• Unimageable AI. Unleashed.

    Bring your NVIDIA GB200 concept to life, backed by ASUS

    Unlock the full potential of your AI infrastructure with the ASUS AI POD powered by NVIDIA GB200 Grace™ Blackwell Superchips. This supercharged AI system is engineered to accelerate real-time trillion-parameter models like LLM and MoE, providing unparalleled performance for large-scale AI projects, hyperscale data centers and research initiatives. Tailored to enterprises and R&D teams driving innovation, ASUS AI POD empowers you to push the boundaries of what’s possible, all while accelerating your time to market with the ASUS-exclusive software-driven approach.

    Watch the video ASUS AI POD whitepaper Apply for Online Test
    One NVIDIA GB200 NVL72 in the data center
    • a square containing a smaller square inside, each with four dots at the top and bottom.
      Blackwell-architecture GPUs

      208-billion transistors with TSMC 4NP process

    • The design features a combination of small dots connected in a pattern, linked to a triangle on the left and a small square on the right
      2nd-gen Transformer Engine

      Doubling the performance with FP4 enablement

    • there are four squares connected in a series, with an X at the center.
      5th-gen NVLink & NVLink Switch

      1.8 TB/s GPU-GPU interconnect

  • The NVIDIA Blackwell GPU breakthrough

    ASUS AI POD contains NVIDIA GB200 Grace™ Blackwell Superchips, each packing 208-billion transistors. NVIDIA Blackwell GPUs feature two reticle-limited dies connected by a 10 terabytes per second (TB/s) chip-to-chip interconnect in a unified single GPU. NVIDIA GB200 NVL72 delivers a computing power of 1,440 PFLOPS in FP4 precision, leveraging its advanced Tensor Cores and fifth-generation NVLink interconnects to achieve high performance for AI workloads.

    Rack System layout Compute Tray Layout NVIDIA GB200 Superchips
    • Front View Rear View
      • Front view of nvidia gb200 nvl72
        1 2 3 4 5 6

        12 x OOB MGMT Switch 1GBe
        1 x OS Switch 1GBe (Optional)
        1 x RMC (Optional)

        23 x 1RU Power Shelf 33KW

        310x Compute Trays

        49x Non-Scalable NVLink
        Switch Trays

        58x Compute Trays

        63 x 1RU Power Shelf 33KW

      • rear view of nvidia gb200 nvl72 rear view of nvidia gb200 nvl72
        1 2

        11400A Rack Busbar

        2NVIDIA Cable Cartridge

    • Top View Front View Rear View
      • This is a line graph that helps people understand the locations of the key components on this server. This is a line graph that helps people understand the locations of the key components on this server.
        1 1 2 3 4 5 5 6 6 7 8 9 10

        1NVIDIA BlueField®-3 DPUs (B3240)

        2PDB

        3Inner Manifold

        4Fan Area

        5NVIDIA ConnectX®-7 Mezzanine Network Board

        6NVIDIA GB200 Grace™ Blackwell Superchips

        7HMC Module

        8Liquid Cooling Outlet

        9Busbar Clip

        10Liquid Cooling Inlet

      • Front view of the compute tray Front view of the compute tray
        1 2 2 3 4 5 6 7 8 9 9

        1NVIDIA BlueField®-3 DPUs (B3240)

        22 x OSFP 400G

        31G PXE

        48* E1.S

        5BMC

        6USB 3.2

        7miniDP

        8NVIDIA BlueField®-3 DPUs (B3240)

        92 x OSFP 400G

      • rear view of the compute tray rear view of the compute tray
        1 2 2 3 4 4 5

        1Liquid Cooling Outlet

        2NVIDIA® NVLink™Connector

        3Busbar Clip

        4NVIDIA® NVLink™Connector

        5Liquid Cooling Inlet

    • GB200 Grace Blackwell superchip
      • 30X
        LLM Inference
        NVIDIA H100 Tensor Core GPU
      • 4X
        LLM Training
        H100
      • 25X
        Energy Efficiency
        H100
      • 18X
        Data Processing
        CPU
      • LLM inference and energy efficiency: TTL = 50 milliseconds (ms) real time, FTL = 5s, 32,768 input/1,024 output, NVIDIA HGX™ H100 scaled over InfiniBand (IB) vs. GB200 NVL72, training 1.8T MOE 4096x HGX H100 scaled over IB vs. 456x GB200 NVL72 scaled over IB. Cluster size: 32,768
      • A database join and aggregation workload with Snappy / Deflate compression derived from TPC-H Q4 query. Custom query implementations for x86, H100 single GPU and single GPU from GB200 NLV72 vs. Intel Xeon 8480+
      • Projected performance subject to change

  • AI Infrastructure at scale

    ASUS continues to demonstrate its expertise in building AI infrastructure by leveraging the powerful NVIDIA GB200 NVL72. This advanced system, housed in a single rack with multiple compute nodes, consumes approximately 120kW of electricity. Given this substantial power requirement, it is crucial for data center operators to reassess and upgrade the existing topology and configuration. Optimization efforts should focus on ensuring stable performance through enhanced heat dissipation, robust networking connections, and the overall architectural integrity of the facility. As AI workloads grow more demanding, these infrastructure improvements are essential for meeting performance and efficiency goals while maintaining operational stability.

  • Maximize efficiency, minimize heat

    Liquid-cooling architectures

    ASUS, in collaboration with partners, offers comprehensive cabinet-level liquid-cooling solutions. These include CPU/GPU cold plates, cooling distribution units and cooling towers, all designed to minimize power consumption and optimize power-usage effectiveness (PUE) in data centers.

    • One NVIDIA GB200 nvl72 is connected to two CDUs (Cooling Distribution Units), with hot and cold water pipelines linked between them.

      Liquid-to-air solutions

      Ideal for small-scale data centers with compact facilities.
      Designed to meet the needs of existing air-cooled data centers and easily integrate with current infrastructure.
      Perfect for enterprises seeking immediate implementation and deployment.

    • Sixteen rack of NVIDIA GB200 nvl72 are connected to two CDUs (Cooling Distribution Units), with hot and cold water pipelines linked between them.

      Liquid-to-liquid solutions

      Ideal for large-scale, extensive infrastructure with high workloads.
      Provides long-term, low PUE with sustained energy efficiency over time.
      Reduces TCO for maximum value and cost-effective operations.

  • Brilliantly Fast

    5th-gen NVLink technology in NVIDIA GB200 NVL72

    The NVIDIA NVLink Switch features 144 ports with a switching capacity of 14.4 TB/s, allowing nine switches to interconnect with the NVLink ports on each of the 72 NVIDIA Blackwell GPUs within a single NVLink domain.

    Each GB200 compute tray is showcased as being interconnected via NVLink.

    NVLink connectivity in a single compute tray to ensure direct connect to all GPUs

    Each GB200 compute tray is showcased as being interconnected via NVLink.

    NVLink connectivity in single rack

  • ASUS Premiere Service Suite

    Transforming data centers with full-scale integration, from hardware to applications

    ASUS has redefined its approach to infrastructure service architecture by providing a full spectrum of services — from design and construction to implementation, validation, testing and deployment. This is seamlessly integrated with cloud-application services, all delivered with a customized, client-specific service model to meet diverse customer requirements.

    • The graphic design of the three servers is arranged in a right-angled configuration.

      Architecture Design

    • The screen displays a management interface showing seven servers connected in series.

      Large-scale Deployment

    • Three gears interlock and operate together.

      Configuration Test

    • A document and a pen, with a checkmark symbol on the document

      Customization

    • Three racks connected to a cloud above them.

      Cloud Services

    • A document with 3 checkmark symbol on it.

      Project Delivery

    Sixteen rack of NVIDIA GB200 nvl72 are connected to two CDUs (Cooling Distribution Units), with hot and cold water pipelines underneath

  • Accelerate your time to market

    ASUS self-owned software and controller

    The NVIDIA NVLink Switch features 144 ports with a switching capacity of 14.4 TB/s, allowing nine switches to interconnect with the NVLink ports on each of the 72 NVIDIA Blackwell GPUs within a single NVLink domain.

    ASUS Control Center
    (ACC)
     ASUS Infrastructure Deployment Center
    (AIDC)
     Rack Management Controller
    (RMC)
    • A tablet, a monitor, and a smartphone, all displaying the ASUS Control Center interface.

      ASUS Control Center (ACC)

      Centralized IT-management software for monitoring and controlling ASUS servers

      • Power Master – Effective energy control for data centers
      • Easy search and control of your devices
      • Enhance information security easily and quickly
    • 4 images all displaying the interface of ASUS Infrastructure deployment center

      ASUS Infrastructure Deployment Center (AIDC)

      One management console to realize centralized remote deployment

      • Automation and systemization
      • Centralized configuration control and management
      • Accelerated deployment for rack-scale infrastructure
    • A monitor display the interface of ASUS rack management controller

      Rack Management Controller (RMC)

      A centralized controller for remote hardware management within server a rack

      • Troubleshooting, management and maintenance
      • Batch monitoring