Most Powerful Rack-scale AI Infrastructure1

AMD Instinct™ GPUs and the AMD Helios™ rackscale solution drive frontier AI and HPC in data centers with leading performance, efficiency, and scalability. Built on AMD CDNA™ architecture and the open AMD ROCm™ software stack, which power frontier AI models, training, inference, simulation, and mixed workloads with predictable performance and strong value at any scale.

What’s New

AMD Helios™ Rackscale Solution Powered by 72 AMD Instinct™ MI455X GPUs: Built for Frontier AI

The AMD Helios design is a fully integrated AI infrastructure, combining the latest AMD Instinct™ GPUs, AMD EPYC™ server CPUs, AMD Pensando™ networking, and AMD ROCm™ software stack, designed using open industry standards enabling large-scale inference, frontier model training and fine-tuning.

Learn About AMD Helios
AMD Helios Rackscale
Helios compute tray lidded

AMD Instinct™ MI400 Series GPUs: Leadership Frontier AI, Sovereign AI, and HPC

AMD Instinct™ MI400 Series GPUs, built on the AMD CDNA™ 5 architecture, are supported by an open, ROCm™ software-based foundation. The portfolio spans the AMD Instinct MI455X GPU for frontier and sovereign AI and the AMD Instinct MI430X GPU for sovereign AI and HPC — enabling customers to select the right platform for their needs.

View AMD Instinct MI400 Series

Workloads

Sovereign AI

AMD Instinct™ MI400 Series GPUs offer nations across the world frontier or sovereign AI compute capabilities based on their needs in an open, proven, and trusted environment to unlock the promise of AI to drive economic growth, protect national interests and reveal new scientific frontiers.

More on Sovereign AI
abstract AI data center

Gen AI, Training & Inference

AMD Instinct™ GPUs are built for generative AI, delivering the performance, memory capacity, bandwidth, and open AMD ROCm™ software foundation needed to scale high-volume inference, training, and fine-tuning from development to production.

 

More on AI Solutions
abstract image

HPC: A Future with AI, Simulation, and Data-Driven Science

AMD Instinct™ GPUs offer industry-leading performance for current and future systems where AI, simulation, and data-driven science converge in a single platform at any scale.2

More on Research Solutions Advancing Science Discovery

Performance Leadership for Frontier AI and HPC

Up To
4X
Gen on Gen AI Performance
More Performance for GenAI and Inference

AMD Instinct MI455X GPUs with peak OCP MXFP4 are expected to offer up to 4X better peak theoretical performance compared to AMD Instinct™ MI355X GPU.3

Up To
50%
More Memory Capacity
Industry Leading Memory Capacity 

AMD Instinct MI455X and MI430X GPUs offer 50% more memory capacity and 6% higher peak memory bandwidth compared to NVIDIA Vera Rubin GPU.4

Up To
15%
More Peak Theoretical AI Performance
Throughput for Language Models

The AMD Helios rackscale solution, powered by AMD Instinct MI455X GPUs, are expected to offer up to 15% better OCP MXFP4 peak theoretical performance compared to NVIDIA Vera Rubin NVL72 with FP4 datatype.1

Up To
8.7X
FP64 FLOPS for HPC & Sovereign AI vs. NVIDIA Vera Rubin
Industry Leading Double Precision Performance for Trusted HPC Results

AMD Instinct MI430X GPUs are expected to offer up to 8.7X more FP64 peak theoretical performance than NVIDIA Vera Rubin.5

short top curved fade divider
AMD Instinct™ MI430X, MI440X, MI455X die

Under the Hood

AMD Instinct GPUs are built on AMD CDNA™ architecture, which offers advanced chiplet, HBM memory and Cache, and AMD Infinity Fabric™ technologies to AMD Instinct GPUs. AMD CDNA architecture offers matrix and vector core technologies for support of a broad range of datatype capabilities—from the highly efficient OCP MXFP 8/6/4 (including INT8, and FP16/BF16 sparsity support for AI), to the most demanding FP64 for HPC.

Learn More About CDNA
short bottom curved fade divider

Portfolio

Find the Right AMD Instinct GPU for Your AI and HPC Needs

Match your workload to the right AMD Instinct GPU for the best balance of performance, memory, interconnect, and energy efficiency for training, inference, or large-scale simulation.

Software

Built on an Open Software Ecosystem

AMD ROCm™ Software

AMD ROCm™ software includes a broad set of programming models, tools, compilers, libraries, and runtimes for AI models and HPC workloads targeting AMD Instinct GPUs.

About ROCm AMD ROCm™ Developer Hub
AMD Enterprise AI Suite

AMD Enterprise AI Reference Stack

By connecting key open-source AI frameworks and Gen AI models with an enterprise-ready Kubernetes platform, enterprises can go from bare metal compute to production-grade AI in minutes, minimizing the time from AI experimentation to large scale production on AMD compute platforms.

Learn About AMD Enterprise AI Reference Stack

Powering the AI & HPC Software Ecosystem with AMD Instinct™ GPUs

At AMD, we believe that advancing AI and HPC requires deep collaboration. Our collaborators span the full spectrum of AI and HPC ecosystem—from model development and large-scale training, to simulation, scientific research, and real-time inference. Together, we power breakthroughs in fields ranging from artificial intelligence to computational fluid dynamics, weather modelling, and beyond.

Ansys
cohere
Deepspeed
deepseek
docker
Essential AI
Exostellar
Hungging Face
JAX
Llama Stack
llm-d
meta
mlir
Modular
ONNX
openxla
PyTorch
Red Hat
seekr
Sglang
TensorFlow
Triton
vLLM
Zyphra
Learn More

Case Studies

View All

AMD Instinct Deployment Options

AMD Instinct™ MI350 Series Accelerator Platform

Cloud Service Providers

Scale AI Faster with AMD Instinct Cloud Partners

Access AMD Instinct GPUs through a growing network of cloud partners delivering optimized environments for AI and HPC workloads. Built on open software and flexible infrastructure, these platforms enable developers to move from experimentation to production seamlessly.

Get Started in the Cloud

Resources

Stay Informed

Sign up to receive the latest data center news and server content.

Sign Up
Footnotes
  1. Based on calculations by AMD Performance Labs in June 2026, for the AMD Helios Rackscale Solution to determine the peak theoretical performance when comparing peak Matrix FP16, BF16, INT8 and peak Open Compute Project MXFP6, MXFP8, FP8 and MXFP4 datatypes vs. NVIDIA Vera Rubin NVL72 (NVFP4 Dense). System manufacturers may vary configurations, yielding different results. MI400-005
  2. Based on engineering projections by AMD performance labs in July 2026, to estimate peak theoretical precision performance of an AMD Instinct™ MI430X GPU using FP64 (Vector) datatype vs. AMD Instinct MI300X, NVIDIA B200, B300 and Vera Rubin GPUs with FP64 (Vector) datatype. Results subject to change when products are released in market. MI400-001C
  3. Based on AMD Performance Labs calculations (June 2026) using an AMD Instinct™ MI455X GPU, peak theoretical performance (FP32, FP16, BF16, MXFP6, MXFP8, FP8, MXFP4 Matrix/Vector), compared to published specifications for AMD Instinct™ MI355X, MI350X, MI325X, MI300X, MI250X, and MI100 GPUs. Results may vary by system configuration and datatype. MI400-006
  4. Calculations by AMD Performance Labs in June 2026, based on published memory capacity and memory bandwidth of AMD Instinct™ MI455X and MI430X GPU when comparing specifications vs. NVIDIA Vera Rubin GPU (NVFP4 Dense).  System manufacturers may vary configurations, yielding different results. MI400-004
  5. Based on engineering projections by AMD performance labs in July 2026, to estimate peak theoretical precision performance of an AMD Instinct™ MI430X GPU using FP64 (Vector) datatype vs. NVIDIA Rubin GPU with FP64 (Vector) datatype. Results subject to change when products are released in market. MI400-001B