What *We* Did Today

Today was basically a major OBS/jefferyTV production-system overhaul.

1. Diagnosed NB3 Performance

Confirmed NB3 has 16 GB RAM and can be upgraded to 32 GB.
Reduced unnecessary startup load.
Established that ChatGPT/Codex can consume roughly 1+ GB RAM, and fully quitting it noticeably reduces memory use.
Investigated the occasional jerky mouse/UI behavior and saw periods around 70% CPU and high-70% memory utilization.

2. Sorted Out the USB-Camera Mess

Got all three external USB cameras operating together.
Determined that the two 4K cameras share the same VID/PID and Windows can confuse their friendly names.
Found a working USB arrangement using the USB-A and USB-C/powered-hub connections.
Got OBS recognizing the cameras reliably again.

3. Made a Major OBS Recording Improvement

We replaced CPU-heavy x264 recording with:

NVIDIA NVENC H.264 → CQP 20 → P5 → High Quality → Single Pass

That was probably today’s biggest performance improvement.

Afterward we were seeing essentially:

30.00 FPS
Extremely low rendering loss
0% encoding lag
Substantially lower OBS CPU demand
Considerably higher recording bitrate and quality than the old 3-Mbps CBR configuration

4. Added All Three Mevo NDI Cameras

OBS was successfully handling approximately:

3 USB cameras + 3 Mevo/NDI cameras = six camera sources

Even with that load, our OBS Stats remained remarkably clean.

5. Backed Up the Working OBS Configuration

Exported the OBS Profile.
Exported the Scene Collection.

That gives us a recovery point for today’s known-good setup.

6. Investigated Audio Synchronization

Found the historical 300/330-ms sync offsets.
Tested them and determined they were no longer appropriate for the current configuration.
Returned the offsets to 0 ms.
Basic camera/audio synchronization became substantially better.
Distinguished actual recording synchronization from what is seen and heard through live monitoring.

7. Investigated Apparent Drift

I ran longer music and clap tests. There appeared to be a small amount of movement in synchronization, but OBS Stats showed that the encoder was not falling behind.

We also recognized that using an internet-streamed video introduces another variable: browser and network buffering. That makes internet video a less reliable reference for measuring OBS clock drift.

8. Mapped the Two-Computer Production System

We established that the i7 is the main music workstation, running Cubase and the Steinberg interface, while NB3 is becoming the OBS production machine.

9. Considered NDI Screen Capture

We looked at replacing VDO.Ninja with NDI to carry desktop video and audio from the i7 to NB3.

That remains an available option, but then I remembered something important.

10. Found Out I Already Owned the Hardware Solution

I pulled out my ClearClick Video2USB capture device.

Instead of buying another capture card, we connected equipment I already had.

11. Successfully Got i7 Video Into NB3/OBS Through HDMI Hardware Capture

This was probably the other major breakthrough today:

i7 HDMI
   ↓
ClearClick
   ↓ USB
NB3
   ↓
OBS

This eliminates VDO.Ninja/network latency from the video path.

12. Established the Actual Audio Architecture

The current arrangement is now:

                    i7 DESKTOP
                        │
              ┌─────────┴─────────┐
              │                   │
            VIDEO               AUDIO
              │                   │
          HDMI OUT          Steinberg interface
              │                   │
         ClearClick          Headphone output
              │                   │
          USB video           Analog cable
              │                   │
              │            USB Audio CODEC
              │              │          │
              │              │          └──→ I MONITOR HERE
              │              │
              └──────────────┴──→ NB3 / OBS

The ClearClick’s HDMI audio is currently not being used. ClearClick supplies the video, while the USB Audio CODEC supplies the program audio.

Where We Ended

The old VDO.Ninja arrangement required something around 300 ms of compensation.

With the new ClearClick hardware video path, the remaining discrepancy is dramatically smaller. I estimated perhaps around 100 ms or less, although we have not measured it precisely yet.

The next step is simple: do a short clap calibration, determine which signal leads, and then adjust OBS in small 25–50 ms increments until the recording is tight.

We did a lot today. Most importantly, we did not merely tweak OBS. We moved considerably closer to a proper two-computer jefferyTV production architecture, with the i7 handling music and Cubase and NB3 handling the multi-camera OBS production.

What We Did Today