Cloud computing GPUs have become popular especially with the emerging Artificial Intelligence (AI). But since demand has outdone supply, cloud providers are embracing the challenge by designing new chips that can fill the gap. These custom accelerators are intended to optimize specific workloads for better and cheaper solutions than those offered by cloud computing GPUs.
Probably the biggest reason why custom chips are becoming popular is because of the issues with GPUs. However, GPUs are critical, especially in training AI models, but they have some severe issues like high power consumption, heat dissipation, and the global shortage of GPUs that affect many industries. Nvidia, for instance, has its GPU pipeline filled for the next year, which underlines the necessity of having more options. As a result, cloud providers such as Amazon web service, Google, and Microsoft are developing custom chips to address emerging market requirements of their clients.
At Microsoft’s Ignite conference the company introduced two new custom chips built specifically for its Azure cloud computing platform. This is a new development as Microsoft follows others such as AWS and Google who have moved to custom silicon with their solutions including AWS Trainium and Google’s TPUs. New chips by Microsoft, Azure Boost DPU, and Azure Integrated HSM work on the enhancement of data processing for AI workloads and security respectively.
The Azure Boost DPU is specially built for data processing with its own operating system and the Azure Integrated HSM is designed to provide additional security to encrypt and manage keys. This places Microsoft squarely in a league with Google’s E2000 IPU and AWS Nitro, which has already been deployed in cloud infrastructure.
Security is another major domain where implementation of custom chips have shown to be beneficial. For example, The Azure Integrated HSM chip in Microsoft helps to lessen the latency and also enhance the scalability of encryption jobs. Both AWS and Google have their own security custom chips; AWS has a Nitro system that does not allow any changes to the firmware without its permission and Google has Titan that sets up a root of trust to check the health of the system.
The drive for cloud computing GPUs and custom chips is not letting up. These innovations give the cloud providers the capacity to meet this ever increasing demand for specialized, high-performance computing that is required by the current market. The future of cloud computing GPUs  is to embrace more of new custom silicon, which will be the next frontier of cloud performance.
source:;Â cloudcomputing-news.net
