3 ms·
Loving those early 3d hardware retrospections. Some corrections: >they started they own company typo >EOM'es only leverage on the cards they produced was the
by rasz 8y ago
Loving those early 3d hardware retrospections. Some corrections:
>they started they own company
typo
>EOM'es only leverage on the cards they produced was the RAM they selected (EDO vs DRAM), the color of the resin and the physical layout of the chips. Pretty much everything else was standardized.
- EDO is DRAM, EDO vs FPM? I didnt know that, always assumed every V1 shipped with EDO, just like every card on your pictures is EDO.
- video signal switching was also up to the vendor (relay/mux)
- and one of the cards has TV encoder section with TV out, pretty neat selling point
>It is not specified if the bus used address multiplexing or if the data and address lines were shared. Drawing it non-multiplex and non-shared makes things easier to understand.
You are addressing 512 kB, but datapath is 16 bit/2 Byte wide, so we only need 18 bit address bus. As for multiplexing thats not how DRAM addressing works IRL (understandable misconception/simplification for non EEs). Row/Column address lines are multiplexed, meaning we are down to 9 address pins +OE/WE. Comes down to ~110 pins assuming full 4 way interleaving. Seems doable with >200pin ASICs.
>21-bit address generates two 20-bit where the least significant bit is discarded to read/write two consecutive pixels.
still too many bits, 2MB at 4 byte granularity is 19 overall, 18 per 1MB bank
>TMU was able to perform per-fragment w-divide
This was a HUGE deal at the time, and achieved by doing serious low level optimizations/tricks (lookup tables/approximation if I remember Oral Panel correctly). 3dfx engineers were big fans of good enough hacks vs slow but correct way of doing things. Another one was color dithering, too bad you didnt mention "24-bit color dithering to native 16-bit RGB buffer using 4x4 or 2x2 ordered dither matrix" - this is the reason straight ram dump screenshots from Voodoo1 dont really look the same as on directly connected monitor. 3dfx called it ~22bit color, it was noticably better than Nvidias pure 16bit.
Btw afaik Quake pushed somewhere between 500-1000 polygons per frame, earlier games like Actua Soccer rarely went up to 500 with fatal consequences of single digit framerate on S3 Virge. You might enjoy Profiling Of 3 Games Running On The S3 ViRGE Chip http://www-graphics.stanford.edu/~bjohanso/index-virge-study.html http://www-graphics.stanford.edu/~bjohanso/index-virge-study...