Support Matrix#

These support matrices list the supported hardware and software requirements of the NVIDIA ACE Agent.

Hardware#

Supported Hardware for NVIDIA ACE Agent Data Center Platforms#
	Hardware Compatibility
GPU Model	NVIDIA Turing T4 GPU NVIDIA Volta V100 GPU NVIDIA Ampere A100 GPU NVIDIA Ampere A30 GPU NVIDIA A10 GPU NVIDIA H100 GPU NVIDIA L4 GPU NVIDIA L40 GPU Note NVIDIA ACE Agent is supported on any NVIDIA Volta or later NVIDIA GPU (NVIDIA Turing or NVIDIA Ampere Architecture) for development purposes. Care must be taken not to exceed the memory available when training or deploying models.
Operating System	NVIDIA ACE Agent requires Linux x86_64 (Preferred Platform Ubuntu 20.04 and 22.04)
Mic and Headset	Any good quality microphone and speaker works with NVIDIA ACE Agent. The following models have been verified to work well. Microphone Linux x86 with USB microphone (for example, a Logitech H390 USB computer headset) Headphones Logitech H340 Logitech H390 Microsoft LX3000
GPU memory needed to deploy models into NVIDIA Triton	ASR model ~5600 MB TTS model ~3500 MB NMT models ~4500 MB per bilingual model ~5500 MB per megatron 500m model ~8500 MB per megatron 1b model NLP models GPU memory requirement for NLP models only includes the incremental memory needed for each model. There is an extra ~1500 MB GPU memory requirement for each Triton server instance. BERT Base NLP model ~600 MB per model
GPU memory needed for deploying an LLM model	In addition to a few tutorials, the LLM and RAG sample bots utilize locally deployed `meta/llama3-8b-instruct` LLM model which require separate A100 80 GB or a H100 GPU for local deployment. If you are planning to use a different LLM model, confirm the GPU compute requirements before deployment.

Software#

For additional information about the supported software stack and containers, refer to the Deep Learning Frameworks Support Matrix.

Supported Software for NVIDIA ACE Agent Data Center Platforms#
	Software Compatibility
Container	Container OS Ubuntu 20.04 NVIDIA Riva 2.17.0
Docker	Docker >= 19.03 with nvidia-docker installed For DGX users, refer to Preparing to use NVIDIA Containers.
TAO Toolkit	TAO Toolkit 21.10 or later
NeMo Guardrails	0.10.1
NVIDIA Driver	510+ 418.40+, 440.33+, 450.51+, 460.27+, 470.57+ for data center GPUs For earlier driver versions, refer to the NVIDIA Driver section in the Deep Learning Frameworks Support Matrix.
NVIDIA NGC CLI	Latest