IBM Cloud to Deploy Intel Gaudi 3 AI Chips in 2025

IBM Cloud is set to introduce Intel's Gaudi 3 AI accelerator chips to its platform early next year, marking a significant collaboration aimed at enhancing AI capabilities in cloud computing. This partnership comes at a crucial time as businesses increasingly seek efficient and cost-effective solutions for their artificial intelligence needs.

The Gaudi 3 chip, unveiled in December 2023, represents Intel's latest advancement in AI technology, designed to compete with leading products from Nvidia and AMD. This chip is built on a cutting-edge five-nanometer process and features a unique architecture that includes 64 Tensor Processing Cores (TPCs) and eight Matrix Multiplication Engines (MMEs). These components are optimized for a variety of AI tasks, enabling the chip to perform up to 1,835 trillion calculations per second (TFLOPS) when processing BF16 data, a format commonly used in AI applications.

Gaudi 3 vs Nvidia H100 GPUs

Intel's Gaudi 3 and Nvidia's H100 are two of the most powerful AI accelerators on the market, but they have some key differences:

Architecture

• Gaudi 3 is built on a 5nm process and features 64 Tensor Processing Cores (TPCs) and 8 Matrix Multiplication Engines (MMEs)
• H100 is built on a 4nm process and features 16,896 shading units, 528 tensor cores, and 528 texture mapping units

Performance

• Gaudi 3 delivers up to 1,835 TFLOPS of BF16 performance
• H100 SXM5 offers 1,920 GB/s memory bandwidth, while the PCIe version offers 1,280 GB/s

Memory

• Gaudi 3 has 128 GB of HBM2e memory capacity
• H100 SXM5 has 80 GB of HBM3 memory

Cost-Effectiveness

• Gaudi 3 is positioned as a more cost-effective alternative to H100
• Early benchmarks suggest Gaudi 3 offers impressive performance per dollar

Interconnect

• H100 is paired with the Grace CPU using Nvidia's ultra-fast chip-to-chip interconnect, delivering 900 GB/s of bandwidth

With IBM Cloud's adoption of Gaudi 3, customers will gain access to a powerful tool for both hybrid and on-premise environments. This integration is expected to enhance the functionality of IBM's Watsonx AI and data platform, providing businesses with the resources needed to develop and deploy advanced AI models more efficiently.

Intel's Gaudi 3 is positioned as a cost-effective alternative to Nvidia's H100 GPU, which has dominated the market. Early benchmarks suggest that Gaudi 3 offers impressive performance per dollar, a critical factor for enterprises looking to optimize their AI investments.

The Gaudi 3 chips will be available through IBM Cloud's Virtual Servers for VPC, enabling businesses to scale their AI workloads flexibly and efficiently. The integration of Gaudi 3 with IBM's existing infrastructure is intended to simplify the management of AI applications, providing enhanced visibility and control over computing resources.

As the demand for AI solutions continues to rise, this collaboration between IBM and Intel could reshape the landscape of cloud-based AI services. By offering Gaudi 3, IBM Cloud not only enhances its service portfolio but also positions itself as a key player in the competitive AI market, catering to enterprises seeking innovative and affordable computing solutions.