By
Godfrey Benjamin
April 26th, 2024
Nvidia has unveiled the H200 chips, which is a significant improvement from H100, and is expected to be available from Q2 2024.
Software and fabless giant Nvidia Corporation (NASDAQ: NVDA) has unveiled a new graphics processing unit called the H200. A new GPU upgrade from the H100, the Nvidia H200 is designed to cater to artificial intelligence (AI) models that are currently the crux of the ongoing push for AI.
According to an official announcement, the H200 can handle large amounts of data for generative AI and process heavy computing workloads. Nvidia says that the H200 is the first GPU with HBM3e, providing a faster and larger memory for handling generative AI and large language models (LLMs). The new GPU can deliver 141 GB of memory at 4.8 terabytes per second. On the need for new and improved chips, Nvidia’s vice president of hyperscale and HPC, Ian Buck, said:
“To create intelligence with generative AI and HPC applications, vast amounts of data must be efficiently processed at high speed using large, fast GPU memory. With NVIDIA H200, the industry’s leading end-to-end AI supercomputing platform just got faster to solve some of the world’s most important challenges.”
Training AI systems and models requires thousands of these chips working together. OpenAI used the H100 to train its GPT-4, its most advanced LLM.
Nvidia to Launch H200 Next Year and Continue H100 Production
There is currently no indication of how much the H200 will cost. However, it is expected to be well more than the average $30,000 price point for the H100. According to Nvidia spokesperson Kristin Uchiyama, the company’s partners set the pricing.
Nvidia expects that the H200 will be about twice as fast as the H100. According to the announcement, the new GPU will be available from cloud service providers and global system manufacturers in the second quarter of 2024. Nvidia named Google Cloud, Amazon Web Services, Oracle Cloud, and Microsoft Azure, as some of the first cloud service providers that will deploy the chips from next year. Other named cloud providers are Vultr, Lambda, and CoreWeave.
Furthermore, the H200 GPU will be compatible with hardware and software currently running H100 systems. This means that organizations can deploy the H200 in any type of data center, without having to dispose of their current H100-compatible server systems.
Uchiyama has noted that Nvidia does not intend to pause H100 production. According to the company spokesperson, Nvidia will continue to add overall supply throughout 2024.
Last month, Coinspeaker reported that Nvidia quietly began designing central processing units (CPUs) for Windows PCs. Nvidia will reportedly use technology from Arm Holdings Ltd, the chip designer owned by Softbank Group. The chips are set for launch in 2025.
Nvidia Partnerships
Nvidia is also working on several partnerships, and is collaborating with Google Cloud to deploy new AI software and infrastructure. In addition, the giant chip designer partnered with electronics manufacturing giant Foxconn to build AI-run data centers for different purposes, including use cases for electric vehicles. Nvidia says that these AI factories will help digitize the manufacturing and inspection processes for robotics and electric vehicles. The plan is also for Foxconn clients to use the AI service to train industrial robots and autonomous machines.
