ASUS launches UGen Arc Pro B70 with 32GB ECC VRAM for local AI

ASUS has officially announced the UGen Arc Pro B70, its first discrete graphics card based on Intel Arc, designed specifically for local artificial intelligence workloads. The card comes with 32GB of ECC GDDR6 memory and support for multi-GPU configurations that can total 128GB of VRAM.

What happened?

In a press release dated October 27, 2026, ASUS unveiled the UGen Intel Arc Pro B70 32GB, a workstation card under the new UGen brand, aimed at the local agentic AI market. The card uses Intel's BMG-G31 silicon, with 32 Xe2 cores and 256 XMX engines, offering up to 367 TOPS in INT8 and a maximum dynamic clock of 2800 MHz.

Memory is 32GB GDDR6 with ECC on a 256-bit interface, delivering 608 GB/s bandwidth. Total board power is 310W, powered by a 12V-2x6 connector, and it adopts a dual-slot blower cooler design, ideal for dense workstations. Outputs include three DisplayPort 2.1 and one HDMI 2.1b.

According to ASUS, a single card supports most models under 27 billion parameters, with dual configurations reaching 80B models and quad-card setups achieving 128GB total VRAM for models up to 120B. European prices start around €1,799.

Why does this matter?

With growing demand for local AI inference, especially for privacy and reduced costs, high-VRAM cards become essential. The UGen Arc Pro B70 offers an accessible alternative to NVIDIA and AMD options in the professional segment, with support for frameworks like Ollama, llama.cpp, vLLM, PyTorch, and OpenVINO.

The ability to scale up to four cards in one system enables enterprise-level setups at a fraction of the cost of cloud solutions or dedicated data center GPUs. This democratizes access to large models for developers, researchers, and small businesses.

What changes in practice?

Users can run LLMs such as Llama 3.1, Qwen, DeepSeek, and others locally without relying on cloud services, with wide context windows. The blower design facilitates use in multi-GPU machines, and ECC memory ensures stability in long training or inference loads.

Although not aimed at gaming, the card maintains support for graphics APIs like DirectX and Vulkan. ASUS recommends a 750W PSU and provides a cable adapter. Availability varies by region, with European listings already active.

By GeekikiBot