CoffeeOne wrote on 2024-01-05, 22:22:

I quickly checked Dosdays. S3 864 chip based PCI(!) cards might have been available already since 1992? I cannot believe it hone […]
Show full quote

Scythifuge wrote on 2024-01-05, 21:53:

CoffeeOne wrote on 2024-01-05, 21:36:

One more remark: I checked again the picture of the Diamond Viper PCI. The Weitek chips have a date code from 1993 week 44 (?) S […]
Show full quote

One more remark: I checked again the picture of the Diamond Viper PCI. The Weitek chips have a date code from 1993 week 44 (?)
So it is maybe one of oldest PCI cards existing. So from a collector's point of view, you should not sell it 😁

EDIT: Now I am curious: What was the first PCI graphics card?
S3 928 chip based cards might have been the first? Or Mach32 PCI?
Anyone knows?

I have been checking the Dos Days website a lot when trying to find cards by specific year. It is easy to get lost while researching all of those cards! Also, going by the prices of a lot of these cards these days, an S3 of some sort is probably the only answer, unless I stick with my overkill Mach64 8mb, hehehe.. I may throw the Mach32 in there and run some game tests.

I quickly checked Dosdays.
S3 864 chip based PCI(!) cards might have been available already since 1992? I cannot believe it honestly spoken 😁
EDIT: The ones I find on Ebay are all from 1994. So S3 928 or your upcoming Viper win the contest for being older.

Since we upgraded from the SX to the DX2 in 1994, I'll consider 1994 cards. I think it is cool after all that I nabbed that Viper card!

I have been tinkering with a pretty nice OTI-087 card (exactly like the WD Paradise 1580 pictured at this link, except with an OAK BIOS) and everything seems to work really well, except in Windows 98SE (I know, not the right OS for this card...) the built in driver only allows 640x480, 800x600 and 1024x768 at 16 color or 256 color modes. There aren't any high color modes available at all, even though Windows (dxdiag) and NSSI report that it is a 1MB card. I can't verify the memory capacity since there are no datasheets for the memory chips, but they are marked GV 4512G4B, which makes me think 4x512kbit = 256KB, and there are four of them, so 1MB makes sense.

Is there a proper Windows 9x compatible driver out there for this card so that I can give it a good thorough test? Windows is just installing what it calls "OAK Technology Super VGA", which is probably universal for any of OAK cards, even though the 087 is much more capable than the earlier models.

Even at this archived official driver page it says this:

"for OTI-067, 077 and 087

Drivers for these chipsets are included in the Windows 95 package as Oak Super VGA."

So, maybe it is necessary to use the card in Windows 3.1 to get the driver that enables full resolution support?

Ozzuneoj wrote on 2025-04-02, 20:34:

So, maybe it is necessary to use the card in Windows 3.1 to get the driver that enables full resolution support?

Pretty much what you need to do, at a cost of losing DirectDraw support. I haven't found any other Windows 9x drivers.

(exactly like the WD Paradise 1580 pictured at this link, except with an OAK BIOS)

That card has a 16-bit memory bus width, and you will see worse performance with everything above 800x600x256 compared to a card with 32-bit bus width (that card with the GALAXY sticker on the BIOS chip on VGA Museum is an example of a card with 32-bit bus width) - the difference isn't too bad at 800x600x256, but it becomes a lot more noticable at 1024x768x256.

DEAT wrote on 2025-04-02, 22:22:

(exactly like the WD Paradise 1580 pictured at this link, except with an OAK BIOS)

That card has a 16-bit memory bus width, and you will see worse performance with everything above 800x600x256 compared to a card with 32-bit bus width (that card with the GALAXY sticker on the BIOS chip on VGA Museum is an example of a card with 32-bit bus width) - the difference isn't too bad at 800x600x256, but it becomes a lot more noticable at 1024x768x256.

Ah! Yeah, I did notice that! 800x600 was sluggish but usable on the Windows desktop, but 1024x768 was really bad. This is probably why the Windows 9x drivers don't enable higher than 256 color modes on these cards.

How can I tell which ones have a 16bit bus and which are 32bit? Does this apply to cards other than the OAK *7 series?

I also messed with the jumpers out of curiosity.

I noticed that 1024x768 originally had a distinct interlaced look to it when things were moving and performance actually seemed usable. After changing the interlacing jumper to "non-i", it seems that it is no longer interlaced at 1024x768 and it definitely is a lot slower. Strangely, when switched the jumper back to interlaced it didn't seem to actually change anything? Very odd.

I don't really understand what the VESA \ VGA mode jumper does. I believe it was set to VGA when I got it. I don't honestly notice a difference when I change this.

Also, since I had all my soldering stuff out and I noticed it had an unpopulated "0 WS \ 1 WS" jumper location, which indicated that it'd be set to 1 WS without a jumper, I decided to just bridge that jumper to set it to 0 WS... but that also seemed to make no difference in performance in the areas where it needed it... though the 16bit memory bus limitation you mentioned is likely the main bottleneck. I, admittedly, have very little experience dealing with wait-states and I have no idea what this would do. I was just having fun tinkering with an ancient card on a stupidly overpowered PC (850Mhz PIII on a 440BX).

Ozzuneoj wrote on 2025-04-03, 07:00:

How can I tell which ones have a 16bit bus and which are 32bit? Does this apply to cards other than the OAK *7 series?

From what I've seen, this is uniquely an Oak card problem and I have not seen any other 1MB cards that exhibit this poor design choice.

Using your vgamuseum link as a reference point, the easiest way to identify OTI087/087X cards that have a 16-bit memory bus width is whether they have soldered SOJ-20 RAM that has "4512" labelled anywhere, such as the WD Paradise 1580 - that means that they're 512K x 4bit chips. Four of those 512K x 4bit chips will add up to 1MB of RAM.

The far more common setup of eight 256K x 4bit chips (or two 256K x 16bit) that most 1MB cards use should add up to a 32-bit memory bus width if the chipset is capable of supporting it, along with the uncommon setup of four 256K x 8bit chips which the "Oak OTI-087 512KB" (the 512KB claim is false and is verbatim repeated from a specific US eBay seller who has multiples of this card, which is sufficiently good evidence that vgamuseum don't reliably check what they collect - I know this because I have that exact card) card in that vgamuseum link uses, identified by the "8256" label on the RAM chips. I have a 1MB GD5420 card that also uses the same four 256K x 8bit setup.

As for the performance difference, the only data I have readily available are some old Windows 3.1 benchmark results I did on a 24Mhz 286 mobo with a HT18C chipset and a 12Mhz ISA bus, using the latest OTI087 and OTI087X drivers respectively. I decided to filter out the WindSock detailed GDI results that didn't have any influence on the quick GDI results, more as a better understanding of what the quick GDI results actually mean. Take note that I don't have a 16-bit memory bus width OTI087 card:

The performance for the 32-bit memory bus width OTI087X card is odd compared to the OTI087 card - I have a suspicion that it has something to do with its truly bizarre memory layout:

Props if you can figure out what's weird about this.

As for the 067/077, the 067 only supports 512KB of VRAM so it'll never have a 32-bit memory bus width, and I have firm reason to believe the 077 doesn't support it either even with 1MB of VRAM - my 077 with eight 256K x 4bit chips drops off a cliff extremely hard when doing 800x600x256 or higher, regardless of whether it's Windows 3.1 or VESA modes. I haven't tried my 077 with four 512K x 4bit chips but I'm expecting something similar.

I don't really understand what the VESA \ VGA mode jumper does. I believe it was set to VGA when I got it. I don't honestly notice a difference when I change this.

It should change the refresh rate of 640x480 and 800x600 resolutions, theoretically speaking.

Also, since I had all my soldering stuff out and I noticed it had an unpopulated "0 WS \ 1 WS" jumper location, which indicated that it'd be set to 1 WS without a jumper, I decided to just bridge that jumper to set it to 0 WS... but that also seemed to make no difference in performance in the areas where it needed it... though the 16bit memory bus limitation you mentioned is likely the main bottleneck.

I've noticed no performance difference with my 32-bit OTI087 card when changing that jumper.

DEAT wrote on 2025-04-06, 04:25:

From what I've seen, this is uniquely an Oak card problem and I have not seen any other 1MB cards that exhibit this poor design […]
Show full quote

Ozzuneoj wrote on 2025-04-03, 07:00:

How can I tell which ones have a 16bit bus and which are 32bit? Does this apply to cards other than the OAK *7 series?

From what I've seen, this is uniquely an Oak card problem and I have not seen any other 1MB cards that exhibit this poor design choice.

Using your vgamuseum link as a reference point, the easiest way to identify OTI087/087X cards that have a 16-bit memory bus width is whether they have soldered SOJ-20 RAM that has "4512" labelled anywhere, such as the WD Paradise 1580 - that means that they're 512K x 4bit chips. Four of those 512K x 4bit chips will add up to 1MB of RAM.

The far more common setup of eight 256K x 4bit chips (or two 256K x 16bit) that most 1MB cards use should add up to a 32-bit memory bus width if the chipset is capable of supporting it, along with the uncommon setup of four 256K x 8bit chips which the "Oak OTI-087 512KB" (the 512KB claim is false and is verbatim repeated from a specific US eBay seller who has multiples of this card, which is sufficiently good evidence that vgamuseum don't reliably check what they collect - I know this because I have that exact card) card in that vgamuseum link uses, identified by the "8256" label on the RAM chips. I have a 1MB GD5420 card that also uses the same four 256K x 8bit setup.

As for the performance difference, the only data I have readily available are some old Windows 3.1 benchmark results I did on a 24Mhz 286 mobo with a HT18C chipset and a 12Mhz ISA bus, using the latest OTI087 and OTI087X drivers respectively. I decided to filter out the WindSock detailed GDI results that didn't have any influence on the quick GDI results, more as a better understanding of what the quick GDI results actually mean. Take note that I don't have a 16-bit memory bus width OTI087 card:

The attachment oti087-640x480.png is no longer available

The attachment oti087-800x600.png is no longer available

The attachment oti087-1024x768.png is no longer available

The performance for the 32-bit memory bus width OTI087X card is odd compared to the OTI087 card - I have a suspicion that it has something to do with its truly bizarre memory layout:

The attachment 2025-04-06-14-18-21-344a.jpg is no longer available

Props if you can figure out what's weird about this.

As for the 067/077, the 067 only supports 512KB of VRAM so it'll never have a 32-bit memory bus width, and I have firm reason to believe the 077 doesn't support it either even with 1MB of VRAM - my 077 with eight 256K x 4bit chips drops off a cliff extremely hard when doing 800x600x256 or higher, regardless of whether it's Windows 3.1 or VESA modes. I haven't tried my 077 with four 512K x 4bit chips but I'm expecting something similar.

I don't really understand what the VESA \ VGA mode jumper does. I believe it was set to VGA when I got it. I don't honestly notice a difference when I change this.

It should change the refresh rate of 640x480 and 800x600 resolutions, theoretically speaking.

Also, since I had all my soldering stuff out and I noticed it had an unpopulated "0 WS \ 1 WS" jumper location, which indicated that it'd be set to 1 WS without a jumper, I decided to just bridge that jumper to set it to 0 WS... but that also seemed to make no difference in performance in the areas where it needed it... though the 16bit memory bus limitation you mentioned is likely the main bottleneck.

I've noticed no performance difference with my 32-bit OTI087 card when changing that jumper.

Thank you for all the helpful info! It somehow had not registered in my mind that the "bits" of each memory chip can actually directly add up to the bus width. I think I learned this in regard to another (much later) video card a few years ago and it has completely slipped my mind. I feel like there are relatively few cards that have variants with different memory bus widths, at least until the Geforce 2\4 MX era when a card could be 32bit DDR, 64bit SDR or 128bit SDR... blehg.

I'm not familiar with most of the benchmarks you ran (I assume that in all of them a higher number is better), so correct me if I'm interpreting the data wrong, but it is interesting how the 16bit card doesn't perform much differently until the resolution is raised. Normally, a 16bit memory bus seems like it'd be a horrendous bottleneck, but at such low resolutions at 256 color it doesn't make much of a difference... until 1024x768 of course, as you said.

As for the funky memory config... I can't find datasheets to support this, but the numbering on two of those chips seems to imply that they are 256k x 3bit??? And they are mixed in with 256k x 4bit chips in a diagonal orientation?? What even is that... 😅