Machine type: RISC-based distributed-memory multi-processor.
Models: T3D.
Operating system: UNICOS MAX (micro-kernel Unix).
Connection structure: 3-D Torus.
Compilers: CFT77_M (Fortran 77 with extensions), C.
System parameters:
Performance:
Note: The values for 
and 
are for a 1,024 node system.
The T3D is the first generation of DM-MIMD systems from CRI. The name T3D refers to its connection structure: a 3-D torus. The system needs one of Cray's larger vectorprocessors as a front-end. A Cray Y-MP/E or Y-MP C90 or M90 will do, but not a Y-MP EL because it lacks the Cray proprietary low-speed (6 MB/s) and high-speed (200 MB/s) channels which are required for coupling with the back-end system.
The computational tasks are executed by DEC alpha MC21064 RISC processors. Cray stresses, however, that the processors are encapsulated in such a way that they can be exchanged easily for any other (faster) processor as soon as this would be available without affecting the macro-architecture of the system.
Each node in the system contains two processing elements (PEs) which in turn contain a CPU, memory, and a communication engine that takes care of communication between PEs. The bandwidth between nodes is quite high: 300 MB/s. Furthermore, care has been taken to hide the latency to a large extent. There is a prefetch queue for asynchronous fetches of small messages and a remote store that allows direct writing in memory addresses of non-local PEs with very low latency. Another device that can be used for general data re-distribution is the Block Transfer Engine. It should make rearrangement of data, for instance in case of matrix transposition, a relatively painless operation. The T3D has hardware support for fast synchronisation. E.g., barrier synchronisation takes only one cycle per check.
Presently the I/O is handled by the front-end processor. Cray intends to make the second generation stand-alone by adding I/O-nodes to the T3D itself.
The T3D supports various programming models. Apart from PVM 3.x for message passing and HPF for data distribution, a Cray proprietary work sharing model, called CRAFT, can be employed. Cray views HPF and Fortran 90 array syntax as subsets of the CRAFT model. Within this model data can be exchanged implicitly, thus looking effectively as a shared-memory system to the user.