The internal code name for this project by was "Hawk". The commercial name Hawk sold under was "GT". Why a two character product name? Very simple: in those days the Sun internal product option ordering software could only handle two character labels (names).
The initial goal for the GT was to use the TACC-2 board (still under design) as the front end board for the GT. For a variety of reasons this was not really feasible, though some concepts were shared. External vendors supplied custom CMOS circuits as well as back-end support, chip fabrication, and testing. This machine also used a lot of standard chip logic. On the software side, other teams supplied OS drivers, XGLTM drivers and support, plus custom PHIGS+ support, and extensive test cases. My role on GT was mostly chief architect; Mike Lavelle was the project lead, and specified the details of many of the chips. Brian Schroder (who has been the design lead on the Applicon implementation of the Triangle Processor) implemented the entire front end DMA chip, Scott Nelson made many contributions, most prominently his design of the antialiased line rendering algorithm; many other people contributed.
The GT was my first Sun 3D graphics accelerator product. When I joined the project it had already been partially designed; I always felt the same way about the GT as Shakespeare did about the play "Marcus Andronicus". (He took over writing the play part way, and forever after has mixed emotions about it.)
At the time of release the GT was the only product in its (price) class the supported true double-buffered 24-bit RGB (8-bit red, 8-bit green, 8-bit blue) plus a 24-bit Z-buffer. (Competing products at the time could support single buffered 24-bit color; when double buffered (animated) they had to drop to 12-bit dithered color (4-bit red, 4-bit green, 4-bit blue)). The GT supported true square pixel stereo video (960x600 @ 108 Hz) and would run the window system in stereo, all but the highest end SGI product was still only supporting the 2-1 non-square over/under 1280x512 stereo format.
The competition for the GT wasn't really SGI, they were still too small at the time and without as much presence in the lucrative mechanical CAD/CAM market. The real competitors were HP and IBM.
Physically the GT consisted of a three board set using the VME form factor; the GT was housed in a separate tower (a re-used Sun deskside workstation enclosure), a ribbon cable connected the GT to driver card in the SPARCstation2TM. The GT used a lot of discrete chips (management was still leery about VLSI), was expensive to produce, and only sold in limited volumes. However, the limited numbers were sold into several major mechanical CAD/CAM end users (automotive, aerospace, consumer design) and the associated CAD/CAM software suppliers. The GT greased the skids for the Sun ZX to roll in a few years later in a major way. In many ways the ZX was "the GT done right": all custom chips except the VRAMs. The GT established that Sun could build full function shaded 3D graphics accelerators, and was also the basis for all my early Sun virtual reality work.
The GT did not have an architecture paper published at SIGGRAPH. Some of the patents describe portions of the overall architecture.