Along with plans to make Nvidia’s most recent chips available to users, Amazon’s AWS cloud division has announced new chips that customers can use to develop and run artificial intelligence applications.
With a range of affordable options, Amazon Web Services is attempting to differentiate itself as a cloud provider. It will do more than just market low-cost Amazon-branded goods, though. Amazon’s cloud will offer premium goods from other vendors, such as highly sought-after GPUs from leading AI chipmaker Nvidia, just like its online retail marketplace.
Since startup OpenAI debuted its ChatGPT chatbot last year and amazed people with its ability to summarize information and write text that looks like it was written by a human, the demand for Nvidia GPUs has increased dramatically. As businesses scrambled to integrate similar generative AI technologies into their products, it caused a shortage of Nvidia’s chips.
Microsoft, the leading rival in cloud computing, may be weakened by Amazon’s two-pronged strategy of producing its own chips and granting users access to Nvidia’s most recent chips. Microsoft used a similar strategy earlier this month when it unveiled the Maia 100, the company’s first AI chip, and announced that Nvidia H200 GPUs will be available in the Azure cloud.
The declarations were made on Tuesday during the Reinvent conference in Las Vegas. In particular, AWS stated that it will provide access to the most recent H200 AI graphics processing units from Nvidia. The general-purpose Graviton4 processor and its new Trainium2 artificial intelligence chip were also announced.
OpenAI trained its most sophisticated large language model, GPT-4, on an H100 chip; the new Nvidia GPU is an improvement. There’s a lot of demand for renting out the chips from cloud providers like Amazon because big companies, startups, and government agencies are all fighting for a limited supply of them. The H200 will produce output almost twice as quickly as the H100, according to Nvidia.
AI chatbots such as OpenAI’s ChatGPT and its rivals are powered by AI models, which are trained on Amazon’s Trainium2 chips. The OpenAI rival Anthropic, backed by Amazon, and startup Databricks intend to develop models using the new Trainium2 chips, which will provide four times the performance of the original model, according to Amazon.
Based on Arm architecture, the Graviton4 processors use less energy than AMD or Intel chips. With 30% more performance than the current Graviton3 chips, Graviton4 is expected to enable better output for a lower price, according to AWS. Central bankers have raised interest rates due to unusually high inflation. Consider switching to Graviton if your company wants to continue using AWS but cut costs in order to better manage the economy.
According to Amazon, more than 50,000 AWS users are currently utilizing Graviton chips.
Lastly, AWS announced that it will run over 16,000 Nvidia GH200 Grace Hopper Superchips, which include Nvidia GPUs and Nvidia’s Arm-based general-purpose processors, as part of its expanding partnership with Nvidia. This infrastructure will be available to both AWS customers and Nvidia’s own research and development team.
Since releasing its EC2 and S3 cloud computing and data storage services in 2006, AWS has introduced more than 200 cloud products. Not all of them has succeeded. A select few are discontinued and some go for extended periods without updates, allowing Amazon to reallocate resources. Nonetheless, the business keeps funding the Trainium and Graviton initiatives, indicating that Amazon recognizes market demand.
AWS did not provide a release schedule for instances of virtual machines running on its Trainium2 silicon or with Nvidia H200 chips. Prior to their eventual commercial release in the coming months, customers can begin testing Graviton4 virtual machine instances right away.