Intel Gaudi 3 PCIe AI Accelerator Card
Brand: Intel | Category: GPUs
SKU: HGT3PCIE1GXPCIE | Part #: HGT3PCIE1GXPCIE | MPN: HGT3PCIE1GXPCIE
Contact for Pricing — Request a Quote
About the Intel Gaudi 3 PCIe AI Accelerator Card
The Intel Gaudi 3 PCIe AI Accelerator Card (part number HGT3PCIE1GXPCIE) operates at PCIe Gen 5 x16 with a 600W thermal design power—critical specifications that determine server chassis compatibility and power supply requirements before deployment. This 5nm process node accelerator delivers enterprise-grade AI inference and training capabilities, designed for AI infrastructure teams deploying large-scale generative AI and machine learning workloads across distributed data center environments.
Built on Intel's third-generation Gaudi architecture, the card features 96 GB of HBM2e memory with 3.7 TB/s bandwidth, enabling rapid tensor operations for demanding AI applications. Peak compute performance reaches 1,835 TFLOPS in BF16 and 3,670 TFLOPS in FP8, making it suitable for both training and inference at scale. The card integrates 24 × 100 GbE RDMA Ethernet ports via onboard network interface cards, supporting RoCEv2 for low-latency cluster interconnects. Active cooling is included, simplifying integration into production environments. Intel provides full software support through the Intel Gaudi Software Suite, with native compatibility for PyTorch and TensorFlow frameworks on Linux operating systems including Ubuntu and RHEL. The PCIe form factor enables rapid deployment into existing server infrastructure without architectural redesign. Network engineers and procurement teams will appreciate the unified compute and networking capabilities consolidated onto a single card, reducing enclosure real estate and simplifying topology management in high-density AI clusters.
Key Specifications
- Manufacturer: Intel
- Part Number: HGT3PCIE1GXPCIE
- Architecture: Gaudi 3 (3rd Generation Intel Gaudi)
- Process Node: 5nm
- Memory: 96 GB HBM2e with 3.7 TB/s bandwidth
- Peak Compute: 1,835 TFLOPS (BF16) / 3,670 TFLOPS (FP8)
- Host Interface: PCIe Gen 5 x16
- Network Ports: 24 × 100 GbE RDMA Ethernet ports with RoCEv2
- Thermal Design Power: 600W with active cooling included
- Supported Frameworks: PyTorch, TensorFlow via Intel Gaudi Software Suite
For competitive pricing and configuration assistance, contact Omnixon Global for your AI accelerator requirements.
Technical Specifications
| Brand | Intel |
| Category | GPUs |
| SKU | HGT3PCIE1GXPCIE |
| Part Number | HGT3PCIE1GXPCIE |
| Condition | New |
| Manufacturer Part Number | HGT3PCIE1GXPCIE |
| Product Family | Intel Gaudi 3 |
| Architecture | Gaudi 3 (3rd Generation Intel Gaudi) |
| Process Node | 5nm |
| Form Factor | PCIe Add-in Card |
| Host Interface | PCIe Gen 5 x16 |
| HBM Capacity | 96 GB HBM2e |
| Memory Bandwidth | 3.7 TB/s |
| BF16 Peak Compute | 1,835 TFLOPS |
| FP8 Peak Compute | 3,670 TFLOPS |
| On-Card Networking | 24 x 100 GbE RDMA Ethernet ports (via onboard NICs) |
| Network Protocol | RoCEv2 (RDMA over Converged Ethernet v2) |
| Thermal Design Power (TDP) | 600W |
| Cooling Solution | Active cooling (included) |
| Supported AI Frameworks | PyTorch, TensorFlow (via Intel Gaudi Software Suite) |
| Operating System Support | Linux (Ubuntu, RHEL) |
Frequently Asked Questions about Intel Gaudi 3 PCIe AI Accelerator Card
What does the Intel Gaudi 3 PCIe AI Accelerator Card do?
The Intel Gaudi 3 PCIe AI Accelerator Card 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.
Is the Intel Gaudi 3 PCIe AI Accelerator Card compatible with my infrastructure?
The Intel Gaudi 3 PCIe AI Accelerator Card 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.