As the need for running massive AI models grows, Nvidia, the industry leader in high-end processors for generative AI, will introduce a chip that is even more potent.
The GH200 super processor, which Nvidia claimed can handle the most complicated generative AI workloads, covering massive language models, recommender systems, and vector databases, was made available, the company announced.
The GH200 will have the same GPU as the H100, Nvidia’s most powerful and popular AI solution at the moment, but three times the memory capacity. According to the business, GH200-powered systems will begin operating in the second quarter of 2024.
The H100 line presently costs about $40,000, while Nvidia withheld the GH200’s price.
For the system to be able to perform the computations required to generate text or a picture of a horse in the style of Banksy, complex AI models need powerful GPUs. It takes a lot of computing power to run these models, and some even need to “break up” the models among different GPUs to operate, even with Nvidia’s H100 chips.
In terms of GPUs with generative AI capabilities, Nvidia practically controls the market. To differentiate themselves, cloud service providers like AWS, Azure, and Google all utilize Nvidia’s H100 Tensor Core GPUs and offer extra services to clients to enable them launch projects utilizing large language models.
Despite Microsoft purportedly wanting to develop AI processors, Nvidia and Microsoft collaborated to create new supercomputers.
AMD, which plans to increase production for its own AI GPU in the fourth quarter of this year, is another rival to Nvidia.
