Rigs

A second PC solves one problem and adds three

The advice is to add a second machine and stop dropping frames. It works. What nobody lists is everything that becomes your job the moment you do.

4 August 2026Rigs

One machine does three jobs on one GPU. Render the game. Compose the scene. Encode the stream. When it runs out, frames get skipped.

Splitting the work fixes that. It is a real fix. It is not a free one.

What the second machine buys

What it costs you

New problemWhat it looks like
Capture pathA card and a cable, or NDI and a network that can carry it
Audio routingGame audio, chat audio and your microphone now live on two machines
SyncCapture adds delay. Audio arriving by a different path does not match.
ControlScene switching, alerts and chat are on the other keyboard
CostA machine, a card, a monitor input, desk space
The one that eats an evening

Audio. Splitting game audio and voice chat onto separate tracks across two machines is the part people underestimate, and it is the part that matters for clipping later.

Capture card or NDI

A card gives you a fixed, low delay. It never touches your network. It costs money and a cable. It also caps you at whatever the card handles.

NDI needs no card. It moves over the network you already have. It costs bandwidth and adds latency. Full NDI at 1080p60 is a serious load on gigabit.

For one feed, a card. For several sources or machines in different rooms, NDI.

Before you buy anything

Check which counter is failing. Skipped frames from encoding lag is the one a second machine solves. Missed frames is scene load. Dropped frames is network.

Two of those three follow you to the new machine. People buy a second PC to fix a browser source that was eating the GPU. They arrive at the same stutter with a lighter wallet.

Try the cheap fixes first. Move encoding to a GPU that is not rendering the game. Drop the x264 preset a step. Both before you buy anything.

Sources

The Rig Wire reports on streaming technology from public documentation, source repositories and vendor releases. We did not bench-test what is described here. When we have tested something, we say so.