NVIDIA doesn't only make supercomputers with powerful GPUs. At Computex in Taiwan today, the company announced a new server design platform, MGX, that should allow server makers to design and build them in tons of different ways, and in much less time.
NVIDIA's press release stated that the MGX platform was able to allow companies to create over 100 server designs. The modular platform also cuts down on a server's development time down to six months and costs down by as much as three quarters.
With MGX, manufacturers start with a basic system architecture optimized for accelerated computing for their server chassis, and then select their GPU, DPU and CPU. Design variations can address unique workloads, such as HPC, data science, large language models, edge computing, graphics and video, enterprise AI, and design and simulation.
NVIDIA added that the MGX server design makes upgrading them far easier in the future. Server builds can also continue to use existing designs with the MGX platform. These servers can be quickly integrated into current cloud and enterprise data centers.
Several server makers have already announced plans to adopt and use NVIDIA MGX. They include ASRock Rack, ASUS, GIGABYTE, Pegatron, QCT and Supermicro. The latter two companies will be the first to offer servers in the MGX format. OCT's server will use NVIDIA's new Grace Hopper GPU which is also being used in NVIDIA's own DGX GH200 supercomputer.
The company added:
Accelerated computing servers from NVIDIA have long provided exceptional computing performance and energy efficiency. Now, the modular design of MGX gives system manufacturers the ability to more effectively meet each customer’s unique budget, power delivery, thermal design and mechanical requirements.
NVIDIA's other announcements at Computex today included a new chatbot AI for games called ACE and a new Ethernet switch for AI applications, Spectrum-X.
