Nvidia Grace Blackwell Ultra

Asus has introduced a desktop workstation that looks far closer to a data center machine than a conventional office PC. The ExpertCenter Pro ET900N G3 carries a price tag of around Rp 1.5 billion, or US$99,999, and is aimed squarely at demanding AI workloads. Its hardware choices make that positioning clear. The system is built around Nvidia’s Grace Blackwell Ultra AI chip and is intended for professionals working on large-scale AI research and large language model development. Memory and AI performance at an extreme level At the core of the machine is Nvidia Grace Blackwell Ultra with 72 ARM Neoverse V2 cores. That puts the ET900N G3 in a very different category from mainstream desktop systems, where the focus is usually general productivity rather than intensive AI computation. The memory configuration is just as unusual. Asus pairs 496 GB of LPDDR5X RAM with 252 GB of HBM3e VRAM, bringing total memory to 748 GB. According to the specifications, the VRAM bandwidth reaches 7.1 TB/s. That level of throughput is important for workloads that move very large datasets and need rapid access to memory. Built for sustained workloads, not everyday use Despite its specialized internals, the workstation is housed in a conventional tower chassis measuring 23.2 x 58.4 x 56.5 cm. That size still allows it to sit on a standard desk, even though the hardware inside is anything but standard. Asus says the internal cooling system is designed to maintain stability under continuous heavy load. The goal is to prevent the processor and GPU from having to slow down during extended AI sessions. That combination of cooling and memory capacity makes the machine suitable for running large AI models locally. It is not positioned as a gaming PC or a general-purpose desktop. Expansion, power, and connectivity To support the hardware, Asus equips the system with a 1,600-watt power supply. For storage and expansion, it includes three PCIe 5.0 slots and four M.2 SSD slots. One of the PCIe 5.0 slots can be used to add an additional Nvidia graphics card. That gives users more room to expand graphics and compute capabilities if their workloads grow. The workstation runs Ubuntu, a choice that fits professional users who need a flexible environment for development and compute tasks. Key Specification Detail Processor Nvidia Grace Blackwell Ultra CPU Cores 72 ARM Neoverse V2 System RAM 496 GB LPDDR5X VRAM 252 GB HBM3e Total Memory 748 GB VRAM Bandwidth 7.1 TB/s Power Supply 1,600 Watts Expansion 3 PCIe 5.0 slots, 4 M.2 SSD slots Operating System Ubuntu Basic ports for a machine in this price class Connectivity includes eight USB 3.2 ports, one USB 2.0 port, and two 10 Gbps Ethernet ports. For a system at this level, the port selection is functional rather than especially expansive. There is no USB 4 or Thunderbolt support, which may matter for users who regularly move very large files on external devices. Even so, the ET900N G3 stands out as an early sign of how local AI desktop systems may evolve. The workstation is already listed by several retailers in the United States, with the same Rp 1.5 billion pricing attached to it.

Asus Puts 748 GB of Memory Into a Workstation Built for Heavy AI Jobs