ASUS Launches UGen Arc Pro B70 with 32GB ECC Memory, Targeting Local AI Inference Market

ASUS Enters Intel Arc Market After Two Generations

ASUS has unveiled the ASUS UGen Arc Pro B70, marking the company's first Intel Arc graphics card after sitting out two previous generations1

1

2

2

. While ASUS previously released a DG1 card in 2021, it remained absent during the Arc Alchemist and consumer Battlemage generations4

4

. The workstation graphics card signals ASUS's strategic shift toward the expanding AI hardware market, joining ASRock and Acer as major manufacturers supporting Intel's discrete GPU efforts.

32GB ECC Graphics Memory Powers Professional Workloads

The ASUS UGen Arc Pro B70 is built around Intel's Xe2 architecture using the Intel Battlemage GPU, specifically the BMG-G31 silicon with 32 Xe cores and 256 XMX engines2

2

3

3

. The most distinctive feature is its 32GB of ECC-enabled GDDR6 memory operating across a 256-bit interface at 19Gbps, delivering 608GB/s of theoretical bandwidth1

1

. This capacity positions the card above typical gaming GPUs for memory-intensive local AI inference tasks. The card achieves 22.94 TFLOPS of FP32 compute performance and up to 367 TOPS of INT8 performance at a maximum operating frequency of 2,800MHz1

1

4

4

.

Multi-GPU Configurations Enable 128GB Total Memory

ASUS explicitly designed the card for multi-GPU configurations, supporting up to four cards in a single system for a combined 128GB of graphics memory1

1

4

4

. This configuration can accommodate models up to 120-billion parameters, though actual scaling depends on model quantization, framework support, and interconnect behavior2

2

. The card uses a two-slot blower cooler design, advantageous for densely packed workstation systems because it exhausts heat through the rear I/O bracket rather than recirculating it inside the chassis1

1

2

2

. The trade-off is potentially higher noise levels under sustained load compared to open-air coolers.

Software Support Targets AI Development Ecosystem

ASUS emphasizes compatibility with major AI frameworks and tools for local AI inference applications1

1

. The card supports Ollama, llama.cpp, vLLM, PyTorch, OpenVINO, and oneAPI, indicating its focus on developers building local inference systems1

1

3

3

. This software ecosystem positions the card for content creation and AI and workstation use cases rather than gaming. The 32GB framebuffer accommodates larger models and datasets than typical 8GB or 16GB gaming cards, though system memory capacity and model quantization remain critical factors1

1

.

Power and Display Specifications

The card draws 310W of total board power through a single 12V-2×6 connector, with ASUS recommending a 750W power supply1

1

3

3

. This exceeds Intel's reference B70 card rating of 230W, though no ASUS-specific benchmarks demonstrate corresponding performance advantages2

2

. For multi-display systems, the card offers three DisplayPort 2.1 UHBR13.5 outputs and one HDMI 2.1b port, supporting up to four simultaneous displays with maximum resolution of 7,680 × 4,320 at 120Hz1

1

3

3

. The card measures 267mm long and 111mm high, weighing approximately 1.1kg1

1

.

Market Positioning and Pricing Remain Unclear

ASUS has not announced pricing, launch date, or regional availability for the ASUS UGen Arc Pro B701

1

2

2

. Intel introduced the reference B70 in March with a $949 starting price, though retail variants typically range from $1,299 to over $2,0004

4

. The card must compete against Intel's own Arc Pro offerings, NVIDIA RTX professional boards, and workstation-oriented Radeon products. Its primary differentiators are the combination of 32GB ECC memory, compact two-slot design, and multi-GPU deployment capabilities1

1

. Earlier benchmarks showed the B70 performing close to NVIDIA's RTX 5060 Ti 16GB at 1440p, despite its professional focus2

2

. ASUS's entry into the Intel Arc ecosystem suggests growing confidence in the platform's viability for professional workloads as demand for local AI hardware continues expanding.