Graphics cards have long been associated with gaming and graphics-intensive work, but since the proliferation of AI, a new breed of GPU has emerged: the AI accelerator. The ASUS UGen Intel Arc Pro B70 32GB GDDR6 is one such piece of AI-age PC hardware.
Remember the agentic AI enabled by chips like NVIDIA RTX Spark? Well, this is exactly what this card is for.
The difference is that instead of buying an entirely new AI computer, the ASUS UGen B70 offers a way to equip a compatible workstation with the hardware needed to run large language models (LLMs) and AI agents locally, without constantly relying on the cloud.
At its heart is Intel’s Arc Pro B70 GPU, equipped with 32 GB of ECC GDDR6 memory. That generous VRAM allocation is particularly important for AI workloads, where running out of graphics memory can cause performance to take a serious hit.
According to ASUS, a single card supports most AI models with fewer than 27 billion parameters. It can also handle a context window of up to 128K tokens with Llama 3.1 8B.
Need more AI muscle? Two cards deliver 64 GB of combined VRAM, while four cards bring that figure to 128 GB. ASUS says a dual-card configuration can even run 80-billion-parameter models at 43.4 tokens per second.
The card consumes 310 W and supports AI software including Ollama, llama.cpp, Intel OpenVINO, PyTorch XPU, and ComfyUI. Compatible models include Llama, Qwen, DeepSeek, Mistral, and OpenAI’s gpt-oss.
For developers, creative studios, and businesses handling sensitive information, the attraction is straightforward: running AI models and agents on their own hardware, with greater control over their data.
ASUS has yet to announce pricing or availability. Meanwhile, you can dig into more details about this new AI accelerator here.
Images: ASUS.
