Brand: Lenovo | Category: GPUs
SKU: 4X67A84530 | Part #: 4X67A84530 | MPN: 4X67A84530
Contact for Pricing — Request a Quote
The Intel Gaudi 2 HL-225H Mezzanine GPU for ThinkSystem SXM Configuration (Lenovo part 4X67A84530) is a high-performance AI accelerator built on Intel's second-generation Gaudi 2 architecture, designed specifically for deep learning training and inference at scale. The HL-225H integrates 96 GB of HBM2e memory across a 3.7 TB/s memory bandwidth envelope, delivering the substantial throughput required for large language model training, computer vision pipelines, and generative AI workloads. Its mezzanine form factor is engineered for direct integration into Lenovo ThinkSystem SXM-based server platforms, enabling dense, high-efficiency AI compute deployments within enterprise datacenter environments.
The Gaudi 2 processor features 24 fully programmable Tensor Processor Cores (TPCs) and a Matrix Multiplication Engine (MME) optimized for mixed-precision training using BF16 and FP32 formats. The accelerator includes 24 integrated 100 GbE RDMA-capable network ports built directly onto the die, eliminating the need for external networking components and enabling scale-out cluster connectivity with low latency and high bandwidth for distributed training jobs. This on-chip fabric architecture supports RoCEv2 and is compatible with Gaudi 2 scale-out topologies using standard Ethernet infrastructure.
As a Lenovo-validated component for ThinkSystem SXM configurations, the HL-225H is subject to rigorous platform-level qualification, ensuring thermal, electrical, and firmware compatibility within supported ThinkSystem server chassis. Enterprises deploying AI infrastructure in UAE, GCC, EMEA, and APAC regions benefit from Lenovo's global supply chain reach and ThinkSystem platform consistency. The accelerator is well-suited for organizations seeking an alternative high-performance AI compute path alongside or in place of traditional GPU ecosystems, with full software support through the Intel Gaudi software stack including Habana SynapseAI and integration with popular ML frameworks such as PyTorch and TensorFlow.
| Manufacturer | Lenovo |
| Manufacturer Part Number | 4X67A84530 |
| Accelerator Model | Intel Gaudi 2 HL-225H |
| Form Factor | Mezzanine (SXM) |
| Architecture | Intel Gaudi 2 |
| Tensor Processor Cores (TPCs) | 24 |
| Memory Type | HBM2e |
| Memory Capacity | 96 GB |
| Memory Bandwidth | 3.7 TB/s |
| On-Chip Network Ports | 24 x 100 GbE RDMA (RoCEv2) |
| Supported Precision Formats | BF16, FP32, INT16, INT8 |
| Network Protocol Support | RoCEv2 (RDMA over Converged Ethernet v2) |
| Platform Compatibility | Lenovo ThinkSystem SXM Configuration |
| Software Stack | Intel Habana SynapseAI |
| Framework Support | PyTorch, TensorFlow |
| Target Workloads | AI Training, Deep Learning Inference, LLM, Generative AI |
| Category | AI Accelerator / Mezzanine GPU |
Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.
| Brand | Lenovo |
| Category | GPUs |
| SKU | 4X67A84530 |
| Part Number | 4X67A84530 |
| Condition | New |
| Manufacturer Part Number | 4X67A84530 |
| Accelerator Model | Intel Gaudi 2 HL-225H |
| Form Factor | Mezzanine (SXM) |
| Architecture | Intel Gaudi 2 |
| Tensor Processor Cores (TPCs) | 24 |
| Memory Type | HBM2e |
| Memory Capacity | 96 GB |
| Memory Bandwidth | 3.7 TB/s |
| On-Chip Network Ports | 24 x 100 GbE RDMA (RoCEv2) |
| Supported Precision Formats | BF16, FP32, INT16, INT8 |
| Network Protocol Support | RoCEv2 (RDMA over Converged Ethernet v2) |
| Platform Compatibility | Lenovo ThinkSystem SXM Configuration |
| Software Stack | Intel Habana SynapseAI |
| Framework Support | PyTorch, TensorFlow |
| Target Workloads | AI Training, Deep Learning Inference, LLM, Generative AI |
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.