ASUS Announces UGen Intel Arc Pro B70 32GB AI Accelerator Cards
ASUS unveiled the UGen Intel Arc Pro B70 32GB AI accelerator card, offering 32GB GDDR6 ECC VRAM to support local AI inference with up to 128K token context windows and dual-card 64GB scaling.
ASUS announced the UGen Intel Arc Pro B70 32GB, a graphics card tailored for local agentic AI workloads. The card features 32GB of GDDR6 ECC memory and an inference-optimized design to meet rising demand for affordable, high-capacity VRAM in desktop and workstation environments. It targets memory-intensive AI tasks such as long-context reasoning and multi-agent deployments, where insufficient VRAM can degrade performance by up to 90% due to data offloading to host memory.
The UGen B70 supports models under 27 billion parameters and enables context windows up to 128K tokens with Llama 3.1 8B. Dual-card configurations scale internal capacity to 64GB, while quad-card setups reach 128GB, catering to small businesses and large enterprises. ASUS emphasizes the card’s role in enabling local, private AI deployments to reduce cloud dependency and associated costs.
The card is backed by a mature AI software ecosystem, including Intel LLM Scaler, vLLM XPU, llama.cpp, Ollama, OpenVINO, and PyTorch XPU. Verified models include Llama 3.1/3.2, Qwen 3.x, DeepSeek-R1 Distill, Mistral, Mixtral, and OpenAI gpt-oss, with dual-card systems achieving 43.4 tokens per second on 80B-class models.
ASUS positions the UGen B70 for three user segments: prosumers and developers gaining access to larger models with a single 32GB card, small businesses and studios using dual-card 64GB setups for in-house deployments, and enterprises leveraging multi-card 128GB systems to meet data-compliance needs while cutting cloud costs.