AMD Instinct™ MI100 Accelerator Features: Designed on AMD CDNA Architecture with 120 Compute Units (7,680 cores) Up to 11.5 TFLOPs Peak FP64 Performance for HPC1 Up to 46.1 TFLOPs FP32 Matrix Peak Performance with All-New Matrix Cores for HPC & AI Workloads1 Up to 184.6 TFLOPs FP16 & 92.3 TFLOPs bFloat16 Peak for Ultra-Fast AI Training2,12 32 GB Ultra-fast HBM2 ECC Memory with up to 1.2 TB/s Memory Bandwidth Open & Portable AMD ROCm™ Ecosystem 2nd Gen Infinity Architecture with up to 340 GB/s of aggregate P2P GPU I/O bandwidth11 PCIe® Gen 4 x16 Ready GPU Specifications: GPU Specifications GPU Architecture CDNA Lithography TSMC 7nm FinFET Stream Processors 7,680 Compute Units 120 Peak Half Precision (FP16) Performance 184.6 TFLOPs Peak Engine Clock 1502 MHz Peak Single Precision Matrix (FP32) Performance 46.1 TFLOPs Peak Single Precision (FP32) Performance 23.1 TFLOPs Peak Double Precision (FP64) Performance 11.5 TFLOPs Peak INT4 Performance 184.6 TOPs Peak INT8 Performance 184.6 TOPs Peak bfloat16 92.3 TFLOPs OS Support Linux x86_64 Requirements External Power Connectors 2x PCIe® 8-pin Total Board Power (TBP) 300W Peak GPU Memory Dedicated Memory Size 32 GB Dedicated Memory Type HBM2 Memory Interface 4096-bit Memory Clock 1.2 GHz Peak Memory Bandwidth Up to 1228.8 GB/s Memory ECC Support Yes (Full-Chip) Board Specifications Form Factor PCIe® Add-in Card Bus Type PCIe® 4.0 x16; PCIe® 3.0 x16 Infinity Fabric™ Links 3 Peak Infinity Fabric™ Link Bandwidth 92 GB/s Cooling Passive Dimensions Board Height Full Height Board Length 10.5" (267 mm) Board Width Double Slot Additional Features Supported Technologies AMD CDNA™ Architecture; AMD Infinity Architecture; AMD ROCm™ - Ecosystem without Borders RAS Support Yes Page Retirement Yes Software API Support OpenMP® Yes OpenCL™ Yes HIP Yes ROCm™ Open Ecosystem Yes Frameworks TensorFlow Yes PyTorch Yes Kokkos Yes RAJA Yes Product Basics Product Family AMD Instinct™ Product Line AMD Instinct™ MI Series Platform Server Launch Date 11/16/2020