Artificial intelligence is moving rapidly out of distant cloud servers and straight onto local office desks. Running large language models and multi-agent workflows locally requires a tremendous amount of video memory. When your system runs short on VRAM, data offloads to host memory, causing inference performance to plummet drastically. ASUS is stepping in to solve this exact problem with a powerful new hardware release. The tech giant has officially announced the ASUS UGen Intel Arc Pro B70 32GB AI accelerator card, built from the ground up for the emerging agent computer era.
Single-Card Prosumer Form Factor
For independent developers, researchers, and AI enthusiasts, a standard 16GB graphics card often creates a frustrating resource ceiling. The single-card configuration of the ASUS UGen B70 shatters that barrier by packing thirty-two gigabytes of ECC-protected GDDR6 memory into a sleek, full-height, two-slot form factor. Driven by the Intel Xe2 architecture with thirty-two Xe-Cores and two hundred fifty-six XMX AI engines, this single card delivers up to 367 TOPS of INT8 AI performance. It effortlessly handles models under twenty-seven billion parameters and supports a massive context window of up to 128K tokens with Llama 3.1 8B. It keeps your sensitive data safely on-premises without any cloud offload penalties.
Dual-Card Small Business Form Factor
Growing creative studios and small businesses often need to run larger models and more complex multi-agent workflows. The ASUS UGen B70 scales smoothly to meet those demands through a dual-card configuration. By pairing two cards together inside a professional workstation, users unlock a combined sixty-four gigabytes of high-speed VRAM. This expanded capacity allows teams to deploy larger models while keeping proprietary data entirely in-house. Supported by a robust software stack including vLLM XPU, Ollama, and OpenVINO, dual-card setups have been proven to run demanding 80B-class models at impressive speeds of over forty-three tokens per second.
Quad-Card Enterprise Form Factor
Enterprise departmental teams in fields like finance, healthcare, and law require massive computing power combined with strict data compliance. The ASUS UGen B70 accommodates heavy enterprise workloads through a scalable quad-card configuration. When four cards work in unison, the system scales up to a staggering one hundred twenty-eight gigabytes of total VRAM. This multi-card setup delivers cloud-level AI capabilities right at the desk’s side at a fraction of the cost of renting cloud APIs. Operating with a thermal design power of 310 watts per card, these workstations provide a cost-effective, long-term foundation for large-scale enterprise artificial intelligence deployment.
Pricing & Availability
The ASUS UGen Intel Arc Pro B70 32GB accelerator card will be available in Malaysia starting from early December 2026. It carries an official suggested retail price of RM8,459.
