Its software was originally written with central processor units, or CPUs, in mind during an era when PCs were common. On Wednesday, Cadence announced that it has reworked many of those core programs to run on Nvidia’s latest “Blackwell” graphics processors, or GPUs. Cadence’s new Millennium M2000 supercomputer will contain about 32 of Nvidia’s newest chips and will cost about $2 million for the most common configurations, though final pricing has not been set, Cadence executives said Wednesday. It follows a supercomputer released last year that ran a more limited set of Cadence’s software.
The price will buy improvements in speed.
Michael Jackson, corporate vice president and general manager of the system design and analysis group at Cadence, said the company worked with Boeing on analyzing turbulence around parts of a 777 jet.
What would have taken eight days on a traditional CPU-based system took less than 24 hours on the new supercomputer, enabling engineers to either get the same work done in less time or use the extra time to carry out further design refinements.
Discover the stories of your interest
“There’s this insatiable need for faster simulation,” Jackson told Reuters in an interview on May 6. At an event hosted by Cadence in Santa Clara, California, Nvidia CEO Jensen Huang joined Cadence CEO Anirudh Devgan on stage and said Nvidia will purchase 10 of the new supercomputers to design the next generation of its chips and AI data center equipment.
“This is a big deal for us. We started building our data center to get ready for it,” Huang said. “We’ll speed it up 50, 60, 100 times,” Huang said the design work Nvidia will do with the machines.