With the T3D, Cray research introduced a new generation of Massively Parallel Processor (MMP) super computers.
The T3D integrated from 32 to 2048 processing elements (PE's). Each PE was composed of a DEC 21064 RISC chip, also known as DEC Alpha, with its own memory area, memory controller, and prefetch queue.
The DEC Alpha contained a floating point unit, an integer unit, both with 32 64-bit registers, a 128-bit bidirectional bus, and both instruction and data caches.
The PE's were combined in nodes of 2 PE's running at 150 MHz, along with two to 64 megabytes of memory and a network switch. The nodes were connected in a 3-D torus network configuration.
Memory was physically distribued among the processors, but was globally addressable (any processor could address any memory location in the system.
The global peak performance of a T3D was from 19 (128 PE's) to 307 (2048 PE's) GigaFlops (billions floating-point operations per second).
The system ran UNICOS MAX, the Cray Research's operating system. Each processor ran a minimal microkernel to handle frequently used functions, interprocessor communication and memory management. Major programming language were C and Fortran compilers
The T3D was tightly coupled to a CRAY C90 or CRAY Y-MP host system through a high speed channel, making host's resources (I/O functionality, I/O subsystem, high-speed disks) directly available to the T3D.
The Cray T3D was replaced by the T3E, the first supercomputer to sustain one teraflop (1 trillion F.P. calculations per second).