/2 % loss profiles: artifacts, stalls, recovery time
Host load	CPU and GPU utilisation while streaming
Time to first game	from a clean Windows install to the first playable frame, with a stopwatch
Security model	which ports are open and when, who holds the keys, whether a third-party account is required

Scenes and settings

The same five scenes for every product: a competitive shooter, an open-world racer, an action RPG, a 2D indie title, and an emulator. 1440p at 60 and 144 Hz, bitrate 20 and 40 Mbit/s, LAN and three WAN profiles. Every product on its default settings first, then on its best documented low-latency settings. All versions and drivers are recorded.

Open data and reproduction

Scripts, raw CSV files and the photodiode firmware will be published in a public repository under the MIT licence, so that anybody with similar hardware can repeat the runs. Where RDP FIDO loses, the numbers are published as they are. Runs are repeated monthly and after every RDP FIDO release; old results stay online with their dates.

Independent checks

Reviewers and bloggers who test remote play get Pro keys, the stand description and the raw data with an invitation to reproduce. Their results are linked from the front page regardless of outcome.