Do I need to forward a port or open my firewall?

No. Both computers open an outbound HTTPS and WSS connection to the relay. Neither machine accepts an inbound connection, so there is no port to forward and no inbound firewall rule to add.

Is my model uploaded anywhere?

No. Model weights stay on the machine that runs them and are never uploaded, relayed or copied to the other PC. What crosses the network is the request, your account email, the public-key fingerprint, and the result artifact.

How does the other computer prove it is mine?

It presents your signed-in account email and a public-key fingerprint. The host sees both in full and must approve the key before that machine can be asked to run anything. Pairings can be revoked at any time.

Is remote benchmarking included in the free trial?

No. The five-day trial covers local benchmarks, the dashboard and both native engines. Remote benchmarking is part of Pro; buying credits once adds remote runs and keeps them.

Benchmark another Windows PC remotely, without opening a port

Q: What does a remote benchmark actually return?

An artifact recorded by the machine that ran the model: model id and SHA-256, tokens per second and total milliseconds, the prefill and decode split, peak RAM and VRAM, the GPU adapter and driver version, and whether a fallback path was used. If the GPU path was not used, the run says so.

Lizard can measure a second computer from the one you are sitting at: its GPU, its model, and its real tokens per second. You approve the other machine by key fingerprint before it can be asked to do anything, and no model is ever uploaded.

How the two computers meet

Both machines open an outbound HTTPS and WSS connection to the Lizard relay. Neither one accepts an inbound connection, so there is no port to forward and no inbound firewall rule to add. The relay queues messages between them and never holds a model.

Pairing, step by step

Create a one-time code. On the computer doing the asking, press Add benchmark computer.
Claim it on the other PC. Enter the code there; that machine sends its account email and public-key fingerprint through the relay.
Approve the key, not the name. The host sees the email and the full fingerprint and approves or rejects it.
Probe or benchmark. Ask the paired PC for a DirectX probe, or a benchmark of the model selected on it. The client confirms before anything runs.
Read the evidence. The result comes back as an artifact rather than a headline number.

What comes back from a remote run

Each completed run returns the model id and its SHA-256, tokens per second and total milliseconds, the prefill and decode split, peak RAM and VRAM, the GPU adapter and driver version, and whether a fallback path was used. That last field matters most: if the GPU path was not actually used, the run says so instead of quietly reporting a CPU number as a GPU result.

What crosses the network, and what does not

Model weights stay on the machine that runs them. Prompts and answers stay local — a remote run measures speed, not text. What crosses the network is the request itself, your account email, the public-key fingerprint, and the result artifact.

Availability

Remote benchmarking is part of Pro. The five-day trial covers local benchmarks, the dashboard and both native engines; buying credits once adds remote runs and keeps them.

See Lizard Chat, the benchmark results, or download Lizard for Windows.