-
Product Basics
- Name
AMD Instinct™ MI350X Platform
- Family
Instinct
- Series
Instinct MI300 Series
- Form Factor
Instinct Platform (UBB 2.0)
- GPUs
8x Instinct MI350X OAM
- Dimensions
417mm x 553mm
- Launch Date
6/12/2025
-
Image Zoom
AMD Instinct™ MI350X Platform
The AMD Instinct™ MI350X Platform is an industry standard UBB 2.0 compatible solution for data centers that sets a new standard for Generative AI, training and high performance computing (HPC).
-
General Specifications
- Total Memory
2.3 TB HBM3E
- Memory Bandwidth
8.0 TB/s Per OAM
- Infinity Architecture
4th Generation
- Bus Type
PCIe® 5.0 x16
- Total Aggregate Bi-directional I/O Bandwidth (Peer-to-Peer)
1,194.8 GB/s
- Warranty
3 Year Limited
-
AI Performance
- Total Theoretical Peak FP4 Performance
73.8 PFLOPs
- Total Theoretical Peak FP4 Performance with Structured Sparsity
147.6 PFLOPS
- Total Theoretical Peak FP6 Performance
73.8 PFLOPs
- Total Theoretical Peak FP6 Performance with Structured Sparsity
147.6 PFLOPS
- Total Theoretical Peak FP8 Performance
36.9 PFLOPs
- Total Theoretical Peak FP8 Performance with Structured Sparsity
73.8 PFLOPS
- Total Theoretical Peak FP16 Performance
18.4 PFLOPs
- Total Theoretical Peak FP16 Performance with Structured Sparsity
36.8 PFLOPs
- Total Theoretical Peak INT4 Performance
36.9 POPs
- Total Theoretical Peak INT4 Performance with Structured Sparsity
73.8 POPs
- Total Theoretical Peak INT8 Performance
36.9 POPs
- Total Theoretical Peak INT8 Performance with Structured Sparsity
73.8 POPS
- Total Theoretical Peak BFLOAT16 Performance
18.5 PFLOPs
- Total Theoretical Peak BFLOAT16 Performance with Structured Sparsity
36.9 PFLOPs
*TF32 support through software emulation.
-
HPC Performance
- Total Theoretical Peak Double Precision Matrix (FP64) Performance
576.8 TFLOPs
- Total Theoretical Peak Double Precision (FP64) Performance
576.8 TFLOPs
- Total Theoretical Peak Single Precision Matrix (FP32) Performance
1.2 PFLOPs
- Total Theoretical Peak Single Precision (FP32) Performance
1.2 PFLOPs
