Brand: AMD | Category: GPUs
SKU: AMD-100000001234X | Part #: 100-000001234X | MPN: 100-000001234X
Contact for Pricing — Request a Quote
The AMD Instinct MI325X Accelerator from AMD is enterprise-grade GPUs hardware built for data centers. AMD Instinct MI325X delivers 256GB of HBM3E memory with 6.0 TB/s bandwidth, purpose-built for frontier-scale AI training and inference workloads. Available in brand new condition. AMD gpus products are trusted by enterprises and data centers worldwide. Available from Omnixon Global. Contact our sales team to request a quote.
| Brand | AMD |
| Category | GPUs |
| SKU | AMD-100000001234X |
| Part Number | 100-000001234X |
| Condition | New |
| Product Family | AMD Instinct MI300 Series |
| Architecture | AMD CDNA 3 (chiplet-based compute + HBM3E) |
| GPU Memory | 256 GB HBM3E |
| Memory Bandwidth | 6.0 TB/s |
| FP8 Peak Throughput | 2,615 TOPS |
| FP16 Peak Throughput | 1,307 TFLOPS |
| BF16 Peak Throughput | 1,307 TFLOPS |
| FP32 Peak Throughput | 653 TFLOPS |
| FP64 Peak Throughput | 163 TFLOPS |
| Matrix Cores | Second-generation AMD Matrix Cores |
| Supported Precisions | FP8, FP16, BF16, FP32, FP64, INT8 |
| AMD Infinity Fabric Bandwidth | 896 GB/s bidirectional |
| Multi-GPU Interconnect | AMD Infinity Fabric with peer-to-peer NVLink-equivalent direct GPU communication |
| Form Factor | OCP Accelerator Module (OAM) / PCIe |
| Thermal Design Power (TDP) | 750W |
| Compute Dies | 3 CDNA 3 GPU compute chiplets + 12 HBM3E stacks |
| ECC Support | Full ECC on HBM3E memory |
| Software Stack | AMD ROCm open-source platform (HIP, MIOpen, rocBLAS, rocDNN) |
| Framework Support | PyTorch, TensorFlow, JAX, ONNX Runtime via ROCm |
| Security Features | Secure Boot, hardware-level memory isolation, multi-tenant capable |
| Launch Generation | Q4 2024 |
| Target Deployment | Data center AI training and inference, HPC clusters |