Thinking Machines Corp of Cambridge, Massachusetts has announced shipments of its new CM-5 vector chips with installations due next month: the company has taken delivery of 100,000 MFLOPS-worth from Texas Instruments Inc and it says that on a three-dimensional irregular finite code, used in aerodynamics, it has achieved performances on a 1,024 node configuration seven times faster than the largest CM-2; each parallel node of a CM-5 can have four vector units, with peak speeds of 32 MFLOPS; on matrix-vector multiply operations the system runs at around 61% of peak; operations taking place within the node itself run at 95% of peak.