Giga Computing Boosts Nvidia Server Line for AI Data Centres
Giga Computing, a subsidiary of GIGABYTE Group, has unveiled its latest addition to the NVIDIA RTX PRO Server portfolio.
The XL44-SX2-AAS1 integrates NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with NVIDIA BlueField-3 DPUs and NVIDIA ConnectX-8 SuperNICs, a combination designed to push data centres further into the AI era.
The server builds on NVIDIA’s MGX modular reference design, offering enterprises the compute power and interconnect speeds required to run workloads at scale.
The XL44-SX2-AAS1 supports up to eight RTX PRO 6000 Blackwell GPUs, each equipped with 96GB of GDDR7 memory. This configuration is aimed at compute-heavy operations such as generative AI, scientific simulations and large-scale 3D rendering.
It also incorporates dual Intel Xeon 6700/6500 CPUs, 32 DDR5 DIMMs and 3+1 redundant 3200W 80 PLUS Titanium power supplies, ensuring the 24/7 reliability expected from modern data centre hardware.
Networking designed for AI
Central to the platform is its ability to handle the east-west data traffic that defines large AI workloads. The BlueField-3 DPU delivers 400Gb/s of total bandwidth, accelerating GPU access to data while maintaining security for AI runtimes.
Meanwhile, the ConnectX-8 SuperNICs provide GPU-to-GPU direct communication across eight PCIe Gen6 x16 channels. Each GPU benefits from up to 800Gb/s InfiniBand or Ethernet bandwidth, plus external connectivity through two 400GbE links.
This combination allows data centres to run distributed AI training efficiently, with the SuperNIC technology helping to synchronise GPUs for demanding multi-node workloads.
Beyond GPUs: a complete enterprise AI platform
While GPU density remains a headline feature, Giga Computing emphasises that the XL44-SX2-AAS1 is not simply a GPU box. It has been positioned as a full enterprise AI platform, capable of running workloads from inference and reasoning to scientific computing and digital twins.
The server comes pre-integrated with Nvidia AI Enterprise software, including NIM microservices. It also supports Nvidia Omniverse for digital twin creation and Nvidia Cosmos for advancing physical AI, extending applications from the virtual world into industrial environments.
Compatibility with Windows, Linux, Kubernetes and virtualisation platforms means the system can be deployed within existing x86 data centres. Giga Computing claims this minimises infrastructure disruption, allowing enterprises to adopt AI workloads more rapidly.
Efficiency and ROI for large-scale operations
For data centre operators, return on investment remains a critical factor. The new server achieves up to 5X improvements in AI inference compared with its predecessor, a performance gain designed to reduce operational costs.
By combining GPU acceleration with optimised networking and system-level reliability, the XL44-SX2-AAS1 positions itself as a practical option for enterprises transitioning into AI-driven infrastructure.
The integration of EcoDesign principles into Schneider Electric’s hardware components further supports sustainability goals, reducing embedded carbon while maintaining long-term operational efficiency.
Industry impact and availability
The applications for such infrastructure cut across multiple industries. Smart manufacturing can leverage robotic simulation and real-time analytics. Financial institutions can accelerate complex modelling. Medical research teams can scale AI-assisted diagnostics and drug discovery. 3D content creators and LLM deployments also stand to benefit from the increased compute density.
By enabling AI workloads to run at scale, Giga Computing’s new platform contributes to the concept of AI factories – data centres designed not just to host applications but to power continuous cycles of training, inference and deployment.
The XL44-SX2-AAS1 will be generally available in October 2025, bringing high-performance compute and network integration to enterprises looking to expand their AI capacity.
