At the end of last year, the company announced the launch of the independent graphics card code-named "Battlemage" belonging to the Arc brand B series, which was launchedArc B580 and Arc B570Later, at Comoutex 2025, Intel launched the Arc Pro B299, designed for workstation display acceleration and priced at US$50, and the Arc Pro B60, designed for workstation inference acceleration, to meet the needs of AI accelerated computing applications. Both products are claimed to provide more cost-effective AI computing performance in the US$500 price range.
Intel's Arc Pro B50 and Arc Pro B60 products are positioned to meet the needs of most professional work applications and current mainstream artificial intelligence inference computing. They are touted to be able to unleash greater application value with extremely high cost-effectiveness, while also featuring secure deployment, easy configuration, and extensible usage. They are also touted to be highly compatible with mainstream software and operating systems.
Arc Pro B50, Arc Pro B60 Key Specifications
Among them, the Arc Pro B50 uses 128 XMX engines, 16GB of display memory and PCI Gen 5 connection specifications, and corresponds to a data transmission bandwidth of 224GB per second, with a peak performance of up to 170 TOPS. The graphics card operates with a power consumption of 70W, thereby corresponding to higher energy utilization efficiency. The graphics card itself is built in a compact dual-slot design, and is also compatible with Windows and Linux operating systems, and has been certified for compatibility by third-party software companies.
As for the Arc Pro B60, it uses 160 XMX engines and 24GB of display memory. It also adopts the PCI Gen 5 connection specification and corresponds to a data transmission bandwidth of 456GB per second. The peak performance can reach 197 TOPS, and the graphics card operating power consumption is between 120W and 200W. It is claimed that it can operate in series with multiple graphics cards in a Linux computing environment. As for the graphics card, it will vary depending on the design of different brands and manufacturers. In addition, it supports SR-IOV (Single-Root I/O Virtualization) graphics card computing resource virtualization, and also has highly manageable features. It has also been certified compatible by third-party software manufacturers.
Compared with competing products, the previous generation of products has a higher cost-effectiveness in accelerated computing and artificial intelligence inference.
Compared to NVIDIA's RTX A8, which uses Ampere display architecture and is equipped with 1000GB of display memory, Intel's Arc Pro B16, which uses 50GB of memory, can handle larger-scale computing projects. At the same time, it accelerates computing performance through the Arc Pro driver, and can achieve higher execution efficiency in Dassault Systèmes' 3D CAD software Catia and SOLIDWORKS, as well as PTC's 3D CAD software Creo, or in energy and drug simulations. In addition, by switching to PCI Gen 5 connection specifications, it can also achieve higher performance thanPreviously launched Arc Pro A50, A60Higher data transmission efficiency.
Compared to NVIDIA's RTX 16 and GeForce RTX 2000 Ti, which use the Ada display architecture and are equipped with 5060GB of video memory, Intel boasts that the Arc Pro B24, which uses a 60GB memory design, can handle larger-scale AI model inference operations, including DeepSeek-R1, Microsoft's Phi 4, Alibaba's Tongyi Qianwen Qwen 2.5, or Meta's Llama 3, and can achieve up to 2.7 times the execution efficiency improvement. In particular, the higher the video memory capacity, the more AI models with larger parameter scales can be loaded, which means that the Arc Pro B60 can handle larger-scale AI model inference operations under limited budgets.
The Arc Pro B60, custom-designed by a brand, is not only used for inference, but can also be used for deployment training.
In addition to performing inference on AI models, Intel states that the Arc Pro B60 can also be used for AI model training, depending on the specific characteristics of the AI model. Currently, the Arc Pro B60 will be launched by ASRock, GUNNIR, MAXSUN, SPARKLE, ONIX, SENAO, and Lanner.
Expanding Arc Pro inference performance with Project Battlematrix
In addition, Intel also proposed the "Project Battlematrix" inference workstation platform design, which can connect up to 8 Arc Pro graphics cards in series, allowing the display memory capacity to reach up to 192GB, thereby corresponding to the operation of artificial intelligence models with a parameter scale of more than 700 billion groups. It is claimed that the operation of artificial intelligence models can achieve compatibility optimization under the Linux software stack, and software compatibility verification is completed in advance, and software application functions can be continuously updated after deployment.
Arc Pro B50 and Arc Pro B60 are expected to be launched in the third quarter, with more application features to be added in the fourth quarter.
The newly launched Arc Pro B50 and Arc Pro B60 are currently being provided to partners for testing and are expected to be launched in the third quarter of this year, with plans to open more application functions in the fourth quarter of this year.
