DJI 20260603031949 0009 D CHAR withad showcased its latest AI hardware solutions at Computex 2026, highlighting the flagship ARTOP 500 system designed to run large language models locally on standard office power. The ARTOP 500 is built on the GRX50 motherboard, which features memory slots supporting up to 2TB of memory. In the demonstrated configuration, the motherboard was equipped with 768GB of memory consisting of eight 96GB modules. The system integrates NVIDIA RTX 5090 GPUs, supporting up to two of these graphics cards to handle demanding computational workloads. For alternative configurations, the platform is also compatible with AMD hardware, including the AMD Ryzen 9700X processor.
—
HDMI® Technology is the foundation for the worldwide ecosystem of HDMI-connected devices; integrated with displays, set-top boxes, laptops, audio video receivers and other product types. Because of this global usage, manufacturers, resellers, integrators and consumers must be assured that their HDMI® products work seamlessly together and deliver the best possible performance by sourcing products from licensed HDMI Adopters or authorized resellers. For HDMI Cables, consumers can look for the official HDMI® Cable Certification Labels on packaging. Innovation continues with the latest HDMI 2.2 Specification that supports higher 96Gbps bandwidth and next-gen HDMI Fixed Rate Link technology to provide optimal audio and video for a wide range of device applications. Higher resolutions and refresh rates are supported, including up to 12K@120 and 16K@60. Additionally, more high-quality options are supported, including uncompressed full chroma formats such as 8K@60/4:4:4 and 4K@240/4:4:4 at 10-bit and 12-bit color.
—
The ARTOP 500 system enables local deployment of extremely large AI models, supporting parameter sizes up to 600 billion. When relying solely on the VRAM of the NVIDIA RTX 5090 GPU, the system can run a 20 billion parameter model. By utilizing the 768GB of system memory, the capacity expands to handle models up to 400 billion parameters. To achieve the maximum 600 billion parameter capability, the hardware is bundled with an ARTOP 100 1TB SSD, leveraging storage to run larger language models that would otherwise require data center infrastructure.
A key design aspect of the ARTOP 500 is its power efficiency and deployment simplicity. The hardware operates on standard household and office voltage without the need for specialized power converters. The baseline configuration draws 1600W of power, while adding a second graphics card utilizes an additional power supply unit to bring the total capacity to 3200W. This standard voltage compatibility allows enterprises to deploy large language models directly on office desks or within standard workspace environments, avoiding the need for dedicated server room infrastructure or high-voltage power installations.
To demonstrate real-world performance, the platform was showcased running a demo kit tailored for an insurance industry use case. The test simulated 80 simultaneous users submitting unique queries to the Llama 3.2 3B model. The ARTOP 500 processed all 80 concurrent chat requests and returned responses in 22 seconds. The system has validated over 100 language models, including DeepSeek V1 671B, Llama, Gemma, and Kimi K2.5. The testing suite also allows users to evaluate model speeds across multiple precision levels, including FP4, FP8, FP16, and FP32, to optimize performance.
For mainstream and mid-tier deployments, the product lineup includes the ARTOP 400s, which features compatibility with both AMD and Intel chipsets, including the Intel 890 chipset. The ARTOP 400s supports up to two NVIDIA RTX 5090 GPUs and runs models up to 400 million parameters. To ensure reliability during continuous 24/7 AI model training and inference workloads, the chassis design features a specialized ventilation door to optimize airflow across key chipsets, along with integrated dust filters to prevent contamination and maintain thermal stability.
