- IBM Cloud incorporates the Gaudi 3 accelerator Intel, available in Frankfurt and Washington DC, with plans for Dallas.
- It offers competitive performance against GPUs of NVIDIA H100 and H200, at a lower cost.
- Several deployment options available: dedicated servers, containerized nodes, and Watsonx licensed nodes.
- Includes security based on hardware and support from the Red Hat and Wipro ecosystem.
IBM Cloud has officially announced the addition of the accelerators of Artificial Intelligence Intel Gaudi 3 to its infrastructure, providing businesses with a viable and cost-effective alternative to traditional NVIDIA GPUs. This collaboration marks a new chapter in the fight to dominate the cloud-based artificial intelligence market, in a context where the demand for powerful and cost-effective solutions continues to grow.
Gaudi 3 accelerators are now available in the IBM Cloud regions of Frankfurt (EU-DE) and Washington DC (US-East), and are planned for arrival in Dallas (US-South) in Q2025 XNUMX. This expansion responds to the growing need for scalability in workloads of Generative AI, offering greater geographic flexibility to global organizations.
Cost-effective alternative and competitive performance
One of the main attractions of Gaudi 3 on IBM Cloud is its attractive price compared to NVIDIA H100 and H200 GPUs, without compromising performance. Various tests conducted by Signal65 showed that the Gaudi 3 outperformed NVIDIA GPUs in specific scenarios, using language models such as IBM's Granite (8B) and Meta's Llama-3.1. Although results vary depending on the type of task, the initial data are encouraging.
Intel claims that the Gaudi 3 provides a 50% more inference performance and a 40% higher energy efficiency compared to the H100. In addition, its architecture allows for better scaling of deep learning tasks through programmable tensor cores, specialized matrix multiplication engines, and a significant memory bandwidth.
The chip is manufactured in a 5 nm process and incorporates 24 high-speed Ethernet ports, which facilitates rapid connectivity between nodes IA. It is also optimized for environments using frameworks like PyTorch and models built into Hugging Face, strengthening its compatibility for developers and data scientists.
Diversity of deployment options
IBM has designed several ways to implement the Gaudi 3 accelerator in its cloud, adapting to different business needs. Available options include:
- Dedicated server in IBM Cloud VPC: Ideal for those who need a private and secure environment with complete control over the software, including support for Red Hat Enterprise images Linux focused on AI.
- Worker node in container: Starting in Q2025 3, IBM will offer Gaudi XNUMX managed nodes for Red Hat OpenShift AI clusters, making it easier to integrate them into native containerized environments.
- Custom license with watsonx: Customers who already own watsonx.ai licenses will be able to use them with Gaudi 3 hardware, preserving their custom toolchain and workflows.
- Deployable Architectures (DAs): Pre-configured modules to accelerate adoption without excessive manual intervention, available in the second half of 2025.
These architectures will be available for both Virtual servers from the VPC environment as well as for solutions based on watsonx and OpenShift on IBM Cloud.
Ecosystem support: Red Hat and Wipro
Red Hat and Wipro have joined this initiative. While Red Hat emphasizes its commitment to open source and collaborative development, Wipro highlights the integration of Gaudi 3 into its center of excellence for generative AI and its previous collaboration with IBM. Although both partners do not provide deep technical details, their support gives solidity to the joint deployment.
Secure and performance-oriented infrastructure
In an area where the protection of sensitive data is critical, IBM has integrated Intel TDX (Trust Domain Extensions) into its VPC virtual servers. This technology allows processing confidential with the help of hardware, ensuring isolation and privacy at the virtual server level. This feature is particularly important in regulated industries, such as healthcare or finance.
Interoperability between Intel Gaudi 3 and IBM Cloud contributes to reduce operating costs by providing a scalable, versatile and secure environment. According to spokespersons for both companies, the collaboration seeks to make enterprise artificial intelligence more accessible and accelerate its global adoption.
Long-term strategic approach
This alliance between Intel and IBM is part of a multi-vendor strategy aimed at diversifying the offering of AI accelerators. While the initial integration may be limited in terms of widespread support or a development ecosystem compared to market leaders, it represents a realistic option for companies looking to break their dependence on NVIDIA.
The availability of multiple deployment paths, along with advantages in cost and energy consumption, reinforce the attractiveness of the Gaudi 3 as an alternative for AI workloads both in testing and production. Long-term success will depend largely on the market's reception and the degree of integration with existing tools.
The deployment of Intel Gaudi 3 on the IBM Cloud marks a strategic move that could change the dynamics of the cloud AI accelerator market. Thanks to its multiple implementation modalities, competitive performance compared to more popular options and more adjusted costs, it is presented as an attractive possibility for those organizations that seek efficiency without relinquishing control. The participation of key ecosystem players and the incorporation of security technologies reinforce its position as a solid alternative in artificial intelligence infrastructures.
Passionate writer about the world of bytes and technology in general. I love sharing my knowledge through writing, and that's what I'll do on this blog, show you all the most interesting things about gadgets, software, hardware, tech trends, and more. My goal is to help you navigate the digital world in a simple and entertaining way.