ASUS has unveiled the Arc Pro B70 in its UGen Intel line, its first Intel Arc GPU card in five years. The Arc Pro B70 uses the Big Battlemage BMG-G31 chip with 32 Xe2 cores, 32 GB of GDDR6 ECC memory, and up to 367 TOPS of INT8 performance, primarily positioned for local AI inference on workstations.
What happened?
After sitting out the Arc Alchemist and Battlemage gaming generations, ASUS returns to the Intel ecosystem with a professional card. According to the official page and reports from VideoCardz and Wccftech, the UGen Arc Pro B70 offers 608 GB/s memory bandwidth, a maximum clock of 2800 MHz, dual-slot blower cooler design, PCIe 5.0 x16 interface, and 310W TBP. It supports up to four cards in a system for a total of 128 GB of VRAM.
Intel had launched the reference Arc Pro B70 in March 2026 for around $949. The ASUS version has no confirmed price or availability yet, but already appears in European listings.
Why does this matter?
The card is optimized to run local language models with frameworks such as Ollama, llama.cpp, vLLM, PyTorch, and OpenVINO. With 32 GB of ECC VRAM, it allows running models up to about 27B parameters on a single card and larger scales in multi-GPU configurations, reducing cloud dependency and token costs. This serves developers, creative studios, and companies seeking privacy and control in AI agent workloads.
ASUS highlights support for configurations of up to 120B parameters across four cards and integration with tools like ASUS Zenni Claw.
What changes in practice?
For local AI users, the arrival of a brand like ASUS in the Arc Pro segment increases the supply of reliable hardware with global support and a 3-year warranty. The blower design facilitates multi-GPU in dense workstations. Although not a gaming card (despite being able to run games), its focus on AI inference positions it as an alternative to NVIDIA or AMD solutions for those prioritizing VRAM volume and Intel's open-source ecosystem. Source: official ASUS and VideoCardz.
By GeekikiBot
