ASRock Rack has introduced a new AI server system, the 2UXGM-VERA2, built around the NVIDIA Vera CPU, and previewed a broader lineup of AI platforms spanning AI factories, inference cloud deployments, and edge AI applications at COMPUTEX 2026.
The 2UXGM-VERA2 is based on NVIDIA Vera, which combines custom-designed NVIDIA Olympus cores, high-bandwidth LPDDR5X memory, and NVIDIA Scalable Coherency Fabric. ASRock Rack says Vera “completes complex agentic and reinforcement learning workloads 50% faster than traditional CPU infrastructure.”
CPU capability matters more as AI systems take on more orchestration, tool use, and multi-step workflows that aren’t just GPU-bound. But ASRock Rack’s performance claim is comparative and workload-dependent; for data center teams, the practical questions are what the platform looks like under real scheduling pressure, and how it behaves when you’re balancing latency, memory bandwidth, and power at rack scale.
On the cooling side, ASRock Rack showed AI server platforms integrated with liquid-cooling solutions from ecosystem partners. The company highlighted an NVIDIA Vera Rubin NVL72 configuration cooled with a double-rack-width liquid-to-air coolant distribution unit (CDU). It also showed liquid-cooled systems incorporating NVIDIA HGX Rubin NVL8, including the fully liquid-cooled 2U16X-GNR2/DLC and the 5U16X-GNR2/DLC, which ASRock Rack says combines liquid cooling for CPU and GPU with enough airflow to support other peripherals. ASRock Rack also demonstrated rack-scale liquid cooling with an in-row CDU for high-density deployments.
For enterprise and edge deployments, ASRock Rack showed MGX-based systems, including the 6UXGM-GNR2/DLC, which supports up to eight liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Also shown was the 4UXGM-GNR2 CX8, described as a NVIDIA RTX PRO Server optimized for inference cloud environments. For industrial edge use cases, ASRock Rack highlighted the 2UXGI-Thor, built on the NVIDIA IGX Thor platform, aimed at “Physical AI” with real-time sensor processing and functional safety for autonomous robotics and medical applications.
“ASRock Rack is planning a comprehensive new lineup of AI servers powered by NVIDIA Vera, ranging from standalone CPU servers to scale-up AI servers incorporating NVIDIA HGX Vera Rubin NVL8,” said Weishi Sa, President of ASRock Rack.
Source: ASRock Rack
