Accelerating AI Transformation: Intel® Gaudi® 3 AI Accelerators on IBM Cloud at IBM Think 2025

By Pauline Essalou and Megan Kuo, Intel Corporation

At Intel’s Vision conference in April, Intel and IBM announced the availability of Intel® Gaudi® 3 AI accelerators on IBM Cloud. This milestone marked the public cloud debut of Intel® Gaudi® 3 AI accelerators for production workloads and a significant leap forward for enterprises seeking to deploy and scale AI workloads more efficiently and cost-effectively. The groundbreaking solution, being showcased this week at IBM Think in Boston, underscores the long-term collaboration between Intel and IBM to deliver efficient, cost-effective solutions designed to scale and deploy AI workloads with enhanced performance, reliability, and speed.

With Intel Gaudi 3 AI accelerators now available on IBM Cloud, enterprises gain access to a robust platform to drive AI inferencing with cutting-edge technology. Built to provide high throughput and low latency, Intel Gaudi 3 accelerators are ideal for handling large-scale applications, including large language models (LLMs), generative AI (GenAI), and complex AI workloads. With 128 GB of high-bandwidth memory and an impressive 3.7 TB/s of memory bandwidth, Intel Gaudi 3 accelerators provide rapid data throughput, drastically reducing bottlenecks and enabling developers to process massive datasets at a much faster rate.

Performance benchmarks underscore the competitive advantage of Intel Gaudi 3 accelerators over traditional solutions. In scenarios running the Llama-3.1-405B-Instruct-FP8 model with large context sizes, Intel Gaudi 3 accelerators outperformed GPU competition by up to 36%.¹ Moreover, Intel Gaudi 3 accelerators offer exceptional price-performance value, achieving up to 92% ¹ more tokens per dollar compared to the competition, making it an incredibly cost-effective solution for enterprises scaling their AI operations. Dive deeper into the performance benchmarks here.

Beyond its raw performance, Intel Gaudi 3 accelerators on IBM Cloud provide flexible deployment options that cater to the diverse needs of enterprise AI infrastructures. Enterprises will be able to bring their own IBM watsonx.ai software license to deploy AI workloads on Intel Gaudi 3-powered virtual servers within IBM Cloud’s Virtual Private Cloud (VPC), giving them full control over their AI stack. To further streamline AI adoption, IBM Cloud will introduce Deployable Architectures (DAs) – pre-configured design modules that accelerate the deployment of AI solutions. These DAs, including options such as Intel AI for Enterprise AI, the OPEA Productivity Suite, and Red Hat OpenShift on IBM Cloud, will empower developers and IT teams to accelerate AI initiatives with minimal configuration and manual intervention, enhancing the overall speed to value.

By combining Intel’s next-gen hardware with IBM’s cloud infrastructure and AI tools, organizations can fully capitalize on the potential of AI, driving innovation and optimizing their return on investment.

For more details on the latest advancements on Intel Gaudi 3 accelerators on IBM Cloud, visit IBM Cloud GPU and AI Accelerator and transform your AI infrastructure today.

¹ Source: Signal65 Lab Insight Whitepaper - Intel Gaudi 3 AI Accelerator at Scale on IBM Cloud, Intel-commissioned study by Signal65, published April 22, 2025. Reported numbers are inferencing results on Intel Gaudi 3 vs. Nvidia H200. See source for workloads and configurations. Results may vary.