- Published on
176W Peak Performance, a Powerful Assistant for AI Creation
- Authors
- Name
- DeepHorus
In May this year, Thunderobot launched a water-cooled mini AI workstation powered by the AMD Ryzen AI Max+ 395 processor. With a peak power consumption of 176W and 128GB of unified memory, it has become a powerful tool for local deployment of large AI models. The device features a square all-aluminum chassis measuring 199×199×199 mm and weighing 6.5 kg, with a built-in 400W Platinum-rated power supply and a water-cooling system, designed specifically for desktop office scenarios.
As AI becomes more integrated into the workplace, cloud-based applications face issues such as privacy leaks, network fluctuations, and high token costs. Thunderobot's workstation addresses these pain points with local computing power, allowing users to run large models without uploading data, while ensuring response speed and data security. Its processor integrates a Zen 5 CPU, RDNA 3.5 integrated graphics, and an XDNA 2 NPU, delivering 50 TOPS of AI performance and supporting allocation of 96GB from the 128GB memory as video memory.
In real-world testing, the workstation uses a 12-layer PCB and 14-phase power delivery, with a water-cooling system covering a 12,000 mm² cooling island. In the CineBench R23 stability test, CPU power consumption spiked to 176W momentarily with a temperature of 93.4°C, then stabilized at an average of 155W. After a 7-minute AIDA64 FPU stress test, power consumption remained at 133W, temperature dropped to 82.2°C, and frequency held at 3.9GHz, showing better thermal performance than similar air-cooled products.
In terms of performance, the CineBench R23 single-core score is 2084 points and multi-core score is 41,350 points. After the stability test, the multi-core score dropped by only 3%, demonstrating excellent sustained load capability. Memory read and write speeds reach 119.45GB/s and 172.55GB/s respectively, providing sufficient bandwidth for parallel AI model computing. For storage, it comes with a pre-installed 2TB YMTC PCIe 4.0 SSD with read/write speeds exceeding 7000MB/s, and supports expansion to three hard drives.
The interface configuration is comprehensive, with USB-C and audio ports on the front, and dual USB4, a 10G Ethernet port, and Wi-Fi 7 antennas on the rear. The built-in power supply saves space compared to traditional external adapters and reduces standby power consumption by 27%. The water-cooling fan uses industrial-grade dual ball bearings with a lifespan of 100,000 hours, meeting long-term operation needs.
This product targets professionals who need local AI capabilities, such as copywriting, code debugging, and data analysis scenarios. Compared to cloud-based solutions, it eliminates privacy concerns and subscription fees, but its 6.5 kg body limits portability, making it more suitable for fixed workstations. By combining water cooling with high-spec hardware, Thunderobot demonstrates the potential of mini PCs in AI computing power, and may encourage more manufacturers to follow this design direction in the future.
