Computex 2025: Intel Arc Pro B-Series: New GPUs for workstation and AI inference at a glance

The Intel® Arc™ Pro B50

With a typical power consumption of 70 watts, the Intel® Arc™ Pro B50 is specifically designed for use in energy-efficient and space-optimized workstation environments. Its mechanical design follows a low-profile half-height, half-length (HHHL) approach and occupies two slots, making the card particularly suitable for compact desktop systems, small form factor (SFF) PCs and 2U rack workstations.

Thanks to the low power consumption, no additional power supply via external PCIe power connectors is required. Power is supplied entirely via the PCIe slot (maximum 75 W), which makes integration into systems with a limited power supply or cooling budget much easier. This can be particularly advantageous in office environments or for densely populated computers with several expansion cards.

Despite the compact design, Intel has not sacrificed any functional features: The card offers four Mini-DisplayPort 2.1 outputs with support for resolutions up to 8K (2× 7680×4320 @ 60 Hz), which also makes it predestined for multi-monitor setups or high-resolution visualizations. In professional use – for CAD, 3D modeling or video editing, for example – the combination of low heat dissipation, mechanical compatibility and ISV-certified driver stack allows easy integration into existing workflows.

The Intel® Arc™ Pro B60

The Intel® Arc™ Pro B60 is significantly more powerful than the B50 and therefore has a higher energy requirement. The specified power consumption is between 120 and 200 watts, depending on the implementation and partner design. This range indicates the need for active and potentially large-volume cooling solutions, which can vary depending on the OEM design. An additional power supply via at least one 6-pin or 8-pin PCIe connector is also a prerequisite in this power range, although Intel deliberately does not provide a standardized solution in the specification.

Intel does not offer a reference design for the B60. Instead, the form factor, cooling system and connectors are left entirely to the board partners. This results in different mechanical and thermal designs – from workstation graphics cards with an active cooling solution in a dual or triple-slot design to passively cooled variants for rack use in the server sector. Thanks to its openness to different designs, the B60 is designed to be integrated as flexibly as possible into existing OEM systems, particularly in environments with scalable GPU arrays or multi-GPU setups under Linux.

The design for variable designs fits in with the B60’s focus as an inference GPU: with its 24 GB GDDR6 memory, 456 GB/s bandwidth and support for SR-IOV, Linux containers and parallel multi-GPU operation, the card is primarily designed for compute-intensive tasks that require a corresponding cooling and supply infrastructure. The B60 is therefore geared more towards the requirements of professional deep learning and high-performance computing environments than classic single-user workstations.

Software support and drivers

Both models use a combined platform of consumer and professional drivers. For workstation use, explicit reference is made to quarterly Pro drivers for Windows. Linux is officially supported, whereby the B60 has been validated in particular for multi-GPU environments with LLM containers (keyword: “Project Battlematrix”).

The software stack includes, among other things:

• OpenCL 3.0, Vulkan 1.3, DirectX 12 Ultimate

• Intel oneAPI, OpenVINO, Level Zero

• Support for SR-IOV and vLLM serving

Market positioning and price-performance ratio compared to NVIDIA

In direct comparison with NVIDIA’s entry-level models for the professional market, in particular the RTX A1000 with 8 GB GDDR6 and the newer RTX A2000 Ada with 16 GB, Intel’s Arc Pro B series is specifically positioned in a price-sensitive niche. Both new Intel models, especially the Arc Pro B50, offer a significantly higher ratio of computing power, memory expansion and range of functions per US dollar at a comparable or lower price.

The “Suggested End User Price” (SEP) quoted by Intel for the Arc Pro B50 is 299 US dollars. According to Intel, this undercuts the NVIDIA RTX A1000, which is listed at around 409 US dollars, as well as the RTX A2000 Ada, whose market price was quoted at around 375 US dollars on May 13, 2025. Despite the lower entry-level price, the B50 offers 16 GB GDDR6 VRAM with 224 GB/s bandwidth, dedicated hardware acceleration for ray tracing and AI inference as well as four Mini-DisplayPort outputs – features that are only available to a limited extent or not at all from NVIDIA in this price range.

The B60 is the more powerful model in the series and has 24 GB GDDR6 with 456 GB/s bandwidth and 160 XMX cores for AI calculations. Although no official SEP has been named, the performance class on offer means that it is likely to compete in a segment that is usually reserved for the RTX A4000 or higher at NVIDIA. It is interesting to note that in the official benchmarks, the B60 itself achieved a higher token rate in several LLM workloads compared to the GeForce RTX 5060 Ti with 16 GB Ada memory and the RTX 2000 Ada (also 16 GB). This indicates that the combination of more memory and more efficient access (e.g. through PCIe Gen5 x8) is advantageous for memory-heavy inference processes.

In the presentation, Intel also emphasizes the price-performance ratio through standardized benchmarks. The SPECviewperf results shown (4K) were divided by the respective product price, which should make the “performance per dollar” visible. In this presentation, the B50 is ahead of both the A50 and A60 as well as clearly ahead of the NVIDIA RTX A1000. For professional users with a limited budget, this results in a potentially attractive ratio of cost, memory size, certifications and functional features.

The B-series is thus very consciously positioned as an answer to the frequently criticized price structure of professional NVIDIA solutions, without sacrificing features such as ISV certification, ECC-free memory, AI acceleration or driver support for Windows and Linux. In particular, the high memory combined with a moderate TDP makes the Arc Pro B50 an alternative for smaller workstations and the B60 a candidate for local LLM inference and container deployments.

Conclusion

With the Arc Pro B series, Intel is expanding its range of professional GPUs with two clearly differentiated models. The B50 is aimed at classic CAD and DCC users with compact workstation hardware and a limited energy budget. The B60 meets the increasing demand for local LLM inferencing with high scalability and VRAM capacity. Both cards rely on ISV certifications and support modern APIs and AI libraries, with the interaction between the Pro driver and the latest hardware forming the main features. The market launch will be staggered with sample deliveries in the third quarter of 2025 and broad availability in the fourth quarter.