Tech Firms Squeeze More Out of AI Chips
Looking to make the most of the AI computer chips they have, technology companies are increasingly “turning to software that can squeeze more performance out of available chips and help reduce costs,” The Wall Street Journal (subscription) reports.
What’s going on: While larger firms are using hard-to-get graphics processing units to build multiple different AI models that “do things such as detect cybersecurity threats and help improve network performance,” other businesses are using central processing units to do similar tasks.
- CPUs aren’t as powerful as GPUs, but they are easier to find.
- And when “tuned with open-source software tools to get more performance out of them,” CPUs can help businesses meet their processing needs.
Boosting performance: As GPU demand continues to outpace supply, companies are using third-party software to squeeze additional performance from existing GPUs, too.
- One Israeli start-up installs optimization software on client GPUs to “automatically put idle computing power to use to gain better processing efficiency,” according to The Wall Street Journal.
- A Seattle-based startup “is betting that most businesses won’t want to deal with owning and managing an array of AI hardware … so it rents out access to processing power from cloud providers that it speeds up on customers’ behalf.”
Renting the cloud: Indeed, cloud-company giants can offer access to much-needed processing power “by renting it out as they do with computer services,” one source told the Journal.