ASRock Rack is showing off a new lineup of liquid-cooled AI servers at NVIDIA GTC 2026 in San Jose, running through 2026-03-19. The centerpiece of the announcement is a pair of systems built around the NVIDIA HGX Rubin NVL8 Platform: the 2U16X-GNR2/DLC RUBIN and 4U16X-GNR2/DLC RUBIN, targeting large-scale AI training and HPC workloads. ASRock Rack is also demonstrating a 44RU liquid-cooled rack housing eight 4U16X-TURIN2 systems as a complete deployment-ready solution.
Rounding out the portfolio are two systems based on the NVIDIA MGX platform. The 6UXGM-GNR2/DLC supports eight liquid-cooled RTX PRO 6000 Blackwell Server Edition GPUs aimed at enterprise AI and visual computing. The 4UXGM-GNR2 CX8 takes a more power-efficient single-slot approach with support for RTX PRO 4500 Blackwell Server Edition GPUs for data processing and AI inference. ASRock Rack is also previewing CPU-based systems built on the NVIDIA Vera architecture, which features custom CPU cores, LPDDR5X memory, and NVIDIA Scalable Coherency Fabric targeting reinforcement learning and agentic AI workloads at data center scale. On the software ecosystem side, NVIDIA's data center drivers provide Linux support for the Blackwell GPU family in the MGX-based systems with CUDA 12.8, though major open-source ML frameworks including PyTorch are still working through stable support for Blackwell's newer SM12x compute capabilities (tracked in an open upstream issue). Looking further ahead to the Rubin-based platforms, NVIDIA engineers have already submitted early patches to the Rust-written Nova open-source driver adding BOOT42 register support targeting Rubin-era GPUs, laying groundwork well ahead of H2 2026 hardware availability.
The full lineup is on display at ASRock Rack booth #110 at the San Jose Convention Center through 2026-03-19, with no pricing announced at this time.
