AMD
AMD Instinct MI210 64GB HBM2e AI Accelerator Card 100-300000008H
AI accelerator card with 64GB HBM2e memory delivering 181 TFLOPs FP16 performance via PCI Express 4.0 x16 interface
- InterfacePCI Express 4.0 x16
- Chipset ManufacturerAMD
- GPUInstinct MI210
- Floating Point FormatsPeak Half Precision (FP16) Performance: 181 TFLOPs Peak Sing
- Memory Size64GB
- Memory Clock1.6 GHz
- 181 TFLOPs FP16 performance for rapid AI training
- 64 GB HBM2e memory with 4096-bit interface for massive datasets
- PCI Express 4.0 x16 interface for high-throughput data transfer
- Fanless dual-slot design fits dense server configurations
- AMD CDNA 2 architecture with ROCm ecosystem for HPC workloads
Product Identity
The AMD Instinct MI210 64GB HBM2e AI Accelerator Card is built for data-centre workloads that demand massive memory bandwidth and matrix throughput. It targets researchers and engineers training large models or running high-performance compute jobs. Buyers who need a graphics card for rendering or gaming should look elsewhere.
Sustained Load Behaviour
The fanless design relies on server chassis airflow to dissipate heat during continuous FP16, bfloat16 or INT8 matrix operations at 181 TFLOPs. FP64 matrix performance holds at 45.3 TFLOPs for double-precision simulations. Without forced-air cooling from the rack, thermal throttling will reduce sustained throughput.
Installation Reality
The full-height dual-slot card installs in any PCI Express 4.0 x16 slot with mechanical clearance for the bracket. No auxiliary power cables are required; the slot delivers all power. Common issues arise when the host chassis lacks sufficient airflow or when the BIOS does not allocate enough MMIO space for 64GB HBM2e.
Highlights
- 181 TFLOPs FP16 performance for rapid AI training
- 64 GB HBM2e memory with 4096-bit interface for massive datasets
- PCI Express 4.0 x16 interface for high-throughput data transfer
- Fanless dual-slot design fits dense server configurations
- AMD CDNA 2 architecture with ROCm ecosystem for HPC workloads
Specifications
| Brand | AMD |
|---|---|
| Model | 100-300000008H |
| Interface | PCI Express 4.0 x16 |
| Chipset Manufacturer | AMD |
| GPU | Instinct MI210 |
| Floating Point Formats | Peak Half Precision (FP16) Performance: 181 TFLOPs Peak Single Precision Matrix (FP32) Performance: 45.3 TFLOPs Peak Single Precision (FP32) Performance: 22.6 TFLOPs Peak Double Precision Matrix (FP64) Performance: 45.3 TFLOPs Peak Double Precision (FP64) Performance: 22.6 TFLOPs Peak INT4 Performance: 181 TOPs Peak INT8 Performance: 181 TOPs Peak bfloat16: 181 TFLOPs |
| Memory Size | 64GB |
| Memory Clock | 1.6 GHz |
| Memory Interface | 4096-bit |
| Memory Type | HBM2e |
| Cooler | Fanless |
| Form Factor | Full Height |
| Slot Width | Dual Slot |
| Features | AMD CDNA 2 Architecture , AMD Infinity Architecture , AMD ROCm - Ecosystem without Borders |
| Shipping weight | 2.27 kg |
| Package size | 381 × 254 × 127 mm |
Questions about this item
How much heat does the fanless MI210 produce when it runs at 181 TFLOPs all day?
The card has no onboard fan, so it relies on server chassis airflow to dissipate the heat generated at full 181 TFLOPs load.
What slot and power connectors must my motherboard provide for the PCI Express 4.0 x16 interface?
You need a full-height PCIe 4.0 x16 slot and the matching auxiliary power cables required by the dual-slot board.
Will the 64GB HBM2e memory work with my existing PCIe 3.0 servers?
The card fits a PCIe 3.0 x16 slot but will operate at PCIe 3.0 speeds instead of the rated 4.0 bandwidth.
New to this? Buying server hardware with crypto — what to check, what it costs to pay, and what is never asked for.
Also in this aisle
People compared these
Same shelf, same checkout — eight coins and a 30-minute rate lock.
Lenovo 4X60N04886 Nvidia Quadro P5000 16GB GDDR5X ai accelerator card
- GDDR5
NVIDIA H200 NVL AI Accelerator Card 141GB HBM3e PCIe 5.0 x16 900-21010-0040-000
- PCI Express 5.0 x16
- NVIDIA



