Support Matrix#

This page lists the support matrices for the NVIDIA Speech NIM microservices. Each support matrix page contains the following sections.

Category

Description

Hardware

Lists the compatible NVIDIA GPUs and supported precision levels (FP16, FP8) for the NIM.

Software

Specifies operating system, NVIDIA driver, and Docker version requirements.

Supported Models

Lists all available models with their publishers and WSL2 support status.
Each model includes deployment profiles that specify the NIM_TAGS_SELECTOR value, inference mode (streaming/offline), batch size, and CPU/GPU memory requirements.

The support matrix pages also include additional model-specific information, such as supported languages, available voices, and speaker diarization profiles.

ASR NIM Support Matrix

Hardware, software, and model profiles for ASR models including Parakeet, Canary, Conformer, and Whisper. Includes inference modes, speaker diarization profiles, and supported languages.

NVIDIA ASR NIM Support Matrix
NMT NIM Support Matrix

Hardware, software, and model profiles for the Riva Translate model. Includes supported translation languages.

NVIDIA NMT NIM Support Matrix
TTS NIM Support Matrix

Hardware, software, and model profiles for TTS models including Magpie. Includes available voices and supported languages.

NVIDIA TTS NIM Support Matrix

Container Image to Served Model Mapping#

For some NIMs, the container image name used in docker run and CONTAINER_ID does not match the string returned by --list-models or the voice catalog at runtime. This table maps the publicly advertised container name to the served-model identifier:

Container Image

Served Name from --list-models or Voice Catalog

Architecture

nvcr.io/nim/nvidia/nemotron-asr-streaming

cache-aware-parakeet-rnnt-en-US-asr-streaming-sortformer

Cache-Aware FastConformer-RNNT with Sortformer diarization head (en-US, streaming-only)

nvcr.io/nim/nvidia/magpie-tts-multilingual

Voices Magpie-Multilingual.{LOCALE}.{Speaker}[.{Emotion}]

Magpie TTS (multilingual, streaming + offline)