ASRock Rack Inc., a leader in innovative server solutions, has launched an impressive lineup of servers harnessing the power of NVIDIA’s Blackwell architecture at COMPUTEX 2024.
Showcased at booth L0818 in Hall 1, these new servers are set to revolutionize the field of accelerated computing and AI workloads.
Highlighted Releases: Liquid-Cooled and Direct-to-Chip GPU Servers
ORV3 NVIDIA GB200 NVL72: A Liquid-Cooled Powerhouse
The ORV3-NVIDIA GB200 NVL72 is a standout in the new range, featuring advanced liquid cooling technology. This rack-scale solution integrates 36 NVIDIA Grace CPUs and 72 NVIDIA Blackwell Tensor Core GPUs. Utilizing NVIDIA NVLink technology, it forms a unified GPU, significantly boosting real-time large language model (LLM) inference capabilities while lowering the total cost of ownership (TCO).
6U8X-GNR2/DLC NVIDIA HGX B200: Direct-to-Chip Cooling for Extreme Performance
Another highlight is the 6U8X-GNR2/DLC NVIDIA HGX B200, which employs direct-to-chip liquid cooling. This server is designed to push the thermal design power (TDP) limits of NVIDIA Blackwell GPUs within a 6U rackmount design, ensuring optimal performance for the most demanding computational tasks.
Additional New Models: Versatile and High-Performance Solutions
6U8X-EGS2 NVIDIA HGX B100: Seamless Transition and Enhanced Efficiency
The 6U8X-EGS2 NVIDIA HGX B100 is designed to serve as a drop-in replacement for existing NVIDIA HGX H100 infrastructures. This server simplifies the upgrade process, allowing users to transition smoothly to the enhanced capabilities of the NVIDIA HGX B100 system.
Air-Cooled Alternatives: 8U8X-GNR2 NVIDIA HGX B200
For environments without liquid-cooling infrastructure, ASRock Rack introduces the 8U8X-GNR2 NVIDIA HGX B200. This air-cooled variant also integrates an NVIDIA HGX B200 8-GPU system, providing robust performance for real-time inference on trillion-parameter models.
Leveraging NVIDIA MGX Modular Reference Architecture
4UMGX-GNR2: Dual-Socket GPU Server
ASRock Rack’s 4UMGX-GNR2 server, compliant with the NVIDIA MGX 4U standard, supports up to 8 FHFL dual-slot GPUs, including the NVIDIA H200 NVL Tensor Core GPU. This server offers multiple PCIe 5.0 x16 slots for enhanced connectivity and is equipped with 16 hot-swap drive bays for E1.S SSDs, ensuring high-speed data processing capabilities.
Supporting Next-Gen AI and Accelerated Computing
All ASRock Rack NVIDIA HGX servers support up to eight NVIDIA BlueField-3 SuperNICs to leverage the NVIDIA Spectrum-X AI-optimized Ethernet platform. These systems will be certified for the comprehensive NVIDIA AI and accelerated computing platform, including the NVIDIA NIM inference microservices as part of the NVIDIA AI Enterprise software platform.
Future-Ready Innovations
ASRock Rack is also developing new products based on the NVIDIA GB200 NVL2 to enable scale-out configurations, ensuring seamless integration with existing data center infrastructures. This initiative underscores ASRock Rack’s commitment to expanding the capabilities of the NVIDIA Blackwell architecture for mainstream LLM inference and data processing.
Showcasing at COMPUTEX 2024
Visitors to COMPUTEX can explore these cutting-edge solutions and more at the ASRock Rack booth. Key exhibits include the MECAI-NVIDIA GH200 Grace Hopper Superchip server, launched earlier this year, demonstrating ASRock Rack’s ongoing innovation in server technology.
Conclusion
ASRock Rack's new lineup, powered by NVIDIA's Blackwell architecture, represents a significant leap forward in server technology. These servers are poised to meet the most demanding AI and computational workloads, providing unparalleled performance, scalability, and efficiency.
For more information, visit ASRock Rack.
This diagram illustrates the key products in ASRock Rack’s new server lineup, each leveraging the advanced capabilities of NVIDIA’s Blackwell architecture.
