AMD Instinct MI250X 128GB HBM2e OAM Accelerator
Brand: AMD | Category: GPUs
SKU: 100-506020 | Part #: 100-506020 | MPN: 100-506020
Contact for Pricing — Request a Quote
About the AMD Instinct MI250X 128GB HBM2e OAM Accelerator
The AMD Instinct MI250X 128GB HBM2e OAM Accelerator delivers 128 GB of HBM2e memory with an aggregate bandwidth of 3.2 TB/s, engineered for large-scale AI, scientific computing, and high-performance data center workloads. Built on the AMD CDNA2 architecture, this dual-GPU accelerator (part number 100-506020) combines two Graphics Compute Dies with 220 total compute units and 14,080 stream processors to accelerate matrix and vector operations across multiple precision formats. The module supports FP64, FP32, FP16, BF16, INT8, and INT4 computation, making it suitable for diverse machine learning training, inference, and simulation tasks. With peak FP64 matrix performance of 47.9 TFLOPS and peak FP32 performance of 47.9 TFLOPS, paired with exceptional 383.0 TFLOPS in both FP16 and BF16 modes, this accelerator delivers the throughput required for transformer models and other compute-intensive applications.
The MI250X integrates AMD Infinity Fabric inter-die connectivity, enabling up to 200 GB/s peer bandwidth per direction between the dual Graphics Compute Dies for efficient multi-GPU coordination. The OAM v1.0 form factor ensures compatibility with Open Accelerator Module-compliant systems, while the PCIe Gen4 x16 host interface provides standard enterprise connectivity. Configured with 8 HBM2e stacks (4 per GCD), the module operates within a 500 W thermal design power envelope. AMD ROCm, the open-source software platform, enables developers and AI infrastructure teams to rapidly deploy applications across heterogeneous compute environments. Omnixon Global supplies this enterprise-grade accelerator for organizations scaling AI clusters and HPC deployments across the UAE and MENA region. Contact us to request a quote for the AMD Instinct MI250X.
Key Specifications
- Manufacturer: AMD
- Part Number: 100-506020
- Memory Capacity & Type: 128 GB HBM2e
- Memory Bandwidth: 3.2 TB/s aggregate
- Peak FP64 Performance: 47.9 TFLOPS (matrix), 23.9 TFLOPS (vector)
- Peak FP32 Performance: 47.9 TFLOPS
- Peak FP16 & BF16 Performance: 383.0 TFLOPS
- Form Factor: OAM (Open Accelerator Module)
- Host Interface: PCIe Gen4 x16
Technical Specifications
| Brand | AMD |
| Category | GPUs |
| SKU | 100-506020 |
| Part Number | 100-506020 |
| Condition | New |
| Manufacturer Part Number | 100-506020 |
| Architecture | AMD CDNA2 |
| Form Factor | OAM (Open Accelerator Module) |
| Compute Dies per Module | 2 (dual Graphics Compute Die) |
| Total Compute Units | 220 (110 per GCD) |
| Stream Processors | 14,080 total (7,040 per GCD) |
| Peak FP64 Performance | 47.9 TFLOPS (matrix), 23.9 TFLOPS (vector) |
| Peak FP32 Performance | 47.9 TFLOPS |
| Peak FP16 Performance | 383.0 TFLOPS |
| Peak BF16 Performance | 383.0 TFLOPS |
| Memory Capacity | 128 GB HBM2e |
| Memory Bandwidth | 3.2 TB/s aggregate |
| Memory Stack Configuration | 8 HBM2e stacks (4 per GCD) |
| Host Interface | PCIe Gen4 x16 |
| Inter-Die Interconnect | AMD Infinity Fabric (up to 200 GB/s peer bandwidth per direction between GCDs) |
| Thermal Design Power (TDP) | 500W |
| Supported Precision Formats | FP64, FP32, FP16, BF16, INT8, INT4 |
| Software Platform | AMD ROCm (open-source) |
| OAM Specification Compliance | OAM v1.0 |
Frequently Asked Questions about AMD Instinct MI250X 128GB HBM2e OAM Accelerator
What server platforms accept the AMD Instinct MI250X 128GB HBM2e OAM Accelerator?
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.
How long is the lead time on AI GPUs?
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.
Do you supply matched networking (Quantum InfiniBand / Spectrum-X)?
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.
Can you help with NVIDIA AI Enterprise licensing?
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.