November 18, 2024, San Francisco, CA-Together AI has disclosed a joint venture with Hypertec Cloud to construct a superGPU cluster incorporating an incredible 36,000 NVIDIA GB200 Graphics Processing Units.
This project, which begins in the first quarter of 2025, is probably going to transform the training and inference of generative AI models by introducing unparalleled scalability, speed, and cost efficiency.
NVidia’s impressive Blackwell architecture, Grace CPUs, NVLink, and their proprietary Together Kernel Collection (TKC) are all features included with the cluster.
These technologies aim to optimize GPU on a cost basis and when used in AI model training and inference cut the costs by as much as 24% in the former and up to 75% in the latter.
“The Together Kernel Collection is a game-changer in the computational AI space where it is a matter of transforming images from text – the turnaround times have heightened significantly,”
Remarked Vipul Ved Prakash, the CEO of Together AI.
The resilient infrastructure of Hypertec Cloud serves to better energy the GPU technologies of Together AI.
“This partnership brings next-gen solutions in AI at scale with a reduced carbon footprint while filling global needs,”
Said Jonathan Ahdoot, President of Hypertec Cloud.
Victor Perez, who is Krea’s Co-Founder, underlined Together AI’s added value:
“Together AI enables us to deliver high-quality visual AI solutions that are real-time and we do not have to compromise creative scope.”
Collaboration additionally gives the possibility to instantly gain access to thousands of H100 and H200 GPUs and establishes Together AI and Hypertec Cloud as AI infrastructure leaders.
With this collaboration, engagements, and innovators globally can fulfill the demand for quick and efficient AI model deployment for mass applications.
Source: https://www.together.ai/blog/nvidia-gb200-together-gpu-cluster-36k
What is your reaction?
