← All hardware
Intel Gaudi 3 logo
hardware Data Center AI Accelerators

Intel Gaudi 3

by Intel

An open, cost-effective AI accelerator priced well below NVIDIA.

Pros

  • Roughly half the price of H100-class GPUs
  • Open Ethernet networking, no proprietary fabric
  • 60% more memory than H100 SXM

Cons

  • Lower raw performance than H100/Blackwell
  • Software ecosystem trails CUDA and ROCm
  • Smaller deployment base and tooling

✓ Where it shines / best for

  • Enterprises wanting a lower-cost alternative to NVIDIA for training/inference
  • Open Ethernet-based cluster scale-out without proprietary fabric
  • On-prem and OEM-server AI deployments

✕ Not the best fit for

  • CUDA-dependent codebases that can't migrate
  • Absolute peak-performance frontier training vs latest NVIDIA
  • Buyers needing the largest software ecosystem and tooling breadth

Features

  • ✓ Data-center scale
  • ✓ API access
  • ✓ Inference
  • ✓ Training
  • ✓ FP8
  • ✓ Ethernet
  • ✓ PCIe
  • ✓ HBM2E
  • ✓ OAM
  • ✓ On-device / offline

Pricing

PlanPriceBillingNotes
PCIe / OAM accelerator~$15,000–$16,000per card (est.)Reported list/OEM pricing per accelerator; sold through OEMs (Dell, Supermicro) and IBM Cloud. Positioned well below comparable NVIDIA.
Universal Baseboard (8-card)~$125,000per UBB (est.)8-accelerator HL-325L baseboard system pricing reported by Intel.
Cloud (IBM Cloud)Usage-basedper hourAvailable as a cloud instance on IBM Cloud; rate by contract/instance.

Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.

Specifications

sram96 MB on-die, 19.2 TB/s
power600W TDP
memory128 GB HBM2e, 3.7 TB/s
networking24x 200 GbE RoCE
architectureGaudi 3 (Intel), dual-chiplet
fp8_bf16_performance1.8 PFLOPS (1,835 TFLOPS BF16)
Sponsored

A full review is being generated for this product and will appear here shortly.

Compare with

Compare
Compare