VOGONS


History of nVIDIA Graphics cards Vol. 2 GPU competition

Topic actions

Reply 81 of 84, by spiroyster

User metadata
Rank Oldbie
Rank
Oldbie
imi wrote on 2021-11-01, 13:15:
Gmlb256 wrote on 2021-09-19, 17:14:
yjfy wrote on 2021-09-19, 16:56:

GPU originally refers to the display chip that supports hardware T&L, Kyro/Kyro II does not support hardware T&L, so it is not listed as GPU.

Indeed, most people doesn't really have any idea about the GPU term anymore. I see it being applied to an entire video card regardless of the capabilities. I believe paradigital mentioned it due to being released around 2000-2001 timeframe.

the term GPU as "graphics processor unit" has been coined in the 70s already for chips dedicated for calculating graphics and you can find multiple mentions of it in computer magazines of the 70s and 80s already.

and I don't see any issue calling graphics/video cards "GPUs" as it's clear what is meant, a card with a GPU on it.

I'm not familiar with anyone using the term GPU before nVidia with the release of GeForce. Do you have a source for this? Any links of anything that called a graphics card a 'GPU' before Geforce came out? (outside of wikipedia that is)

To my knowledge, originally they were called 'graphics processors', 'graphics chips', 'video chips' or 'graphics accelerators' etc as there was not 1, but usually many which collectively made a Graphics System (or Graphics subsystem) and usaully had a dedicated purpose which together implemented a 'Graphics pipeline'.

What we call TnL (transform, clipping lighting) was originally done by a Geometry Engine (GE), which wasn't originally a single chip, rather multiple processors, and in the case of early GE (GE4), had many FP processors (Weitek 3332) each processing a single stage of the Geometry pipeline. These were never called GPU's though, since it wasn't a single 'unit' like a CPU is, and their roles were clearly defined.

On PC's, solutions like Voodoo did not perform the TnL aspect, rather let the CPU do it, which left the Graphics card simply being a glorified rasterizer. Other solutions which accelerated this were rarely a single chip doing everything of the pipeline (apart from DAC) and relied on distinct stages (although there were examples, high-end CAD, although these were never called GPU's either... afaik). TnL was probably the last aspect of the traditional immediate mode graphics pipeline on main stream PC's that wasn't accelerated. Thats essentially what nVidia did with GeForce.

Arcades called them 'Video boards' or sometimes 'graphics boards', but again these were made up of many DSP's and later dedicated FPU's for 3D, never a single unit and neither Namco or Sega (originators of 3D acceleration in the Arcades) called them GPU's thats for sure.

NitroX infinity wrote on 2021-11-01, 21:10:

3DLabs' Wildcat VP series seems absent from this timeline.

They came out in 2002, although they should rightly claim the GPGPU crown since thats what the 'VPU' techincally did. A number of years before the unified shader architecture became a thing.

Reply 82 of 84, by imi

User metadata
Rank l33t
Rank
l33t
spiroyster wrote on 2021-11-01, 21:42:

I'm not familiar with anyone using the term GPU before nVidia with the release of GeForce. Do you have a source for this? Any links of anything that called a graphics card a 'GPU' before Geforce came out? (outside of wikipedia that is)

To my knowledge, originally they were called 'graphics processors', 'graphics chips', 'video chips' or 'graphics accelerators' etc as there was not 1, but usually many which collectively made a Graphics System (or Graphics subsystem) and usaully had a dedicated purpose which together implemented a 'Graphics pipeline'.

I mean the wikipedia article cites two papers mentioning GPUs but those don't seem to be easily accessible, but if you check google books with a quick search you'll find articles as early as '76 i.e.
https://books.google.com/books?id=Jzo-qeUtauo … processing+unit

yes, we called graphics cards and chips by many different names, and nvidia popularized the term in consumer space, but they did not make the "first GPU" and GPU was not "originally" used only for chips with hardware TnL ^^

Reply 83 of 84, by spiroyster

User metadata
Rank Oldbie
Rank
Oldbie
imi wrote on 2021-11-01, 23:05:
I mean the wikipedia article cites two papers mentioning GPUs but those don't seem to be easily accessible, but if you check goo […]
Show full quote
spiroyster wrote on 2021-11-01, 21:42:

I'm not familiar with anyone using the term GPU before nVidia with the release of GeForce. Do you have a source for this? Any links of anything that called a graphics card a 'GPU' before Geforce came out? (outside of wikipedia that is)

To my knowledge, originally they were called 'graphics processors', 'graphics chips', 'video chips' or 'graphics accelerators' etc as there was not 1, but usually many which collectively made a Graphics System (or Graphics subsystem) and usaully had a dedicated purpose which together implemented a 'Graphics pipeline'.

I mean the wikipedia article cites two papers mentioning GPUs but those don't seem to be easily accessible, but if you check google books with a quick search you'll find articles as early as '76 i.e.
https://books.google.com/books?id=Jzo-qeUtauo … processing+unit

yes, we called graphics cards and chips by many different names, and nvidia popularized the term in consumer space, but they did not make the "first GPU" and GPU was not "originally" used only for chips with hardware TnL ^^

Fair enough, can't argue with that. Certinalty 'GPU' was not a common phrase until GeForce came along. I think the 'Unit' phrase was important, but granted 'unit' could apply to descriptions of the subsystems prior to the phrase becoming mainstream I guess.

Reply 84 of 84, by Wes1262

User metadata
Rank Member
Rank
Member
yjfy wrote on 2021-09-15, 03:55:

Such an amazing thread! Love all the info and history.

I spotted something strange though. You write that the Quadro 2 pro and the Ultra have identical cores (except that the first is binned).
But in the attached table the Quadro shows as a 150nm productive process, while the Ultra is 180nm. What gives?

Thanks!