Export
The link always opens the latest version of this page.
Ask an AI about it
Open in ChatGPTA new chat that reads this entry from the site Open in ClaudeThe same, in ClaudeShare the link instead of a file, and they will always see the latest version.
Files are a snapshot of today. The live page has anything newer.
AMD MI450
GPU accelerator for AI workloads with 19.6 TB/s HBM bandwidth, TDM tensor engine, and workgroup clusters
- Category
- GPU accelerator 1 source
The AMD MI450 is a GPU accelerator designed for AI workloads 1.
Memory and Bandwidth
The MI450 features 432 GB of HBM (high-bandwidth memory) with 19.6 TB/s bandwidth 1, enabling high-throughput data movement for large-scale model inference and training.
Compute Architecture
Each SIMD (single-instruction multiple-data) unit contains 1024 vector general-purpose registers (VGPRs) 1. Workgroup processors (WGPs) are equipped with 320 KB of local data share (LDS) memory 1, providing fast on-chip storage for inter-thread communication within workgroups.
Workgroup Clusters
The MI450 introduces workgroup clusters with hardware-supported cluster barriers 1. This architectural feature enables efficient synchronization and communication across multiple workgroups, improving performance for algorithms requiring inter-workgroup coordination.
Facts 1
| Attribute | Value | Confidence | Sources | Cited |
|---|---|---|---|---|
| Category | GPU accelerator | low | 1 | 1 |
Developed by 1
Official links
Sources 1
- Attention Decode on AMD MI450 GPUs: A Gluon Kernel Optimization Guide · Hacker News (YC)