It is absolutely possible to recreate a chip like the VSA-100. There are companies that will analyse and reverse engineer the chip for you if you supply them with a couple of chips and pay their fees.

However I don't think demand will justify the expense:
-There are pretty good Glide wrappers that let you play Glide games on non-3dfx graphics cards.
-There are still plenty of NOS VSA-100 chips left. You can currently just buy some on ebay.
-Anthony has recently come up with his own design of a Voodoo 6000 PCI card using said NOS VSA-100 chips. With the expected demand he estimates the cost for an assembled board to be around USD1500. Prices will be similar for other reproductions. If you have to make new chips or reverse engineer and implement in FPGA it'll be much worse.
-The VSA-100 is less compatible than earlier 3dfx chipsets. It is absolute overkill for earlier Glide games while most (if not all) later games supporting Glide will also look and play fine on other APIs and newer cards are plenty and often much faster.
-No 3dfx card is a good choice for any other API besides Glide because even VSA-100 cards are mediocre performers by early 2000's standards and painfully slow compared to later cards with Win9x drivers.

Again I'm not saying this is impossible but right now this is just not economically feasible. Projects like the ones you listed are much simpler designs and they use avaliable and well documented chips. Anthony's V6000 PCI design using NOS chips ended up costing so much it's only attractive for hardcore enthusiasts.
If you end up having to make your own VSA-100 chips (or substitute them in FPGA) you also get problems with IP rights so at this point it may be cheaper to just ask Nvidia for licensing to run a few batches of VSA-100 chips at your favorite chip manufacturer instead of reverse engineering. I guess if there was enough demand to justify that Nvidia would already be making those cards since they already have everything they'd need.

Doornkaat wrote on 2021-02-22, 15:44:

-Anthony has recently come up with his own design of a Voodoo 6000 PCI card using said NOS VSA-100 chips. With the expected amount he estimates the cost for an assembled board to be around USD1500. Prices will be similar for other reproductions. If you have to make new chips or reverse engineer and implement in FPGA it'll be much worse.

I have never heard of that until now. Is there a place where I can get the blueprint for this board design? I would love to do this.

https://www.modlabs.net/forum/topic/60674/
smaller thread on Vogons: Voodoo 5 6000 Clone PCB
Anthony is not sharing his scematics though and with the work that has gone into this I think I can understand that.
He has also done stuff like an SLI Voodoo 1 btw.

TelamonLivesOn wrote on 2021-02-22, 14:37:

Note: I am not in any way implying that I will start work on this right now, but I just wanted to throw the thought out there.

🤣

All those examples you've listed are either extremely simple (Covox) or use existing chips. As Doornkaat stated, someone has already used the same approach to build new 3dfx cards.
Low-level 3dfx emulation is also already a thing (pcem), in theory someone could take that code and port it over to e.g. MiSTer or another FPGA-based solution.
As for why nobody's done that, or is selling kits to build new 3dfx cards, I guess it's a lack of interest. 3dfx cards might be expensive, but they're not that great/useful.

Orpheus/AWE64 Legacy fill a niche by creating a product that doesn't really exist to begin with, namely high-quality ISA sound cards with broad game support. In that case it's worth it to build something that's more expensive than an old piece of hardware.

all the listed examples also use original chips and no reproductions, i.e. OPL2/OPL3 or the AMD InterWave none of these use FPGAs, and while the OPL3 at least has been done in FPGA the "original" sound of the OPL3 is what is thought after in replica soundcards and these chips are still available in abundance.

also it would be far more difficult to implement a big chip like the VSA-100, it would probably end up more expensive than hunting down an original card.

also it would be far more difficult to implement a big chip like the VSA-100

I doubt you can recreate it on original 250m lithography. And it's fairly simple chip for modern not top-tier lithography. But yes, nobody gives a damn about VSA100, what everyone really want is reliable Glide emulation, preferably platform agnostic.

isn't the original voodoo or the voodoo 2 a far more realistic target? also probably more useful?

and it would open the door for changes like higher clocks that would be impossible with OG hardware, and that could be fairly interesting.

still, this is far beyond my level, and sounds really complex.

SPBHM wrote on 2021-02-22, 16:23:

isn't the original voodoo or the voodoo 2 a far more realistic target? also probably more useful?

and it would open the door for changes like higher clocks that would be impossible with OG hardware, and that could be fairly interesting.

still, this is far beyond my level, and sounds really complex.

I really just want a fast 3DFX card with support for higher resolutions, which an enhanced voodoo 2 might be able to do. I also would like the voodoo 2 to support more than 2 cards in sli.

Edit: I'm thinking something like a Quantum3D Mercury Brick

TelamonLivesOn wrote on 2021-02-22, 16:26:

I really just want a fast 3DFX card with support for higher resolutions, which an enhanced voodoo 2 might be able to do. I also would like the voodoo 2 to support more than 2 cards in sli.

I'm really not trying to be a dick or nay sayer here but what would that card really be good for?
Most Win9x games are designed to be run at 640*480, 800*600 or 1024*786. If you choose higher resolutions textures often look bad and UI elements will be too small. DOS is even more limited.
A decent scaler makes much more sense for Win9x gaming on high resolution screens in my opinion.

Also it sounds like you just want a Glide compatible card, not necessarily a 3dfx one. So why not see what a Wrapper can do for you? nGlide is a great place to start. Your satisfying Win9x gaming experience doesn't have to be expensive! 😎👍

Edit: Win9x isn't listed as supported by nGlide but it works anyway.

Doornkaat wrote on 2021-02-22, 17:28:

I'm really not trying to be a dick or nay sayer here but what would that card really be good for? Most Win9x games are designed […]
Show full quote

TelamonLivesOn wrote on 2021-02-22, 16:26:

I really just want a fast 3DFX card with support for higher resolutions, which an enhanced voodoo 2 might be able to do. I also would like the voodoo 2 to support more than 2 cards in sli.

I'm really not trying to be a dick or nay sayer here but what would that card really be good for?
Most Win9x games are designed to be run at 640*480, 800*600 or 1024*786. If you choose higher resolutions textures often look bad and UI elements will be too small. DOS is even more limited.
A decent scaler makes much more sense for Win9x gaming on high resolution screens in my opinion.

Also it sounds like you just want a Glide compatible card, not necessarily a 3dfx one. So why not see what a Wrapper can do for you? nGlide is a great place to start. Your satisfying Win9x gaming experience doesn't have to be expensive! 😎👍

Edit: Win9x isn't listed as supported by nGlide but it works anyway.

Really what I'm trying to do is write custom software for the hardware that I can use in my vintage high-end workstation. I am also trying to stretch the limits of 3d rendering in dos. I also want a reason to change my 2D video card, since my Matrox Millennium II has issues at 800x600 and is generally incompatible with several DOS games. I also like the coolness factor of a custom multi-sli card. As for having glide support, I could care less if early glide games don't work, since I also have a Voodoo 1, but I do want a beast of a system with 3DFX as its main claim to fame 😎. For more background, this system runs Windows NT 4.0 on a Full AT Dual Pentium III 600mhz, 512mb to 2gb SDRAM, 600w AT PSU, around 11 usable slots, a 47gb 5.25" 68-pin SCSI HDD, and much much more.

Best thing is not recreate the VSA-100 with bugs, make a revision 2 of this that will work properly with PCI 2.1, AGP 2x or 4x and another for PCIe and revise the chip to work with 64MB 128bits. 4x the texture and render processor as well then give option of 4 chiplets on one silicon die to allow for less number of packages to 4.

Do same based on rampage chipset bring it to realization with these suggestions.

Cheers,

pentiumspeed wrote on 2021-02-22, 19:42:

Best thing is not recreate the VSA-100 with bugs, make a revision 2 of this that will work properly with PCI 2.1, AGP 2x or 4x and another for PCIe and revise the chip to work with 64MB 128bits. 4x the texture and render processor as well then give option of 4 chiplets on one silicon die to allow for less number of packages to 4.

Do same based on rampage chipset bring it to realization with these suggestions.

Cheers,

I like this idea. What type of fpga would be appropriate for this? I can't seem to find any real information on the individual chips, only on the boards (getting a single chip would be better). The Voodoo 5500 ran at 166mhz and could often be overclocked to 200mhz, so this would probably be the most accurate speed.

The specs to "3dfx Napalm", which was the codename for the VSA-100, were leaked around the same time as (or even included in?) the big 3dfx leak of 2003.
You might want to take a look here (for example): http://vgamuseum.info/index.php/companies/ite … x-voodoo-5-5500

Stiletto wrote on 2021-02-23, 03:19:

The specs to "3dfx Napalm", which was the codename for the VSA-100, were leaked around the same time as (or even included in?) the big 3dfx leak of 2003.
You might want to take a look here (for example): http://vgamuseum.info/index.php/companies/ite … x-voodoo-5-5500

Can't believe I didn't see that spec sheet near the bottom. Thank you very much for pointing that out! I might just consider buying some VSA-100 chips now and then build a card later. One other question though, what kind of ram was used with the Voodoo 5500? I know it is 166mhz SDRAM, but I need to know the other specs so that I can actually source some RAM chips.

considering the fact that even such operation as xl9572 cpld cracking costs a lot (it requires precise work on a special equipment), i can't even imagine how much complete reversing of vsa100 silicon would cost. that's only the first step. asic silicon production costs a really big money.

TelamonLivesOn wrote on 2021-02-23, 03:36:

what kind of ram was used with the Voodoo 5500? I know it is 166mhz SDRAM, but I need to know the other specs so that I can actually source some RAM chips.

Hyundai HY57V65322DB are the most common chips on Voodoo 5500 cards I think.

You could also try locating the later revision 320 of the VSA-100 that supposedly runs cooler and with higher clocks btw.

All possible memory configs are listed in a vsa100 datasheet

Quadcore VSA-100 hehe

anthony wrote on 2021-02-23, 08:23:

All possible memory configs are listed in a vsa100 datasheet

Thanks! I will look there. Also, awesome job on the Voodoo 6000 board replica!