Level up your AI and HPC Applications with NVIDIA H100 GPUs

In this article you read,

• what you can expect from NVIDIA’s new H100 GPU generation,
• which role they play for artificial intelligence,
• and when the Open Telekom Cloud will introduce H100 GPUs.

As the last quarter of 2024 is already underway, the Open Telekom Cloud team is getting ready to provide a special Christmas gift to their users: The NVIDIA H100 Tensor Core GPUs with their increased computing power will be available at the end of the year.

Anyone who is in Artificial Intelligence (AI) and Machine Learning knows: GPUs are at the heart of development, training, and operations of pretty much any AI service. As AI models grow, get better and diversify, the hunger for GPUs grows immensely – and given the high price and operations costs (power) for a H100 GPU, these high-performance devices are best used on demand from the cloud.

In December 2024, the Open Telekom Cloud is enhancing their GPU portfolio with the H100 as the “p5 ECS” flavor family. This next-generation GPU delivers unprecedented performance, scalability, and efficiency for a wide range of demanding workloads, from AI training and inference to high-performance computing and data analytics. The Open Telekom Cloud will add more than 4 petaflops additional GPU computing power in December to support their customers’ rising needs.

AmberSearch is a German provider of AI services and a user of Open Telekom Cloud resources. Their amberSearch team sees new possibilities with the arrival of the H100 GPUs: “For us, Telekom was the best German partner from the very beginning to provide a secure cloud infrastructure for the operation of our AI solution,” managing director Philipp Reißel explains.

According to NVIDIA, the new GPUs generation can accelerate exascale workloads considerably. The combined technological innovation speeds up Large Language Models 30-fold (AI inference test on a Megatron Chatbot with 530 billion parameters) and empowers model training by a factor of 4 compared to A100 GPUs (GPT 3).

Customers will be able to select from four different flavors of ECS instances, leveraging one single GPU or up to eight GPUs in one virtual machine.

Details about the pricing are available in the service description. More details will be published with the ECS release notes, once the new instances are available.

Get ready to elevate your cloud solutions with NVIDIA’s most advanced GPU technology!