Hi nvidia,
What is the maximum number of encoder instances that can be created concurrently on Jetson AGX Orin?
Hi,
Please check
Jetson AGX Orin FAQ
And it is fine to have 8+ encoding tasks, since this is handled through the patches.
When the encoder is created, can we change the bitrate in the running mode?
Hi,
Yes, runtime bitrate change is supported.
Hi,
How to change runtime bitrate in AGX Jetson Orin? Is there any sample codes?
Hi,
Please refer to the sample:
/usr/src/jetson_multimedia_api/samples/01_video_encode
On AGX Orin the binding limit isn’t the instance count — it’s NVENC throughput. Two separate things:
Session count is a software limit, and it’s the one answered above: 8+ concurrent encode sessions are fine with the JetPack 6 patches.
Hardware ceiling is the encoder’s rated pixel rate. AGX Orin has 2 NVENC engines, rated (profile UHP):
| Module | Codec | 4K60 | 4K30 | 1080p60 | 1080p30 | Rated |
|---|---|---|---|---|---|---|
| AGX Orin 64GB | H.265 | 2 | 4 | 8 | 16 | 1,000 MPix/s |
| AGX Orin 64GB | H.264 | 1 | 3 | 7 | 14 | 930 MPix/s |
| AGX Orin 32GB | H.265 | 1 | 3 | 6 | 12 | 795 MPix/s |
| AGX Orin 32GB | H.264 | 1 | 2 | 5 | 11 | 740 MPix/s |
Source: NVIDIA Jetson AGX Orin Series Modules Data Sheet, DS-10662-001 v1.8, §4.5 Multi-Standard Video Encoder — Table 4-5 (64GB), Table 4-6 (32GB).
The arithmetic, so you can run it for a profile the datasheet doesn’t print: one 1080p30 stream is 1920 × 1080 × 30 = 62.2 MPix/s. On the 64GB module, 1,000 / 62.2 = 16.1 — which is exactly where the datasheet’s “1080p30 (16)” comes from. So for 2560×1440 @ 30: 2560 × 1440 × 30 = 110.6 MPix/s, and 1,000 / 110.6 ≈ 9 H.265 streams on the 64GB. That last one is pixel-rate interpolation, not a datasheet-printed figure — use it as a planning number, not a guarantee.
Two things that bite in practice:
- NVDEC and NVENC share memory bandwidth. If you’re decoding and encoding simultaneously, the real combined total sits below the sum of the two tables.
- Max NVENC/NVDEC clock is the same across nvpmodel presets on Orin, so these counts hold at 15 W just as at MAXN — power mode moves CPU/GPU, not the codec ceiling.
So: create as many sessions as your pipeline needs (8+ works fine), but size the deployment against MPix/s rather than session count. That’s the number that actually runs out.
Per-module, per-codec tables with the datasheet section and revision behind every row, if you want the reproducible version: Jetson AGX Orin 64GB H.265 decode capacity: 22 × 1080p30, 7 × 4K30 – Edge AI Stack