“12K production” is not one fixed workload. Raster dimensions, frame rate, codec, bit depth, chroma sampling, RAW compression and effects determine the data rate and compute requirement. A procurement specification must name the exact format and workload rather than simply request a “12K workstation”.
For the network-capacity choices that affect high-resolution post-production, see our guide to 10GbE versus 40GbE.
Define a representative 12K format
For a theoretical example, use a raster of 12,288 × 6,480 pixels, 24 frames per second and 16-bit data. This is an illustrative calculation, not a claim that every 12K camera or codec uses this exact raster.
Pixels per frame: 12,288 × 6,480 = 79,626,240 pixels.
16-bit single-channel Bayer data: 79,626,240 × 2 bytes ≈ 159.3 MB per frame, or about 3.82 GB/s at 24 fps. That is approximately 13.75 TB per hour before compression and overhead.
16-bit RGB data: 79,626,240 × 3 channels × 2 bytes ≈ 477.8 MB per frame, or about 11.47 GB/s at 24 fps. That is approximately 41.3 TB per hour before compression and overhead.
Real camera RAW and video codecs compress or pack data differently. The calculation illustrates why codec, bit depth and colour representation must be stated before sizing storage and network capacity.
Workstation requirements
Specify the application, codec, effects, number of simultaneous streams, colour pipeline, AI tasks and export target. Test CPU decode, GPU effects, VRAM, system RAM, scratch storage and thermal behaviour using the actual project. Avoid recommending a fixed GPU or CPU configuration without benchmarking the named software and media.
For current workstation GPU memory and power specifications that can inform a procurement shortlist, consult NVIDIA’s professional workstation GPU specifications.
Storage and network budget
Measure sustained read/write throughput, queue depth, latency and concurrent access under the production workload. Size the working set for source media, decoded caches, AI inference inputs, proxies and render outputs. Keep source, cache and archive policies separate. Provide headroom for concurrency and backup rather than using nominal interface speed as the target.
For independent backup copies, immutability and restore testing, see our multi-terabyte backup architecture guide.
For a storage cost model covering NAS, cloud object storage and DAM, see NAS versus cloud DAM.
Example acceptance gates
The following are suggested internal starting gates, not industry standards. Adjust them to the project and record approval:
- Playback: zero dropped frames during a 30-minute representative timeline at the required format.
- Frame processing: p95 render or processing time remains within the 24 fps frame budget of 41.7 ms for real-time tasks where real-time output is required.
- Storage: sustained measured throughput retains at least 30% headroom over the observed peak workload.
- Memory: peak use remains below 80% of available system or GPU memory during the acceptance run.
- Reliability: no out-of-memory errors, thermal shutdowns or unhandled job failures during the defined sustained test.
- Delivery: output frame count, codec, colour metadata and checksum validation pass.
Procurement test plan
- Lock the raster, frame rate, codec, bit depth and project complexity.
- Run a representative edit, playback, render and AI-enhancement workload.
- Measure p50/p95 frame time, storage throughput, memory, power and temperature.
- Repeat with expected concurrent users and background tasks.
- Record failures, bottlenecks and recovery time.
- Purchase against the measured configuration, not a generic 12K label.
Conclusion
A 12K station is a system-level specification. Define the actual media format, calculate theoretical data rates, benchmark the complete application and use explicit acceptance gates. The numerical example above is a capacity model, not a camera benchmark or a universal workstation recommendation.