Skip to main content

Intel Gaudi (Habana)

Vendor: Intel Habana

Category: ASIC Dedicated Accelerator

Architecture: Gaudi / Xe HPC

Introduction

Intel Gaudi AI accelerator (formerly Habana Labs), purpose-built for AI training and inference optimization. Gaudi 2 and Gaudi 3 excel at LLM training, support PyTorch and TensorFlow, with competitive pricing.

Specifications

ModelComputeMemoryInterfaceTDPProcess
Gaudi 31,678 TFLOPS (FP8) / 1,678 (BF16)128GB HBM2e (3.7 TB/s)OAM + 24x 200GbE900W5nm
Gaudi 2865 TFLOPS (FP8) / 432 (BF16)96GB HBM2e (2.45 TB/s)OAM + 24x 100GbE600W7nm

Official Website

Visit Official Website

Driver Downloads

Linux

OS Support

WindowsLinuxmacOSAndroid

Version History

VersionRelease DateDescription
SynapseAI 1.182025-Q1Full Gaudi 3 support
SynapseAI 1.162024-Q2PyTorch 2.1 native integration

Performance Benchmarks

ModelTaskPerformance Metric
Gaudi 3 × 8GPT-3 175B Training~1.2 days (estimated)
Gaudi 3Llama 2 70B Inference~100 tok/s (FP8)
Gaudi 2 × 8BERT-Large Training~90% H100 efficiency

Pricing

ModelReference PriceNotes
Gaudi 3Contact vendorAvailable via Intel Developer Cloud or ODMs
Gaudi 2$8,000-12,000Market price below H100

Quick Installation

Linux (Ubuntu 22.04)

# 1. Install Habana driver
sudo ./habana-driver-*.run --install

# 2. Install SynapseAI SDK
sudo ./SynapseAI-*.run --install

# 3. Verify
hl-smi

Gaudi 2 can also be trialled for free via Intel Developer Cloud.

Code Examples

Python (PyTorch on HPU)

import torch
import habana_frameworks.torch.core as htcore

# Use HPU backend (API consistent with CUDA)
device = torch.device("hpu")
x = torch.randn(1024, 1024, device=device)
y = torch.matmul(x, x)
print(f"HPU matrix multiply: {y.shape}")

# Habana Graph mode (similar to CUDA Graphs)
htcore.hpu.graphs()

Architecture Highlights

  • Gaudi Architecture: AI accelerator purpose-built for large model training, integrated 24x RoCE 2.0 network interfaces, supporting direct large-cluster interconnect
  • SynapseAI: Intel AI software stack, native PyTorch/TensorFlow support
  • Competitive Pricing: Gaudi 2 priced at roughly 1/3 of H100, suitable for budget-constrained training scenarios

Model Compatibility

Model/FrameworkSupportNotes
PyTorch✅ NativeHPU backend, CUDA-API compatible
TensorFlowHabana backend
JAX⚠️Experimental support
Llama / Qwen and similar LLMsDeepSpeed / FSDP both supported
Stable DiffusionHPU backend

Large-Scale Cluster Deployments

Based on global AI supercomputing cluster statistics, Intel Gaudi has accumulated over 4,000 chips deployed across 1 publicly disclosed clusters.

Chip Model Statistics

Chip ModelTotal DeployedCluster Count
Intel Habana Gaudi24,0001

Notable Deployment Clusters Top 10

#Cluster NameTotal ChipsChip ModelOperator
1Intel Stability Gaudi 24,000Intel Habana Gaudi2 ×4,000Intel, United States of America

If you're evaluating alternatives, the following products may also fit your scenario: