At Computex 2025 in Taiwan, Intel has unveiled a new lineup of graphics processing units (GPUs) and AI accelerators designed for professionals and developers.
Intel’s announcements include:
- New Intel Arc Pro B-Series GPUs: Intel announces Intel Arc Pro B60 and Intel Arc Pro B50 GPUs, expanding the Intel Arc Pro family with configurations tailored for AI inference and professional workstations.
- Intel Gaudi 3 AI Accelerators: Intel Gaudi 3 AI accelerators are now available in PCIe and rack scale systems, offering scalable, open solutions for enterprise and cloud AI inferencing.
- Intel AI Assistant Builder: Now publicly available on GitHub, Intel AI Assistant Builder enables developers to create local, purpose-built AI agents optimized for Intel platforms.
New Intel Arc Pro B60 and B50 GPUs – based on the Xe2 architecture – feature Intel Xe Matrix Extensions (XMX) AI cores and advanced ray tracing units, bringing high-performance capabilities to creators, developers and engineers.
Expanding Intel’s GPU lineup for professionals, new Intel Arc Pro B60 and Intel Arc Pro B50 GPUs are designed for demanding AI inference workloads and workstation applications. With AI-ready features, 24 gigabytes (GB) and 16GB of memory, respectively, and multi-GPU scalability, the Intel Arc Pro B-Series offers a versatile solution for creators, AI developers and professionals.
The GPUs are optimised for AEC (architecture, engineering, construction) and inference workstations, offering stability and performance through a wide range of independent software vendor (ISV) certifications and optimized software. The Intel Arc Pro B-Series GPUs are compatible with consumer and pro drivers on Windows. On Linux, the GPUs support a containerized software stack to simplify AI deployments, and will be progressively upgraded with features and optimizations. By combining high memory capacity with key software compatibility, the Intel Arc Pro B-Series offers creators and AI developers scalable, cost-effective solutions. Intel also revealed a configurable workstation-class Intel® Xeon®-based platform (code-named Project Battlematrix) that is designed to reduce AI developer friction points. It supports up to eight Intel Arc Pro B60 24GB GPUs to enable medium-size (up to 150 billion parameters) and accurate AI models with up to 192GB of video random-access memory.
“The Intel Arc Pro B-Series showcases Intel’s commitment in GPU technology and ecosystem partnerships,” says Vivian Lien, vice-president and GM of Client Graphics at Intel. “With Xe2 architecture’s advanced capabilities and a growing software ecosystem, the new Arc Pro GPUs deliver accessibility and scalability to small and medium-sized businesses that have been looking for targeted solutions.”
The Intel Arc Pro B60 GPU will be sampling from add-in board partners including but not limited to ASRock, Gunnir, Lanner, Maxsun, Onix, Senao and Sparkle starting in June 2025. The Intel Arc Pro B50 GPU will be available from Intel-authorized resellers starting in July 2025.
Intel Gaudi 3
Expanding on its AI strategy, Intel introduced new deployment options for Intel Gaudi 3 AI accelerators:
- Intel Gaudi 3 PCIe cards support scalable AI inferencing within existing data center server environments. Using AI models such as Llama, customers from small businesses to large enterprises benefit from the flexibility to run models ranging from Llama 3.1 8B to full-scale Llama 4 Scout or Maverick because of the scalable configurations available. Intel Gaudi 3 PCIe cards will be available in the second half of 2025.
- Intel Gaudi 3 rack scale system reference designs are built for flexibility and scale, supporting up to 64 accelerators per rack and offering 8.2 terabytes (TB) of high-bandwidth memory. An open, modular design helps avoid vendor lock-in. And with liquid cooling, these systems deliver strong performance and help keep total cost of ownership (TCO) in check.
The Intel Gaudi rack scale architecture is optimized for running large AI models and excels in real-time inferencing with low-latency performance. These configurations reinforce Intel’s commitment to open, flexible and secure AI infrastructure, supporting both custom and Open Compute Project (OCP) designs for cloud service providers (CSPs).
Intel AI Assistant Builder
Following its debut at CES 2025, Intel AI Assistant Builder – a lightweight, open software framework for building and running custom AI agents locally on Intel-based AI PCs – is now available in public beta release on GitHub. With new innovative solutions being unveiled at Computex by Acer and ASUS as examples, Intel AI Assistant Builder allows developers and partners to quickly create and deploy AI agents for their own organizations and direct customers.