Otdfbt

Plain and simple, we know that consumer level 3D acceleration devices roughly appeared in the middle of the 90s, with brands like Matrox, 3DFX, Nvidia and so on.

However, such technology for sure must have been available way before for professionals, governments or even military.

When and what was the first 3D acceleration device ever released for a computer?

To clarify after Stephen Kitt comment,

By 'device' I more or less mean any approach or solution that significantly improved the speed of a 3D simulation, when say, you would compare it to the performance of a software rasterizer on a desktop computer of that time.

If I could give an example it would be like when you used to play software rasterized 3D games with a low frame rate, against when you installed a 3DFX card, you instantly felt another level of performance/rendering.

The first appears to be Evans & Sutherland's LDS-1, introduced in 1969. The first one was delivered to Bolt, Beranek and Newman Inc., in August 1969. This was an analogue vector system, with depth cuing, and could draw and manipulate complex wireframe models in real time.

The first 3D engine produced as a VLSI microchip was the Geometry Engine, developed at Stanford University and commercialised by SGI from the early 1980s.

SGI's failure to address the Windows/Intel market led various SGI staff to leave and found Nvidia or join ATI in the 1990s, laying the foundations for today's market, where integrated geometry engines are a standard part of display systems.

You can't really answer this question without acknowledging the parallel development paths for high-quality 3D graphics, which includes not only the dominant Polygon/Texture-mapping approach, but also the Ray Tracing approach. This parallel development path continues through today, with recent announcements like Nvidia's next generation RTX platform, that features real-time Ray Tracing.

While the LDS-1 is a very early example of being purpose-built for 3D graphics, it was incapable of producing anything photo-realistic or "natural". At roughly the same time, Ray Tracing was being used for high quality static images that, while taking many hours to render, produced results with a level of quality that would still impress 3D artists today.

In 1982, Osaka University created a purpose-built supercomputer designed for 3D graphics rendering using Ray Tracing. The supercomputer was known as the LINKS-1 Computer Graphics System, and was an early contender for fastest computer in the world. It is briefly described on Wikipedia as:

a massively parallel processing computer system with 514 microprocessors (257 Zilog Z8001's and 257 iAPX 86's), used for rendering realistic 3D computer graphics with high-speed ray tracing.

The Japanese supercomputer predates SGI's commercialization of the Geometry Engine by 5 years, and could produce far better 3D graphics.

Of course, both supercomputers and SGI systems were far out-of-reach for users of Desktop computer systems of the 1980s. However, Ray Tracing enjoyed a brief popularity at this time. Most 3D artists relied on affordable Amigas with accelerator boards that included FPUs to render 3D graphics and animations using ray tracing. The Macintosh II, with its Motorola 68020/68881 CPU/FPU, was also popular for this purpose. There is available a tremendous amount of early 3D desktop computer content created with such equipment, using both ray tracing and polygon rendering techniques. In a sense, any 1980s machine with an FPU was an "accelerated 3D graphics workstation", if 3D rendering software made use of it. My own Amiga 1000 had a 68881 FPU attached via the side-expansion bus for just this purpose, in 1988, at the same time SGI was selling the IRIS 3000-series.

In the mid-1980s I worked for a company that was using satellite imagery as one of the tools geologists used to do studies for firms searching for oil. The computer+software system we used was called Earthviews, based on an LSI-11/73 CPU with an RSX-11 operating system. (The actual software we used most was a FORTH program called AIMS ("Advanced Image Management System"), which ran separate from RSX-11 -- i.e. it was its own operating system.) Physically the system was quite large, consisting of two tall 19"-wide racks and a tape-drive system just as large. One of the optional hardware features was in one of the racks, about 18" tall, called a "vector processor". I believe the system was developed in the early '80s, maybe even late '70s. If I remember correctly, what it did was calculate a stream of a large number of multiply-plus-add operations in parallel.

While this hardware peripheral was not actually dedicated to be a "3D accelerator", it could operate as a major component of such a subsystem. I'm not sure, but I believe what Earthviews used it for was to process what was essentially sonar data from appropriate ground scans. I don't remember this very well, because our company didn't use sonar data. I think if you had the data, Earthviews would build a 3D rendition of the ground the sonar passed through.