Coming Soon

MI355X

ULTIMATE AI & HPC PERFORMANCE.
Built on the cutting-edge 4th Gen AMD CDNA™ architecture to drive the next era of innovation.

🔥 Scroll to the bottom to reserve 🔥

DIRECT LIQUID COOLED DESIGN

The global adoption of AI technology demands ever greater amounts of processing capacity. The direct liquid cooling option enables AMD Instinct MI355X GPUs to consume up to 1400W.

Liquid cooling helps reduce GPU die temperatures, expanding the accelerator’s power envelope, facilitating higher clock speeds and delivered performance. The use of cold plates enables the UBB to be integrated into standard and open compute servers, helping customers achieve higher densities.

INDUSTRY-STANDARD FORM FACTOR

The platform combines the power of eight accelerators on an industry-standard universal baseboard (UBB 2.0).

  • • 8 OAMs connected with AMD Infinity Fabric™ mesh
  • • 153.6 GB/s bidirectional links
  • 288 GB of HBM3E memory per accelerator
  • 8 TB/s of memory bandwidth
  • • Massive 2.3 TB of coherent, shared memory per platform

AMD MI355x Platform Architecture

Seven links per MI355X plus one PCIe® Gen 5 x16 connection per OAM device for upstream server and/or I/O connectivity.

Multi-Chip Architecture

The MI355X uses the 4th Gen AMD CDNA™ multi-chip architecture based on 3nm process technology.

Breakthrough AI Acceleration

With new expanded FP6 and FP4 datatype support, MI355X maximizes computational throughput and energy efficiency. Enhanced FP16 and FP8 processing positions the MI350 Series to deliver exceptional performance for advanced generative AI models.

K8s Integration

AMD GPU Operator simplifies deployment in Kubernetes. Day-0 support enables optimized models upon release.

Advanced Security

Verify hardware authenticity, secure multitenant GPU sharing, and encrypted high-speed communication.

Scalable Networking

Ethernet-based AI and HPC networking enables massive hyperclass scalability and low TCO.

Optimized Software & Framework Integration

The foundation of AMD accelerated computing, AMD ROCm™ software empowers AI developers to fully leverage AMD Instinct GPUs. With Day-0 support for PyTorch, TensorFlow, JAX, and ONNX Runtime.

OpenAI
PyTorch
Hugging Face
Databricks

Reserve Your Compute

Demand for next-generation accelerators is at an all-time high. By placing a reservation today, you secure your priority spot in the deployment queue.

Priority Allocation
Exclusive Hardware Updates
$100 Deposit

By reserving, you agree to our Terms.

Footnote explanations are available at: AMD Legal Claims