GIGABYTE’s New Liquid Cooling GPU Server

Brand-new NVIDIA HGX H100 systems constructed using the Intel Xeon platform

The industry pioneer in high-performance servers, server motherboards, and workstations, Giga Computing, a subsidiary of GIGABYTE, today unveiled two new GPU servers for the 4th Gen Intel Xeon Scalable CPU and NVIDIA HGXTM H100 4- and 8-GPU platforms. The GIGABYTE G363-SR0 excels at both optimal performance and cooling efficiency by readily supporting processors and GPUs with high TDP requirements. The GIGABYTE G593-SD2 server is aimed at data centers that prefer air-cooled equipment. Both provide the most GPU processing power that can be accommodated by a single server of their respective capacities.

The new NVIDIA L40S GPU, which is excellent for multipurpose applications including AI, graphics, and video, will also be qualified by Giga Computing. New workstation-focused NVIDIA Ada Generation GPUs like the NVIDIA RTX A5000, NVIDIA RTX A4500, and NVIDIA RTX 4000 are also eligible.

Direct Liquid Cooling, G363-SR0

Greater performance necessitates more creative solutions. The four NVIDIA GPU modules and the Intel Xeon processors are both efficiently and quickly cooled by the G363-SR0’s direct liquid cooling (DLC). The 3U G363-SR0 provides optimal computing performance for the majority of AI and HPC workloads at a 1:2 CPU-to-GPU ratio. There are eight 2.5″ bays available to accommodate the massive amount of high-speed data access needed. Gen5 NVMe/SATA/SAS drives are supported.

There are six PCIe Gen5 low-profile slots available to take use of the best networking technology. Additionally, a 2 + 1 redundant power supply setup using 3000W titanium power supplies is used to power the system.

DLC technology was chosen because it has excellent all-around performance and can swiftly dissipate heat from the CPU and four GPUs. When compared to a conventional air-cooled system, this enables the server to function faster while also allowing for higher energy savings.

Alternatives to the conventional hot-cold aisle server rooms that exhaust hot air back into the room must be developed in order to counteract this tendency, as data centers will in the near future consume a growing portion of the world’s electricity. When using direct liquid cooling, the heat produced can be transmitted to the facility water and then rejected using a cooling tower or another similar equipment outside the data center. As a result, the cost of the accompanying cooling infrastructure can be decreased, which lowers OPEX, TCO, and improves ROI.

8 GPU System, G593-SD2

There has been a ton of interest in our G593 series, which is built for the top tier of GPU-accelerated computing, due to the high demand for AI training and other compute-intensive workloads. The twin Intel Xeon CPUs are housed in the top 1U of the 5U chassis, and the NVIDIA GPUs are housed in the bottom 4U. Without the need of sophisticated cooling techniques, this design guarantees that top-of-the-chart performance will not be throttled in a packed chassis.

GIGABYTE is constantly in contact with its cherished partners and consumers, and as a result, frequently makes minor tweaks to give what the market demands. In this instance, a minor but incredibly favored modification to the form factor of several of the expansion slots in the G593 gave rise to the G593-SD2. The G593-SD0, which had room set aside on the back top side for low-profile slots, came before this new server. These slots were eliminated and replaced in the new -SD2 model with the FHHL slots needed for NVIDIA BlueField DPUs.

A minor adjustment was performed to enable the G593-SD2 to utilize the numerous BlueField cards to speed up the circulation of data for AI and HPC as well as security isolation. The high level of computations for generative AI or demanding HPC workloads will be provided by both G593 servers.

Encouragement of Accelerators

This week, NVIDIA expanded its already sizable GPU range for computing or visualization by adding four additional accelerators. Giga Computing will support the new GPUs by certifying its business solutions. Additionally, these new GPUs will be offered for purchase when paired with our server platforms.

The most potent general-purpose GPU for data centers is the NVIDIA L40S GPU, which is based on the NVIDIA Ada Lovelace architecture and offers ground-breaking multi-workload acceleration for inference and training, graphics, and video applications.

The latest NVIDIA RTX Ada Generation GPU workstations are intended for various professional user levels who require high-performance, real-time ray tracing, AI-accelerated computing, and expert graphics rendering. The Ada-based RTX 5000, RTX 4500, and RTX 4000 GPUs are PCIe Gen4 air-cooled cards that enable innovations such NVIDIA GPUDirect Remote Direct Memory Access (RDMA) for quicker and more effective data transfer to other systems and NVIDIA GPUDirect for video support.

As the year continues, more servers with GPU baseboard modules will enable liquid cooling via a kit supplied with the company’s servers. Giga Computing has just unveiled more potent solutions for GPUs. Customers will be able to lower energy expenses and greener their data centers by adopting this.