the voodoo 5 9000, a neat concept with potantial or pure myth?

ScubaSteve404 · January 17, 2020

i got curious about sli and it's history again and wondered would it be possible to bring back and make better 3dfx's version of SLI( Scan Line Interleave ) with modern hardware? while old SLI has the benifet of potentialy working in most if not all games no matter what with almost perfect performance scaling it brings back a major issue, interleave vs progrsive resolution scaning aka X_Resolution_P vs X_Resolution_i. while interlaced video looked good enough as video unless there was high motion still frames from interlaced video will have issue with alignment of the "fields", while 3dfx's SLI didb't show halves and sequancial frames at the same time field alignment can still pose to be an issue because while it will be the one frame if the cards are synced the fields that make up the image may still be able to shift like a stack of loose paper, meaning evan if someone made something like a "voodoo 5 9000" would keeping timings tight and synced be enough to counter/prevent that or would the output be an incoherent scrolling mess? if somehow the major issues with old SLI could be fixed frame times/resalutions once thought absolutly insane could be made as easy to run as 1080P with a card as absolutly insane as the most likly fake "voodoo 5 9000" 4k( 3840 by 2160 ) reduced 32 times verticaly is 67.5 by 3840 which is a mere 259.2 thousand pixels vs the original 8.29 million of the orginal resolution, now imagine 67.5 by 3840 being rendered on 32 gpus at the same time and than stacked verticaly in order. i know this sounds absolutly insane but lets scale this back some, doing the same thing but with only 4 instead of 32, while yes thats still alot its not completly insane. 2160 divided by 4 is 540, 540 by 3840 is the same pixel count as 1080P. stack four of that verticaly and IF this can be done using four gpus 4K could be rendered with similar frame times as 1080P evan if it cost 4 graphics cards it would potentialy work with most if not all games provided the timing/sync issues could be solved vs nVidia's SLI which reqiures the game to directly support it and the frame time improvments varying wildly. and before anyone asks "how will you combine the four independent video signals into one, theres two ways i thougt of that could work. either it gets passed through each graphics card until eventualy it gets the the last one which will be the only one plugged into via a display cable or a sort of "display squid" could be made to plug into all of the cards and combine the signals that way and if nVidia has control over the pantents for old SLI they could intheory build this into the g-sync module in g-sync monitors meaning you'l only need four cables. i think one way to improve apon old SLI would to be intead of interleaving it GPU1, GPU2, GPU3, GPU4, repeat, each gpu gets one big block and it would be GPU1, GPU2, GPU3, GPU4 end frame instead of repeate till the frame was constructed, having the GPUs work in blocks instead of lines may help a ton with timing and or syncing becuase you just need to sync the first line of the next block with the last line of the previous. i think it be preatty cool though if someone made something akin to the "voodoo 5 9000" graphics card becuase the little poof of its existance show this MASSIVE beast of a card with 32 chips working in tandem evan if its fake a card like that if real would be in a luage all of its own. if that modification to 3dfx's version of SLI works i think i'd just call it something like "SLI 2.0" or think of some way to mention 3dfx in the name without making the name too long like "3DFX SLI 2.0"

Samfisher · January 17, 2020

Less of a TLDR, more of a wall of text didn't read. Please, please, use paragraphs.

GoldenLag · January 17, 2020

27 minutes ago, ScubaSteve404 said:

i got curious about sli and it's history again and wondered would it be possible to bring back and make better 3dfx's version of SLI( Scan Line Interleave ) with modern hardware?

not impossible, but its dumb.

its a niche market where very few run dual GPUs and very few have reasons to do so. trying to make multi-die GPUs is a thing, but AMD hasnt found a good way to implement it for consumers, and in the server space they dont care if they need to treat each GPU as seperate all that much.