Large Language Models (RC15)

Support Matrix

NVIDIA NIMs for large-language models will run on any NVIDIA GPU, as long as the GPU has sufficient memory, or on multiple, homogenous NVIDIA GPUs with sufficient aggregate memory and CUDA compute capability > 7.0 or higher. Some model/GPU combinations are optimized. See the following Supported Models section for further information.

  • Linux operating systems (Ubuntu 20.04 or later recommended)

  • NVIDIA Driver >= 535

  • NVIDIA Docker >= 23.0.1

These models are optimized using TRT-LLM and are available as pre-built, optimized engines on NGC and should use the Chat Completions Endpoint.

Llama 3 8B Instruct

Optimized configurations

The GPU Memory and Disc Space values are in GB; Disk Space is for both the container and the model; Profile is for what the model is optimized.

GPU

GPU Memory

Precision

Profile

# of GPUs

Disk Space

H100 80 FP8 Throughput 1 8.5
H100 80 FP8 Latency 2 8.5
H100 80 FP16 Throughput 1 16
H100 80 FP16 Latency 2 16
A100 80 FP16 Throughput 1 16
A100 80 FP16 Latency 2 16
L40S PCIe 48 FP8 Throughput 1 8.5
L40S PCIe 48 FP8 Latency 2 8.5
L40S PCIe 48 FP16 Throughput 1 16
A10G PCIe 24 FP16 Throughput 1 16
A10G PCIe 24 FP16 Latency 2 16

Non-optimized configuration

The GPU Memory and Disc Space values are in GB; Disk Space is for both the container and the model.

GPUs

GPU Memory

Precision

Disk Space

Any NVIDIA GPU with sufficient GPU memory or on multiple, homogenous NVIDIA GPUs with sufficient aggregate memory and compute capability > 7.0 or higher 24 FP16 16

Llama 3 70B Instruct

Optimized configurations

The GPU Memory and Disc Space values are in GB; Disk Space is for both the container and the model; Profile is for what the model is optimized.

GPU

GPU Memory

Precision

Profile

# of GPUs

Disk Space

H100 80 FP8 Throughput 4 85
H100 80 FP8 Latency 8 85
H100 80 FP16 Throughput 4 155
H100 80 FP16 Latency 8 155
A100 80 FP16 Throughput 4 155
L40S PCIe 48 FP8 Throughput 4 85
L40S PCIe 48 FP8 Latency 8 85
L40S PCIe 48 FP16 Throughput 8 155

Non-optimized configuration

The GPU Memory and Disc Space values are in GB; Disk Space is for both the container and the model.

GPUs

GPU Memory

Precision

Disk Space

Any NVIDIA GPU with sufficient GPU memory or on multiple, homogenous NVIDIA GPUs with sufficient aggregate memory 240 FP16 100

The following LoRA formats are supported:

Foundation Model

HuggingFace Format

NeMo Format

Meta-Llama3-8b-Instruct fused/grouped - Query Attention fused/grouped - Query Attention
Meta-Llama3-70b-Instruct fused/grouped - Query Attention fused/grouped - Query Attention
Previous Configuring a NIM
Next API Reference
© Copyright © 2024, NVIDIA Corporation. Last updated on May 30, 2024.