Nov 14 (Reuters) - Silicon Valley artificial intelligence (AI) computing startup SambaNova Systems said on Monday it delivered eight units of its latest AI system to the U.S. Argonne National Laboratory, which is expanding its AI offering to researchers.
Each of the systems has eight DataScale SN30 chips, launched in September, and triples the speed of the previous generation.
With AI work taking center stage in research, Argonne National Laboratory has been testing out various AI chips and systems, Rick Stevens, associate lab director for computing, environment and life sciences at Argonne lab told Reuters.
In addition to SambaNova, AI systems from startups including Cerebras Systems, Groq Inc, Graphcore, and Intel Corp (INTC.O) owned Habana Labs have been tested.
"We're working with new emerging AI hardware architectures, and we get early hardware and then we play with it, we use it on our science applications," said Stevens.
"When we get to that point, we make internal decisions - 'Is this useful enough and performant enough and interesting enough that we should take the next step?' And so SambaNova has gotten to that milestone."
Steven said the lab was evaluating AI hardware for its next supercomputer to see if SambaNova and others can be included. The lab, which is a Department of Energy national lab run by the University of Chicago, is currently building a supercomputer dubbed Aurora that will be a two exascale speed machine - able to do two billion calculations in a second.
Stevens said the lab was in discussions with Cerebras and Habana about scaling up those systems as well. "What we try to do is understand the sweet spots of these different systems," he said.
SambaNova's first-generation AI system has been used to predict how tumors respond to various drug combinations, improve weather forecasting accuracy, and speed up fluid dynamics simulations, he said.
Our Standards: The Thomson Reuters Trust Principles.