Brand: AMD | Category: GPUs
SKU: AMD-100-300000001 | Part #: 100-300000001 | MPN: 100-300000001
Contact for Pricing — Request a Quote
For AI training and large-language model inference workloads, the AMD Instinct MI300A APU Accelerator combines integrated CPU and GPU compute in a single package, built on the advanced CDNA 3 architecture. This unified design eliminates traditional CPU-GPU data movement bottlenecks, delivering exceptional performance for demanding enterprise AI applications. The MI300A integrates 24 AMD EPYC Zen 4 CPU cores with 228 CDNA 3 GPU compute units, all sharing 128 GB of unified HBM3 memory with 5.3 TB/s memory bandwidth—a configuration purpose-built for complex model training and real-time inference at scale.
IT infrastructure teams and AI platform architects will find the MI300A's performance metrics compelling: peak FP64 matrix (MFMA) throughput reaches 122.6 TFLOPS, while BF16 matrix operations deliver up to 1,307.4 TFLOPS (scaling to 2,614.9 TFLOPS with sparsity enabled). FP8 matrix workloads achieve 2,614.9 TFLOPS, extending to 5,229.8 TFLOPS with sparsity. The accelerator ships in OAM form factor with liquid-cooling support for data center thermal environments, connects via PCIe Gen 5 host interface, and runs AMD ROCm software with native support for PyTorch and TensorFlow frameworks. Part number 100-300000001 represents AMD's expanded availability push for this strategic accelerator. With full AMD ROCm stack compatibility—including HIP, OpenCL, rocBLAS, MIOpen, and rocFFT—the MI300A integrates seamlessly into existing high-performance computing and machine learning pipelines.
| Brand | AMD |
| Category | GPUs |
| SKU | AMD-100-300000001 |
| Part Number | 100-300000001 |
| Condition | New |
| Capacity | 128GB |
| Manufacturer Part Number | 100-300000001 |
| Product Name | AMD Instinct MI300A APU Accelerator |
| Architecture | CDNA 3 |
| Device Type | APU (Accelerated Processing Unit) — integrated CPU + GPU |
| CPU Cores | 24 (AMD EPYC Zen 4) |
| GPU Compute Units | 228 CDNA 3 Compute Units |
| Memory | 128 GB HBM3 (unified CPU/GPU) |
| Memory Bandwidth | 5.3 TB/s |
| Peak FP64 Vector Throughput | 61.3 TFLOPS |
| Peak FP64 Matrix (MFMA) Throughput | 122.6 TFLOPS |
| Peak BF16 Matrix (MFMA) Throughput | 1,307.4 TFLOPS (with sparsity: 2,614.9 TFLOPS) |
| Peak FP8 Matrix (MFMA) Throughput | 2,614.9 TFLOPS (with sparsity: 5,229.8 TFLOPS) |
| Form Factor | OAM (OCP Accelerator Module) |
| Thermal Design | Liquid-cooled (data center managed thermal environment) |
| Interconnect | AMD Infinity Fabric (die-to-die); PCIe Gen 5 host interface |
| Software Platform | AMD ROCm (HIP, OpenCL, rocBLAS, MIOpen, rocFFT) |
| AI Framework Support | PyTorch, TensorFlow (via ROCm) |
| Availability | In stock at Omnixon Global; RFQ response within 48 hours |
The AMD Instinct MI300A APU Accelerator (Expanded OEM Availability) accelerates AI/ML training, inference, scientific HPC, and virtualization (vGPU) workloads. Typical deployments include LLM training clusters, computer-vision pipelines, financial risk modeling, and rendering farms.
Key specifications for the AMD Instinct MI300A APU Accelerator (Expanded OEM Availability): capacity 128GB; 3 years support; new condition; support 3 years standard (manufacturer); gpu support AMD MI300A; ai optimized Yes. Manufacturer part number 100-300000001. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
The AMD Instinct MI300A APU Accelerator (Expanded OEM Availability) requires a PCIe Gen4 or Gen5 x16 slot, server power adequate for the card's TDP, and CUDA/ROCm driver support in your hypervisor or bare-metal OS. Sales engineering will confirm chassis fit (1U/2U/4U), PCIe lane count, and PSU headroom before quoting.