Brand: HPE | Category: GPUs
SKU: P48851-B21 | Part #: P48851-B21 | MPN: P48851-B21
Contact for Pricing — Request a Quote
The HPE AMD Instinct MI250X 128GB OAM GPU Accelerator (P48851-B21) is an OCP Accelerator Module v1.0 form factor GPU equipped with direct liquid cooling, engineered for high-performance compute and AI workloads requiring extreme throughput and precision flexibility. Built on AMD's CDNA2 architecture, this accelerator pairs a dual graphics compute die configuration via AMD Infinity Fabric to deliver 220 total compute units—110 per GCD. The 128 GB HBM2e memory subsystem achieves 3.2 TB/s aggregate bandwidth, ensuring data-hungry algorithms maintain steady compute utilization. Peak performance spans FP64 at 47.9 TFLOPS, FP64 Matrix Core operations at 95.7 TFLOPS, and mixed-precision workloads up to 383.0 TFLOPS in both FP16 and BF16 formats. Support for FP64, FP32, FP16, BF16, INT8, and INT4 precision modes grants flexibility across scientific simulation, deep learning training, and inference pipelines. The 500W thermal design power is managed by direct liquid cooling, while AMD Infinity Fabric enables seamless multi-GPU peer-to-peer communication. PCIe Gen4 x16 connectivity via the OAM baseboard provides host interface capability.
HPE's MI250X accelerator targets artificial intelligence infrastructure teams, HPC centers, and hyperscale research organizations deploying large-scale model training and scientific computing clusters. These buyers prioritize memory capacity and bandwidth density to minimize training iteration time and reduce total cost of ownership in mixed-precision environments. The modular OAM design integrates directly into OCP-compliant systems, simplifying deployment and reducing integration risk in standardized data center architectures.
The AMD ROCm software stack—supporting PyTorch, TensorFlow, JAX, and OpenMPI—ensures compatibility with mainstream frameworks and distributed training protocols. For pricing, availability, and system integration support for the HPE AMD Instinct MI250X with part number P48851-B21, contact Omnixon Global to submit your request for quotation.
| Brand | HPE |
| Category | GPUs |
| SKU | P48851-B21 |
| Part Number | P48851-B21 |
| Condition | New |
| Manufacturer Part Number | P48851-B21 |
| Product Name | HPE AMD Instinct MI250X 128GB OAM GPU Accelerator |
| GPU Architecture | AMD CDNA2 |
| GPU Die Configuration | Dual Graphics Compute Die (GCD) via AMD Infinity Fabric |
| Total Memory Capacity | 128 GB HBM2e |
| Memory Bandwidth | 3.2 TB/s aggregate |
| Peak FP64 Performance | 47.9 TFLOPS |
| Peak FP64 Matrix (FP64 Matrix Core) | 95.7 TFLOPS |
| Peak FP32 Performance | 47.9 TFLOPS |
| Peak FP16 Performance | 383.0 TFLOPS |
| Peak BF16 Performance | 383.0 TFLOPS |
| Compute Units | 220 (110 per GCD) |
| Form Factor | OAM (OCP Accelerator Module) v1.0 |
| Cooling | Direct Liquid Cooling (DLC) |
| Interconnect | AMD Infinity Fabric (inter-die and multi-GPU peer-to-peer) |
| PCIe Interface | PCIe Gen4 x16 (host interface via OAM baseboard) |
| Supported Software Stack | AMD ROCm (PyTorch, TensorFlow, JAX, OpenMPI compatible) |
| Precision Support | FP64, FP32, FP16, BF16, INT8, INT4 |
| Thermal Design Power (TDP) | 500W |
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.
Highly model-dependent. L40S / RTX-class: typically 3-6 weeks. H100/H200/B200 in SXM form factor: 12-16 weeks for whole-platform allocations. We quote genuine-channel ETAs only — no grey-market promises.
Yes — Omnixon stocks 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.
Yes. We hold genuine-channels for NVIDIA AI Enterprise software subscriptions. Add it to your RFQ and we quote node-aligned licensing along with the hardware.