Stay Cool, Scale Fast
Supercharging AI Factories for Enterprise
ASUS AI POD with NVIDIA HGX™ Rubin NVL8
1Air-Cooled CPU Sled
Simplified Design, Faster deployed time
2Flexible Storage Options
Bigger space for U.2/E1.S
3Fully Cable-less system
Board-to-Board Design, High Computing Density
4Up to 50% Less Maintenance Time
Faster Servicing, Simplified Stock & Less Downtime
Faster Time-to-Train
- Massive parallel GPU scaling
- Ultra-low latency interconnect
- Sustained performance under full load
Balanced Performance & Efficiency
- Low-latency processing
- High throughput
- Cost-efficient operation
Deploy with Confidence
- Faster time-to-deployment
- Consistent uptime and reliability
- Simplified maintenance and flexibility
NVIDIA HGX™ Rubin NVL8,
Powering Every AI and HPC Workload
Boost Token Factory Throughput
*Source: NVIDIA. Projected performance, subject to change. Token throughput based on Kimi K2-Thinking (FTL≤500ms, ISL=4K, OSL=4K); HGX Rubin NVL8 with Sparse NVFP4 vs. HGX B200 with Dense NVFP4. Training comparison based on DeepSeek-R1 pretrained on 15T tokens, 4K sequence length.
Accelerating the Next Generation of Agentic AI
A 3U Compute Tray Built for Flexibility and Serviceability
-
-
1 2 3 3 4 5 6 7 7 818 x NVMe SSD
2Front I/O Board
32 x Riser Card
4Mainboard
5Storage Backplane
6Baseboard - HGX Rubin NVL8
7CPU Heat Sink
8Middle Plane
-
1 2 31UQD08
(Hot Liquid Outlet)2600A Busbar Clip
3UQD08
(Cold Liquid Inlet) -
1 2 3 4 5 6 7 8 9 1011 x PCIe Gen5 x16 (FHHL)
(for BlueField-4)2NVMe U.2 x 8 *
31 x PCIe Gen5 x16 (FHHL)
(No signal while BF4 installed)4Power Button
5DM LAN
62 X USB3.2 Gen1
7Locate Button
8Debug Port
9Mini DP
102 x 10G LAN
*NVMe E1.S x 8 design reserved. Please contact Sales if needed.
-
-
-
Leading the Era of Super-Intelligent AI Reasoning
XR R1I12B-M1BHFS Key Features
-
-
CPU18x Intel Xeon6 6776P ProcessorsGPU72 NVIDIA Rubin GPUCoolingHybrid Cooled
GPU sled: Liquid Cooling
CPU sled: Air CoolingNetworking2 x 10GbE RJ45 LAN module (Intel-x710 Based-T)1 x Mgmt LANRack Power
ShelfPower Shelf: 3+1 Redundant 110kW @415VAC~480VACPower Whip: 8*100A IEC60309
1x OOB MGMT Switch 1GBe
2 x 3RU Power Shelf 110KW
9 x Compute Trays
72 x NVIDIA Rubin NVL8 GPUs
18 x Intel Xeon6 6776P CPUs2 x 3RU Power Shelf 110KW
-
ASUS AI POD with NVIDIA HGX™ Rubin NVL8
North-South Networking
North-South Networking
Power Shelves
Management Node
File System
Object Storage
Management switch
Power Shelves
9 x Compute Trays
• 18 x Intel® Xeon® 6 processors
• 72 x NVIDIA Rubin GPUs
Power Shelves
ASUS Professional Services
Your Trusted Partner from Factory to Future
-
Testing and Validation
-
Packaging and Delivering
-
Deployment and Management
FAQs
-
Why does XR R1I12B-M1BHFS use a 3U hybrid-cooling design instead of full liquid cooling like the 2U XR R1I12B-M1BLFS ?Why does XR R1I12B-M1BHFS use a 3U hybrid-cooling design instead of full liquid cooling like the 2U XR R1I12B-M1BHFS ? XA NR1I-E12L is built for data centers that need to run HGX Rubin NVL8 GPU compute but aren't yet ready for full direct liquid cooling (DLC) across every component. It applies direct liquid cooling to the GPU sled only, while the CPU sled uses air cooling (6 x system fans). This hybrid architecture, housed in a 3U (447mm x 800mm x 131mm) chassis with 54V busbar power, lets customers adopt Rubin NVL8 GPU density while reducing the facility-side liquid-cooling infrastructure required at initial deployment.
-
Can XR R1I12B-M1BHFS be deployed in an existing air-cooled data center?Can XR R1I12B-M1BHFS be deployed in an existing air-cooled data center? Not entirely — the HGX Rubin NVL8 GPUs always require liquid cooling regardless of chassis design. XR R1I12B-M1BHFS 's hybrid-cooling architecture minimizes the liquid-cooling footprint by confining it to the GPU sled, while the CPU sled and other components remain air-cooled. Customers still need CDU and liquid-supply infrastructure for the GPU loop, so facility cooling capacity and readiness should be assessed before deployment. ASUS can help evaluate whether an existing environment can support this hybrid configuration.
-
What storage flexibility does XR R1I12B-M1BHFS offer compared to the E1.S-based XR R1I12B-M1BFS ?What storage flexibility does XR R1I12B-M1BHFS offer compared to the E1.S-based XR R1I12B-M1BFS ? XA NR1I-E12L provides 8 x 2.5" hot-swap NVMe drive bays (with E1.S support reserved as a design option), offering more capacity and format flexibility than the LR's fixed 8 x E1.S (4TB) configuration. The larger 2.5" bay footprint gives customers more options when sizing local storage for data-intensive AI workloads.
-
What are the power and cooling requirements per rack for XR R1I12B-M1BHFS ?What are the power and cooling requirements per rack for XR R1I12B-M1BHFS ? A full 48RU MGX rack supports 9 XA NR1I-E12L compute trays, totaling 72 NVIDIA Rubin GPUs and 18 Intel Xeon 6776P processors. Each compute tray draws approximately 25.7kW (22.2kW GPU + 3.5kW CPU incl. fan), for a rack-level system power budget of approximately 237.2kW. Power is delivered through 2+2 redundant 110 kW power shelves (415–480VAC) with 8 x 100A IEC60309 power whips, and a 5000A liquid-cooled rack busbar. Cooling is hybrid — liquid for the GPU sled, air for the CPU sled (liquid CPU cooling under evaluation).
-
What deployment and after-sales support does ASUS provide for global rollouts of XR R1I12B-M1BHFS ?What deployment and after-sales support does ASUS provide for global rollouts of XR R1I12B-M1BHFS ? ASUS provides deployment planning and technical support for HGX Rubin NVL8 rack-scale solutions, including infrastructure assessment, rack integration, and system validation. Deployment and after-sales service options may vary by region and project requirements. ASUS teams can work with customers to define the appropriate service scope for each deployment.