Announced as a product family, not as a specific accelerator. AMD previewed the Instinct MI500 Series at CES 2026 on 5 January 2026 and restated it at Advancing AI on 23 July 2026, saying it is planned to launch in 2027, but it has not yet named a single MI500 SKU. The MI400 Series eventually resolved into the MI430X, MI440X and MI455X; the MI500 Series has no equivalent breakdown yet, so this page describes the family. AMD has published only three specifications for it, all of them categorical: CDNA 6 architecture, a 2nm process and HBM4E memory. It has published no memory capacity, no bandwidth, no board power, no compute unit count and no performance figure at any precision, so every numeric field here is blank by design rather than pending research. The 1,000x performance figure widely quoted for this part is AMD's projection for an MI500-powered rack against an MI300X platform, not a per-GPU specification, and AMD names no precision on either side of that comparison.

AMD Instinct MI500 NEW

CDNA 6 2027 2nm

Overview

The AMD Instinct MI500 Series is AMD's next generation of data center accelerators, previewed by Dr. Lisa Su at CES 2026 and planned to launch in 2027. It is the successor to the MI400 Series that ships through 2026 and 2027 as the MI455X, MI440X and MI430X. AMD has disclosed three things about the silicon. It is built on a new architecture, CDNA 6, the sixth generation of AMD's data center compute design and the successor to the CDNA 5 architecture used across the MI400 line. It is manufactured on an advanced 2nm process, a full node beyond the mix of 2nm and 3nm chiplets used in the MI455X. And it uses HBM4E memory, a step beyond the HBM4 introduced with the MI400 Series. AMD has also placed it in its rack roadmap: MI500 Series GPUs will power the Helios 500 rack-scale system alongside EPYC "Verano" processors and Pensando "Como" and "Monza" networking, and AMD has already named the generation after it, the MI600 Series in a Helios 600 rack. Beyond those attributes AMD has published nothing. There is no announced capacity, bandwidth, board power, chiplet configuration, clock speed, transistor count or throughput figure at any precision, and there is no MI500 datasheet or product page. AMD's only performance statement is that the MI500 Series is on track to deliver up to a thousandfold increase in AI performance over the MI300X GPUs of 2023. That claim deserves care: AMD's own endnote defines it as a comparison between an MI500-powered AI rack and an MI300X platform, based on engineering projections made in December 2025, and it names no precision on either side. It compares a rack to a platform, over a period in which AMD added FP8, FP6 and FP4 datapaths that the MI300X never had, so it cannot be reduced to a per-accelerator number and no performance data is recorded here. Individual MI500 SKUs, and the specifications that come with them, are expected to be named closer to launch.

Performance Metrics

No performance metrics available for this GPU.

Power Specifications

TDP

--

Max Power

--

Power Connector

PCIe Slot

Cooling

Air

Memory Specifications

Capacity

--

Type

HBM4e

Bandwidth

--

Interface

--

Hardware & Design

Form Factor

--

Architecture

CDNA 6

Process Node

2nm

Launch Year

2027

Variant

Standard

Market Segment

Professional

Full Specifications

Memory
Memory Type HBM4e
Power & Thermal
Enterprise Features
Compute APIs ROCm
General
Architecture CDNA 6
Launch Year 2027

Documentation & Resources

Common Use Cases

General Compute AI/ML Workloads Data Processing

The AMD Instinct MI500 is optimized for high-performance computing tasks with CDNA 6 architecture delivering high TFLOPS of compute power.

Where to Rent

Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.

Browse GPU Cloud Providers

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs