SERVCHIP

Interactive view

ARCHITECTURE

Gaudi 2

SERIES

Gaudi 2

Memory

96 GB HBM2e

Bandwidth

~2.45 TB/s

TDP

450W (PCIe) / 600W (OAM)

In StockGaudi 2·Best for: Cost-effective deep-learning training with open-source framework support.

Intel Gaudi 2 AI Accelerator

Name: Intel Gaudi 2 AI Accelerator
Brand: Intel
SKU: intel-gaudi-2
Availability: InStock

Gaudi 2

|Intel Gaudi 2

Overview

The Intel Gaudi 2 AI accelerator is the predecessor to Gaudi 3, offering 96 GB of HBM2e memory with a proven architecture for deep-learning workloads. Designed for cost-conscious AI deployments, Gaudi 2 provides excellent price-performance for medium-to-large training clusters. Its open software platform and support for mainstream AI frameworks make it a versatile choice for research institutions, enterprises, and cloud providers building AI infrastructure.

Key Features

96 GB HBM2e memory
Efficient deep-learning training and inference
Open software stack (PyTorch, TensorFlow)
Scalable multi-node architecture
Integrated Ethernet networking
Cost-effective AI acceleration

Ideal Use Cases

Deep learning training (CV, NLP, speech)AI inference servingRecommendation enginesResearch and education AI clustersHybrid cloud AI workloads

Full Specifications

Complete technical specifications for the Gaudi 2.

Memory

Memory: 96 GB HBM2e
Memory Bandwidth: ~2.45 TB/s

Compute

Tensor Cores: N/A (dedicated matrix engines)
CUDA Cores: N/A
FP8 TFLOPS: N/A
FP16 TFLOPS: ~400 TFLOPS (BF16)
TF32 TFLOPS: ~200 TFLOPS
FP64 TFLOPS: N/A

Interconnect

Interconnect: Integrated Ethernet, up to 24 ports per node

Physical

TDP: 450W (PCIe) / 600W (OAM)
Form Factor: PCIe Gen4 / OAM
Cooling: High-performance air or liquid

Manufacturing

Launch Date: 2022
Manufacturing Process: TSMC 7nm

Ready to deploy the Gaudi 2?

Get a personalized quote within 24 hours. Our certified engineers will help you architect the optimal configuration for your workload.