Intel will launch Arc Pro B50 and B60 GPUs at Computex 2025. These GPUs enable advanced AI inference and workstation graphics with up to 24GB VRAM.
Intel has significantly strengthened its professional graphics portfolio by introducing new Intel Arc Pro GPUs: the Arc Pro B50 and Arc Pro B60. Announced at Computex, these GPUs are designed to meet the demanding requirements of modern graphics workstations and the rapidly growing field of enterprise AI inference. This launch builds on the success of Intel’s consumer-focused Arc Battlemage series, which received positive reviews for its gaming performance after debuting last December.
Intel Arc Pro B50
The Arc Pro B50 is a compact graphics card targeting entry-level and mainstream workstation markets. The B50 features 16GB of VRAM, a 2.7x increase over its predecessor, the Arc Pro A50, and double the memory of competing cards in its class. It also offers 128 XMX Engines, delivering up to 170 TOPS for AI-accelerated tasks, PCIe Gen 5 support for faster data access, and workstation-optimized Pro Drivers with ISV certifications for over 50 professional applications. According to Intel, the B50 delivers 2.3 times the performance and 2.4 times the efficiency of the A50. Compared to the NVIDIA RTX A4000, Intel claims a 50% performance advantage for the B50, offering twice the performance per dollar at a $299 price point.
Intel Arc Pro B60
The Arc Pro B60 is positioned as a more powerful and scalable solution, specifically tailored for local LLM inferencing. Professionals can use proprietary data and domain expertise to fine-tune custom AI models in-house. The B60 comes equipped with 24GB of VRAM for larger AI models, 160 XMX Engines providing 197 TOPS of AI performance, and a memory bandwidth of 456 GB/s double that of the B50.
It also supports multi-GPU configurations on Linux, allowing users to combine multiple B60 cards to significantly increase memory capacity for more complex models. Intel is working with partners to offer various form factors to meet diverse market needs, with additional features like SR-IOV and enhanced manageability for large-scale deployments planned for release in the fourth quarter.
Project Battlematrix
To support advanced AI workloads, Intel is introducing a new class of systems called “inference workstations.” These are designed to help small and medium-sized businesses run and formalize their specialized knowledge locally, using their own data securely without relying on third-party cloud services. Central to this strategy is “Project Battle Matrix,” an inference workstation platform that can accommodate up to eight Intel Arc Pro GPUs in a single system (4 PCIe cards, 2 GPUs per card). This high-performance setup can deliver up to 192GB of VRAM, enabling the execution of AI models with up to 70 billion parameters at full precision, all within a PCIe Gen 5 Xeon platform.
Intel has also outlined a clear roadmap for software features. The current phase focuses on enablement, providing foundational Windows and Linux drivers to partners. This will be followed by ISV certifications for key workstation applications. In the third quarter of 2025, Intel plans to roll out its initial containerized solution, featuring an optimized LLM inference stack, VLM staging capabilities, and basic GPU telemetry. Future updates will bring improved LLM performance and enhanced VLM serving. By the fourth quarter of 2025, Intel aims to introduce advanced features such as SR-IOV, support for virtual desktop environments, and comprehensive manageability tools for large-scale deployments.
This launch is backed by a robust ecosystem, with seven board partners set to announce their Arc Pro B60 GPU products during Computex. While Intel has not set a fixed MSRP for the B60, the company estimates its value at around $500. Complete systems featuring these GPUs are expected to range from $5,000 to $10,000, depending on design, features, and partner configurations.
Intel is currently providing samples of the Arc Pro B50 and B60 GPUs to customers and is working to enable the first platforms and end-to-end workflows. General availability is expected to begin in the coming quarter. The company plans to introduce support for new workloads and capabilities over time, with major feature updates including enhanced manageability for larger deployments scheduled for the fourth quarter.
As the price of high-VRAM, high-performance systems continues to fall, personal research, small-scale training, and private inference are becoming increasingly accessible. This release marks an exciting step forward for professionals and researchers alike.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
