Source in, normalised
Mezzanine files and contribution feeds accepted, inspected and normalised before anything downstream has to cope with a surprise.
Transcoding on a GPU you already own
A five-rendition ladder means every minute of source is billed five times, every time you re-encode. It is a predictable, batchable workload with no latency requirement — close to the ideal case for hardware you own.
The bill that multiplies
Per-minute transcoding pricing looks reasonable until it is multiplied by ladder depth, and then again by every re-encode. Adding a rendition for low-bandwidth viewers, or changing a codec, turns from an engineering decision into a budget event — which is how catalogues end up stuck on ladders nobody would choose today.
On owned hardware the constraint becomes queue depth instead of invoice. Re-encoding the back catalogue overnight costs electricity, and that changes what you are willing to improve.
What the pipeline does
Ladder tuned to the material
Sport and dialogue do not want the same ladder. The profile follows the content rather than a vendor default.
HLS, segmented and encrypted
Packaged for adaptive playback with AES-128 applied at packaging time, so an unencrypted rendition never exists on disk.
Into the origin
Written to the S3-compatible origin with a sharded layout, so a large catalogue does not turn into a directory that nothing can list.
Encoding characteristics
| Concern | Managed service | Owned pipeline |
|---|---|---|
| Cost model | Per minute, per rendition | Capital already spent, plus power |
| Re-encoding | A budget decision | A scheduling decision |
| Ladder changes | Billed again in full | Queue time only |
| Codec adoption | Waits for vendor support | Waits for your hardware |
| Encryption | Applied downstream | Applied at packaging time |
| Throughput limit | Vendor concurrency quota | GPU count |
The things people ask first
How much GPU do we need?
It follows catalogue hours, ladder depth and how quickly new content must be live. For most regional services this is a smaller number than people expect.
What about live?
Live encoding runs on the same hardware with a latency-sensitive profile and priority over batch work, so an overnight re-encode cannot starve a match.
Can we keep using a managed encoder for overflow?
Yes. Burst to a managed service for exceptional load and keep the baseline on your own hardware — the pipeline supports both targets.
How much do you re-encode?
Catalogue hours, ladder depth and how often the ladder changes. Those three numbers usually make the case on their own.