Jon Peddie has been a graphics market analyst for about 35 years. We first met back when I was working on the TMS340 graphics processor family and the Video-DRAM (VRAM) in the mid-1980. Jon just published a retrospective article about the TMS34010 and VRAMs. I wanted to add some background information to his article and an old Byte Magazine article that gives some perspective as to the graphics accelerator board market of that era.
Introduction – My Background Graphics and Memory
I have lived two somewhat different lives in high tech. My first 20 years were with Texas Instruments where I became the youngest TI Fellow in the history of the company. At TI, I worked on the architecture and logic design of the TMS9918 VDP (the first “Sprite Chip”), the TMS9995 and TMS99000 (16-bit microprocessors), the TMS34010 and TMS34020 (the first programmable graphics processors), and the TMS320C80 (image processor with, four VLIW DSPs and a RISC CPU on a chip).
The TMS9918 design started in 1977 was copied by the display chips used in the early Nintendo and Sega game systems and was the first consumer device interfacing directly with DRAMs. One of my first jobs was to figure out the timing to interface with DRAMs, which before the Synchronous DRAM (SDRAM) had very complex timing with over 30 critical timing parameters. The TMS34010 graphics processor was the first fully programmable processor to directly interface with DRAMs and the first processor to use the VRAM (a DRAM with a high-speed shift register and the precursor to today’s GDRAM and the SDRAM). The TMS320C80 was the first processor to directly interface with the Synchronous DRAM.
While we had to use DRAM to support the large memory demands of graphics processors, the bandwidth of DRAMs was becoming too slow to keep up with our needs. This caused me to work with TI’s memory division on the first practical VRAM in the early 1980s. My first-hand experience with the problems of interfacing to DRAMs and the work with VRAMs, in turn, led to my helping out TI’s Memory Group on the development of the Synchronous DRAM. In many ways, today’s SDRAM and GDRAM are direct descendants of the VRAM.
BTW, I make the distinction of the first “practical VRAM” because the idea of putting a shift register on a DRAM for supporting video display has multiple overlapping claims of invention including TI, IBM, and AT&T. What my team did was make a series of improvements to the basic concept to make it work in a system and help pushed the idea out into the market.
TMS34010 Graphics Cards Circa 1989 in Byte Magazine
It so happens, I’m in the process of going through many thousands of pages of old articles about work I did back in my days at TI and I came across a November 1989 article, “The Brains Behind the Graphics” in Byte Magazine about graphics cards based on the TMS34010. I thought this reference would make a good companion to Jon Peddie’s retrospective.
While few today would know it, in the mid to late 1980s and early 1990s, before NVidia (founded 1993) and in the early days of ATI (founded 1985), the TMS34010 family dominated graphics cards used in Photoshshop® and CAD as discussed in the Byte article (see quote from the article below).
TI’s TMS34010 is probably the most popular graphics processor in the current PC market. Its powerful graphics capabilities have earned it a place in every thing from frame grabbers to printerByte Magazine Nov. 1989 “The Brains Behind the Graphics,” by Steve Apiki, Howard Eglowstein, and Rick Grehan
controllers, including, of course, display controllers.
I recently installed at NVidia Quadro P4000 board in my PC for doing some CAD work that costs about $750. I had forgotten how much graphics cards cost back in 1989 and was reminded by the article. Below is a table showing the various cards you could buy back then for doing Photoshop and CAD. The “Video Memory” was VRAM and the “Additional Memory” was either left over space in the VRAM or additional DRAM on the graphics card. The adjustment for inflation between 1989 and January 2019 is 2.03x, so double the prices below for comparison. Also note, we were talking in terms of one to two megabytes of memory back then and frame buffers were mostly 8-bits per pixel with a “color palette” (RAMDAC) to convert to red, green, and blue because memory was so expensive.
Off to CES Next Week
I’m heading to CES and my schedule is almost fully booked. I will be spending most of my time in the AR area of South Hall and in the Sands Expo along with some private meetings in the various hotels.