AMD Instinct MI350X
The air-cooled accelerator in AMD's Instinct MI350 Series, built on the CDNA 4 architecture and deployed eight to a platform.
AMD · Current generationShipping
MI350X
CDNA 4 · Instinct MI350 Series
- Memory
- 288 GB HBM3E
- Bandwidth
- 8 TB/s
- Power
- 1,000 W
Specifications at a glance
As published by AMD, checked 24 September 2026. Configurations vary by system builder.
- Form
- Passive OAM module; 8 GPUs per platform, up to 64 GPUs per air-cooled rack
- Cooling
- Air-cooled
- GPU memory
- 288 GB HBM3E per GPU; 2.3 TB per 8-GPU platform
- Memory bandwidth
- 8 TB/s per GPU
- FP4 compute
- 9.2 PFLOPS MXFP4 per GPU (dense); 73.8 PFLOPS per 8-GPU platform
- Interconnect
- Fourth-generation Infinity Fabric: 7 links at 153.6 GB/s per GPU; PCIe 5.0 x16
- Power
- 1,000 W total board power
- Status
- Shipping; launched June 2025
Configuration and site considerations
The MI350X is the air-cooled option, so standard air-cooled rack plans apply.
- Air-cooled platforms at 1,000 W board power per GPU.
- Air-cooled racks hold up to 64 GPUs.
- Document the platform, the GPU count and the system builder.
Prepare this system for a financing review
Document the configuration, the supplier and the deployment so the request can be assessed against the actual equipment.**
System configuration and quantity
Supplier and purchase order
Or fleet details and ownership information for deployed servers.
Deployment site and colocation agreement
Power and cooling readiness
Customer contract or documented on-demand revenue
