Brand: AMD | Category: GPUs
SKU: 100-000000846 | Part #: 100-000000846 | MPN: 100-000000846
Contact for Pricing — Request a Quote
PCIe Gen 5 and 550 W TDP are the primary considerations when integrating the AMD Instinct MI300A 128GB HBM3 APU OAM Accelerator into enterprise infrastructure—these specifications determine server chassis compatibility and power supply dimensioning. This unified-memory APU combines AMD's CDNA 3 GPU architecture with 24 AMD EPYC Zen 4 CPU cores, delivering a single coherent memory pool of 128 GB HBM3 that eliminates the traditional discrete CPU/GPU memory split. The result is a purpose-built accelerator for workloads requiring high-bandwidth, low-latency data movement between compute and memory.
The MI300A OAM (OCP Accelerator Module) form factor integrates 228 CDNA 3 compute units with aggregate memory bandwidth of 5.3 TB/s, supporting a comprehensive precision palette including FP64, FP32, FP16, BF16, INT8, and FP8. For AI infrastructure teams and HPC procurement specialists evaluating dense, coherent-memory compute platforms, this accelerator delivers matrix and vector performance across multiple numerical formats. AMD ROCm software enables HIP, OpenMP offload, and OpenCL programming models. Part number 100-000000846 is the exact SKU for this configuration. Request a quote and our team will respond within 48 hours.
| Brand | AMD |
| Category | GPUs |
| SKU | 100-000000846 |
| Part Number | 100-000000846 |
| Condition | New |
| Manufacturer Part Number | 100-000000846 |
| Product Name | AMD Instinct MI300A APU OAM Accelerator |
| Architecture | AMD CDNA 3 (GPU) + AMD Zen 4 (CPU) |
| Form Factor | OAM (OCP Accelerator Module) |
| CPU Cores | 24 x AMD EPYC (Zen 4) |
| GPU Compute Units | 228 CDNA 3 Compute Units |
| Unified Memory Capacity | 128 GB HBM3 |
| Memory Type | HBM3 |
| Memory Bandwidth | 5.3 TB/s (aggregate) |
| Memory Architecture | Unified CPU+GPU shared memory pool (no discrete CPU/GPU memory split) |
| FP64 Vector Performance | Up to 61.3 TFLOPS |
| FP64 Matrix Performance | Up to 122.6 TFLOPS |
| FP32 Performance | Up to 122.6 TFLOPS |
| BF16 / FP16 Matrix Performance | Up to 1,457.9 TFLOPS (sparse) |
| FP8 Matrix Performance | Up to 2,915.9 TFLOPS (sparse) |
| Supported Precisions | FP64, FP32, FP16, BF16, INT8, FP8 |
| TDP | 550 W |
| Host Interface | PCIe Gen 5 |
| Software Platform | AMD ROCm (HIP, OpenMP offload, OpenCL) |
| Target Platform | OCP Universal Baseboard (UBB) systems |
Reference servers include Dell PowerEdge XE9680 / XE9712, HPE Cray XD670, Lenovo ThinkSystem SR685a / SR675 V3, Supermicro AS-A21GE / SYS-821GE, Gigabyte G593 / G894, ASUS ESC. Share your target platform in the RFQ and we will confirm chassis-to-GPU compatibility and recommended NIC pairing.
Yes — Omnixon supplies the full NVIDIA networking lineup (Quantum-2 / Quantum-X InfiniBand, Spectrum-X Ethernet, ConnectX NICs, BlueField DPUs) so we can quote a complete training-cluster BOM, not just the GPUs.
