Computex 2025: Intel Unveils BattleMatrix AI Workstations with New Arc B60 Pro GPUs

At the Computex 2025 event, Intel announced the upcoming release of its new line of workstation graphics cards, the Arc Pro B50 and B60, designed to enhance performance in CAD and AI inference applications. Utilizing the Battlemage architecture, which is also featured in Intel's gaming GPUs like the Arc B580 and B570, these new cards will be available starting in the third quarter of 2025.

The Arc Pro B50 model will be released as a compact, dual-width card requiring no external power due to its low thermal design power (TDP) of 70 watts. Intel has set a suggested retail price of $299 for this model, which represents a price increase of approximately $40 compared to the Arc Pro A50. In contrast, this price point is significantly lower than Nvidia's older RTX A1000 model, which retails for around $395.

In the case of the Arc Pro B60, Intel will collaborate with various partners for distribution, including Asrock and Sparkle, with additional designs from Gunnir, Maxsun, Onix, Senao, and Lanner already in the pipeline. The specific pricing for the B60 will be determined by these partners, with the cards expected to launch in Q3 2025. However, full software features, including virtualization and remote management capabilities, will not be available until the fourth quarter of the year. At launch, optimized drivers will be certified for use with specific design and engineering software, and Intel promises quarterly updates and validated support for Linux systems.

Intel anticipates that the new graphics cards will deliver significant performance improvements over previous models and competitors, primarily due to their ample memory capacity. For example, when running AI models that fit within the local graphics memory, the Arc Pro B60 is projected to outperform the RTX A2000 Ada with 16 GB of memory by as much as 60%. However, it is also noted that it may be slower than the GeForce RTX 5060 Ti with 16 GB of memory under certain conditions. When workloads exceed the 16 GB capacity, the B60's 24 GB will provide a distinct advantage, potentially leading to performance increases of up to 2.7 times compared to the A2000 Ada.

In addition to the B50 and B60 models, Intel aims to position the B60 as a key component in AI inference workstations. These workstations can support up to eight Arc Pro B60 cards, totaling 192 GB of graphics memory across four cards, which will facilitate more complex AI applications. With appropriate optimizations, larger models can be distributed across the graphics memory of multiple cards, allowing for a higher volume of simultaneous requests without exhausting memory resources.

Intel plans to provide a container-based software solution for large language models (LLMs), including tailored Linux drivers and optimized LLMs with appropriate quantization to ensure immediate usability upon deployment.

Pricing for the complete workstations featuring these graphics cards is anticipated to range between $5,000 and $10,000, although Intel has not yet disclosed specific launch dates for these systems.