Hello,
We are experiencing a failure when attempting SVO recording with any hardware-accelerated compression mode (H264, H265, H264_LOSSLESS, H265_LOSSLESS) on a Blackwell-generation GPU. CPU-based LOSSLESS mode works fine.
Environment:
* ZED SDK: 5.0.7 (installed 2026-03-31)
* Camera: ZED-2i (S/N 39253932), firmware 1523
* GPU: 2x NVIDIA RTX PRO 6000 Blackwell Server Edition (PCI device 2bb5)
* Driver: 590.48.01 (NVENC API version 13.0, Video Codec SDK 13.0)
* CUDA: 13.1 (nvidia-smi) / toolkit 12.0 (apt)
* OS: Ubuntu 24.04.3 LTS, kernel 6.8.0-106-generic, x86_64
* Python: 3.10 (conda), pyzed from pip
* libnvidia-encode-590 is installed and loads correctly
*
Symptoms:
We wrote a diagnostic script that tests every compression mode in isolated subprocesses. Results:
Compression Mode
SVO v1 (ZED_SDK_SVO_VERSION=1)
SVO v2 (default)
H264
Error 13 (SVO_UNSUPPORTED_COMPRESSION)
Segfault (signal 11)
H265
Error 13 (SVO_UNSUPPORTED_COMPRESSION)
Segfault (signal 11)
H264_LOSSLESS
Error 13 (SVO_UNSUPPORTED_COMPRESSION)
Segfault (signal 11)
H265_LOSSLESS
Error 13 (SVO_UNSUPPORTED_COMPRESSION)
Segfault (signal 11)
LOSSLESS
OK
OK
The SDK log shows:
[ZED][SLHW] Failed to create encoder. Err : CreateEncoder : m_nvenc.nvEncInitializeEncoder(m_hEncoder, &m_initializeParams) returned error 8
NVENC error 8 is NV_ENC_ERR_UNSUPPORTED_PARAM, meaning the encoder is found but rejects the initialization parameters.
Analysis:
The RTX PRO 6000 Blackwell has 9th-generation NVENC, which requires Video Codec SDK 13.0 APIs. The driver correctly reports NVENC API 13.0 via NvEncodeAPIGetMaxSupportedVersion. We suspect the NVENC wrapper bundled inside libsl_zed.so was compiled against an older Video Codec SDK and passes initialization parameters that the Blackwell encoder does not accept.
Additionally, the SVO v2 code path does not handle the NVENC initialization failure gracefully and segfaults instead of returning an error code.
Questions:
1. Is Blackwell NVENC encoding expected to work with ZED SDK 5.0.7, or is a newer release required?
2. Is there a patched SDK version or hotfix available that addresses this?
We can provide the full diagnostic script, core dumps, or any additional logs if needed.
Please note - we cannot roll back to older NVIDIA driver versions as they do not support the capabilities we need of the Blackwell GPUs