In a highly anticipated move, Nvidia—known for its dominance in high-end processors for generative AI applications—has teased the release of an even mightier chip. The relentless demand for larger and more robust AI models is driving this tech evolution.
Introducing the GH200 Super Chip
Nvidia’s new offering, aptly named the GH200 super chip, has been engineered to manage some of the most intricate generative AI tasks. These range from expansive language models to recommenders and vector databases.
Comparing GH200 to its Predecessor
Building on the legacy of the H100, currently the most revered AI chip from Nvidia’s arsenal, the GH200 comes armed with the same GPU. However, what sets it apart is its tripled memory capacity. Users eager to harness its power won’t have to wait long, with systems running on the GH200 slated for a Q2 2024 release.
As of now, the price tag for this technological marvel remains under wraps. To provide some context, Nvidia’s current top-tier, the H100, is priced at around $40,000.
Why the Need for More Power?
Sophisticated AI models command powerful GPUs to perform complex tasks, whether it’s crafting text or generating images, like a horse painted in the unique Banksy style. These operations demand colossal processing strength. Even with the robust H100, certain tasks necessitate distributing the load across multiple GPUs for seamless execution.
Nvidia’s Dominance and the Cloud Connection
Nvidia’s near monopoly in the AI-capable GPU domain is evident. Industry giants such as AWS, Azure, and Google lean on Nvidia’s H100 Tensor Core GPUs, adding unique services to facilitate users in deploying large language models.
Moreover, Nvidia has teamed up with Microsoft for new supercomputer projects, despite whispers of Microsoft’s plans to produce AI chips.
Facing Off Against AMD
While Nvidia might be at the forefront, competition isn’t far behind. AMD is prepping to accelerate its AI GPU production, eyeing a Q4 launch this year.