Google Cloud Next ’24: Upgrades to AI Hypercomputer, workload infrastructure announced

Google has announced that it is bringing some significant performance improvements to its AI Hypercomputer architecture. The company also launched the 4th gen C and N machines that help optimise AI workloads.

By: Tech Desk

New Delhi | Updated: April 10, 2024 08:23 AM IST

3 min read

Google Cloud Next 24 | Google AI ypercomputer | Google Cloud AI

NVIDIA's recently launched Blackwell GPUs will also be available on Google Cloud. (Image Source: Google)

Google has announced that it is bringing numerous enhancements to its AI Hypercomputer architecture. The company says services like Character.AI use the same architecture to meet its processing demands.

The tech giant also seems to be bringing the recently launched NVIDIA Blackwell GPUs to Google Cloud, which will be available in two configurations. While the HGX B200 GPU-powered VM is designed to handle the most intensive AI, data analytics and HPC workloads, the liquid-cooled VM powered by the GB200 NVL72 GPU offers enhanced real-time LLM inference and the ability to train models with up to 1 trillion parameters.

On the software side of things, the company introduced JetStream – an inference engine for large language models that offers higher performance per dollar and MaxDiffusion – a new high-performance and scalable interface for diffusion models Google Distributed Cloud also got a new generative AI search package solution powered by Gemma, which offers flexible deployment options.

The tech giant also announced its first-ever ARM-based CPU called Google Axion which combines “Google’s silicon expertise with Arm’s latest compute core designs to deliver up to 30 per cent better performance than current Arm-based instances, 50 per cent faster performance than current generation x86-based VMs while improving the energy efficiency by 60 per cent.

Also Read | OpenAI models trained on over million hours of YouTube videos: Report

To optimise AI workload, Google has launched the fourth generation C and N machine families powered by the 5th Generation Intel Xeon processors, which is an industry first from a cloud hyperscaler. These new machines are designed to support general-purpose workloads and offer a balance between performance, flexibility and cost.

Google claims C4 offers “industry-leading price-performance and enterprise-grade tooling”, whereas N4 is designed to handle performance for general-purpose workloads that do not require peak processing power all the time.

While C4 VMs are an ideal fit for databases and caches, network appliances, high-traffic web services and online gaming, N4 VMs help with “cost optimisation and price-performance through an efficient architecture of streamlined features, shapes, and next-generation Dynamic Resource Management.”

Story continues below this ad

The company also announced the general availability of the latest Hyperdisk capabilities that let users purchase and manage block storage capacity in a shared workload pool and said it is the first and only cloud hyperscaler to offer the ability to pool storage.

Tags:

Google Google Cloud