Intel Gaudi 3 AI Accelerator OAM
Brand: Intel | Category: GPUs
SKU: HLS-GAUDI3-OAM | Part #: HLS-GAUDI3-OAM | MPN: HLS-GAUDI3-OAM
Contact for Pricing — Request a Quote
About the Intel Gaudi 3 AI Accelerator OAM
Large-language-model training and inference workloads demand the architectural efficiency of purpose-built AI accelerators, and the Intel Gaudi 3 AI Accelerator OAM delivers exactly that through a 5nm processor design optimized for deep learning operations. This OAM (OCP Accelerator Module) form factor card is engineered for enterprise AI infrastructure teams seeking scalable, high-throughput compute density without the constraints of GPU-centric platforms. With 64 Tensor Processing Cores and 8 Matrix Multiplication Engines, the Gaudi 3 accelerator (part number HLS-GAUDI3-OAM) combines FP8 peak performance of 3,670 TFLOPS with BF16 compute capability of 1,835 TFLOPS, enabling efficient mixed-precision model training and inference at enterprise scale.
The accelerator integrates 96 GB of HBM2e memory paired with 3.7 TB/s memory bandwidth for rapid data movement, while 24 on-die 200 Gb/s Ethernet ports (RoCEv2) and 4.8 Tb/s of bidirectional scale-out bandwidth per accelerator enable seamless multi-card clustering. Up to 8 accelerators can be deployed per node within a 4U Universal Baseboard chassis, with direct Gaudi-to-Gaudi fabric interconnect for tightly coupled distributed training. Intel's comprehensive software stack—including the Intel SynapseAI SDK, PyTorch and TensorFlow integration, and Hugging Face Optimum Habana—reduces deployment complexity and accelerates time to production. The module supports FP8, BF16, FP16, TF32, and FP32 precisions, operates under 900W TDP with direct liquid cooling, and runs Intel Gaudi Software on Linux with native Docker and Kubernetes support for containerized environments. IT procurement teams and infrastructure architects can deploy this accelerator card via standard PCIe Gen 5 x16 host connectivity, simplifying integration into existing data center architectures. Contact Omnixon Global to request a formal quotation for the Intel Gaudi 3 AI Accelerator OAM and discuss your enterprise AI acceleration requirements.
Key Specifications
- Part Number: HLS-GAUDI3-OAM
- Manufacturer: Intel
- Form Factor: OAM (OCP Accelerator Module)
- Process Node: 5nm
- HBM2e Memory: 96 GB with 3.7 TB/s bandwidth
- Peak Compute (FP8 / BF16): 3,670 TFLOPS / 1,835 TFLOPS
- On-Die Network: 24
Technical Specifications
| Brand | Intel |
| Category | GPUs |
| SKU | HLS-GAUDI3-OAM |
| Part Number | HLS-GAUDI3-OAM |
| Condition | New |
| Product Line | Intel Gaudi 3 |
| Form Factor | OAM (OCP Accelerator Module) |
| Process Node | 5nm |
| BF16 Compute Performance | 1,835 TFLOPS |
| FP8 Compute Performance | 3,670 TFLOPS |
| Tensor Processing Cores (TPCs) | 64 |
| Matrix Multiplication Engines (MMEs) | 8 |
| HBM Memory Capacity | 96 GB HBM2e |
| Memory Bandwidth | 3.7 TB/s |
| On-Die Network Ports | 24 × 200 Gb/s Ethernet (RoCEv2) |
| Scale-Out Bandwidth (per accelerator) | 4.8 Tb/s bidirectional |
| Scale-Up Interconnect | Gaudi-to-Gaudi direct fabric via OAM Universal Baseboard |
| Max Accelerators per Node | 8 (in 4U Universal Baseboard chassis) |
| PCIe Interface | PCIe Gen 5 x16 (host connectivity) |
| TDP | 900W |
| Supported Precisions | FP8, BF16, FP16, TF32, FP32 |
| Software Stack | Intel SynapseAI SDK; PyTorch, TensorFlow integration; Hugging Face Optimum Habana |
| Driver & Firmware | Intel Gaudi Software (Linux; container-ready with Docker/Kubernetes support) |
| Cooling | Direct Liquid Cooling (DLC) via OAM module interface |
| Launch Generation | Gaudi 3 (2024) |
Frequently Asked Questions about Intel Gaudi 3 AI Accelerator OAM
What does the Intel Gaudi 3 AI Accelerator OAM do?
The Intel Gaudi 3 AI Accelerator OAM 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.
What are the headline specs of the Intel Gaudi 3 AI Accelerator OAM?
Key specifications for the Intel Gaudi 3 AI Accelerator OAM: new condition; raid support H100; gpu support NVIDIA H100; ai optimized Yes. Manufacturer part number HLS-GAUDI3-OAM. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.
Is the Intel Gaudi 3 AI Accelerator OAM compatible with my infrastructure?
The Intel Gaudi 3 AI Accelerator OAM 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.