Without more boundaries this is sort of pointless because many downright antique cards could be made to work under 9x with a 9x specific driver as some companies regurgitated out new windows drivers for a moderately long while even with windows built in support.

S3-911 and non-svga cards in svga mode come to mind as being slow

https://www.os2museum.com/wp/more-isa-vga-benchmarks/

I definitely experienced RTG-3105IEH in Win9x. May have used the Super VGA driver as I couldn't get any DirectDraw going on that card

Well, maybe an early Super EGA card that supported pre-VGA high resolutions and color depths (“high” meaning 640x480 and 256 colors)

rmay635703 wrote on 2025-05-12, 04:08:

Without more boundaries this is sort of pointless because many downright antique cards could be made to work under 9x with a 9x specific driver as some companies regurgitated out new windows drivers for a moderately long while even with windows built in support.

S3-911 and non-svga cards in svga mode come to mind as being slow

https://www.os2museum.com/wp/more-isa-vga-benchmarks/

that page does't benchmark in win9x, just vga mode in dos. also the chips450/realtek3105 are quite slow in dos aswell, although not essentially the slowest.

leileilol wrote on 2025-05-12, 04:11:

I definitely experienced RTG-3105IEH in Win9x. May have used the Super VGA driver as I couldn't get any DirectDraw going on that card

can you remember whether you got 256color modes(probably in 640*480 only) in win9x with the 3105, or find the driver that you had been using?
i have only found one version of 3105 driver for win9x on the web, and it just doesn't work, there is an exclamatory mark on it in the device manager.

i finally got the realtek3105's win9x driver running, there are TWO mistakes in the driver's inf file that prevented it from working. the one i found is incorrect isa pnp device id, and the other found by AI is a spelling mistake in copying files.
i am now quite certain that realtek3105 is the slowest video card for win9x, save for cards without win9x driver or with 256kb max that can only run windows with 16colors.

noshutdown wrote on 2025-06-23, 14:43:

the one i found is incorrect isa pnp device id

Care to share what the correct PnP device ID is?

i am now quite certain that realtek3105 is the slowest video card for win9x, save for cards without win9x driver or with 256kb max that can only run windows with 16colors.

Even more so than the Trident 8900S/9000i? Several cards do perform significantly worse at 800x600x256 than 640x480x256 - Trident 9000i, RTG3105iEh (but not RTG3105e), HMC HM86314 and OTI067 are all examples I can think of with Windows 3.1, but I've had no success with getting my HM86314 to work with the one Win9x driver that exists.

noshutdown wrote on 2025-06-23, 14:43:

i finally got the realtek3105's win9x driver running, there are TWO mistakes in the driver's inf file that prevented it from working. the one i found is incorrect isa pnp device id, and the other found by AI is a spelling mistake in copying files.
i am now quite certain that realtek3105 is the slowest video card for win9x, save for cards without win9x driver or with 256kb max that can only run windows with 16colors.

I screwed up the ability of mine to use 512k, so it's stuck at 256... some DOS utils were picking it up as an ET3000 though, so wonder if you want to try an ET3000 driver for giggles to see if Realtek just kinda copied.

noshutdown wrote on 2025-06-23, 14:43:

i finally got the realtek3105's win9x driver running, there are TWO mistakes in the driver's inf file that prevented it from working. the one i found is incorrect isa pnp device id

Wait a minute...
There's an ISA graphics card that implements ISA PNP ?

DEAT wrote on 2025-06-23, 23:27:

Care to share what the correct PnP device ID is?

sure, the correct pnp device id is *pnp0900, the inf file mistakenly wrote *rtg3105 in the [mfg.realtek] section.

Even more so than the Trident 8900S/9000i?

sure, the 8900c is mediocre but far from the slowest, at least it has true 16bit isa bus and 32bit ram width when filled with 1mb. the 9000 may be slower than 8900c in 640*480 and higher resolutions for having only 16bit ram width, but at least it still have 16bit isa bus over those with only 8 bus data pins.

Several cards do perform significantly worse at 800x600x256 than 640x480x256 - Trident 9000i, RTG3105iEh (but not RTG3105e), HMC HM86314 and OTI067 are all examples I can think of with Windows 3.1, but I've had no success with getting my HM86314 to work with the one Win9x driver that exists.

this is reasonable as these cards all have 8 bit isa bus and 16bit ram width at no more than 40mhz, which becomes scarce in 800*600*256c mode as the video chip needs to send more data to the ramdac and used up ram bandwidth. the 3105e may be better as it comes with a faster 48mhz crystal.
i have tested two 3105 cards so far, one is a regular 3105i with 70ns dram, and the other is a strange 3105ieh with 16pieces of outrageously slow 120ns 1*256kbit dram, so slow that screen keeps messing up in 640*480*256c mode. performance wise, the 3105ieh is faster(although unstable) than 3105i in 640*480*256c mode, but slower in 800*600*256c mode.
the hmc86304/314 looks similar to the realtek3105i/ieh, and they both belong to the same class of 8bit isa and 16bit ram. i didn't try it simply due to short of drivers, and i doubt if it can get much slower than the 3105.

noshutdown wrote on 2025-06-24, 05:29:

sure, the correct pnp device id is *pnp0900, the inf file mistakenly wrote *rtg3105 in the [mfg.realtek] section.

I've tested this, and it doesn't work with my RTG3105E or RTG3106. I've noticed other minor discrepancies in REALTEK.INF compared to known working third-party drivers (ATi mach64, C&T F65545, S3, Trident, Tseng W32) which I cleared up and still was not able to get them to work. My RTG3105iEh has a dead RAM chip from testing it on an overclocked ISA bus and I haven't got around to replacing it yet, so I can't test that card. What was the reasoning behind choosing *PNP0900 specifically? Which specific RTG3105 chipset are you working with? Can you provide a version of this driver that supposedly works?

sure, the 8900c is mediocre but far from the slowest, at least it has true 16bit isa bus and 32bit ram width when filled with 1mb. the 9000 may be slower than 8900c in 640*480 and higher resolutions for having only 16bit ram width, but at least it still have 16bit isa bus over those with only 8 bus data pins.

I agree with this. Except I wasn't talking about the 8900C - I was specifically talking about the 8900S, which was released between the 8800CS and 8900B.

UniVBE detects this as a 8900B, but the performance is exactly the same as a 9000i-3, which are the worst overall cards that I have - slower than a 8900B with 512KB of RAM.

this is reasonable as these cards all have 8 bit isa bus

False, at least when it comes to the 9000i-3, RTG3105E and OTI067. My 9000i-3 card doesn't have jumpers to set it to 8-bit ISA mode like most 8900CL/D/9000 cards, and it refuses to work in an 8-bit ISA slot.

the hmc86304/314 looks similar to the realtek3105i/ieh, and they both belong to the same class of 8bit isa and 16bit ram. i didn't try it simply due to short of drivers, and i doubt if it can get much slower than the 3105.

My 286 Windows 3.1 benchmarks shows that its much slower than a RTG3105E, but it is comparable to a RTG3105iEh. But as I've stated multiple times in the past that's only relevant for GDI, and not for DirectDraw.

BitWrangler wrote on 2025-06-24, 02:28:

some DOS utils were picking it up as an ET3000 though, so wonder if you want to try an ET3000 driver for giggles to see if Realtek just kinda copied.

The real kicker is that the DOS utilities for the Avance Logic ALG2101 are visually identical to Realtek's DOS utilities. Both of the Realtek Win9x drivers that exist have a copyright belonging to Avance Logic. I've never believed that Avance Logic were ever their own company at any point, and were only a subsidiary to Realtek.

Grzyb wrote on 2025-06-24, 05:09:

Wait a minute...
There's an ISA graphics card that implements ISA PNP ?

Nope - most of the ISA/VLB video card drivers for Win9x refer to those devices in their INF files as *PNP09xx, with the last two digits being unique per company. The only exception from that naming convention that I've seen for working drivers are S3 drivers which use S3911, S3801, S3928, S3864 etc.

DEAT wrote on 2025-06-24, 14:00:

I've tested this, and it doesn't work with my RTG3105E or RTG3106. I've noticed other minor discrepancies in REALTEK.INF compared to known working third-party drivers (ATi mach64, C&T F65545, S3, Trident, Tseng W32) which I cleared up and still was not able to get them to work. My RTG3105iEh has a dead RAM chip from testing it on an overclocked ISA bus and I haven't got around to replacing it yet, so I can't test that card. What was the reasoning behind choosing *PNP0900 specifically? Which specific RTG3105 chipset are you working with? Can you provide a version of this driver that supposedly works?

hardware id is only the first change that i made. the second mistake is found by AI that you need to search and change "[TRG_CopyFiles]" to "[RTG_CopyFiles]", as this spelling mistake prevents driver files(vxd and drv) from being copied. after this it worked for me, my card is a rtg3105i whose pcb looks more similar to generic xfx pt505s low pcb cards in this page:
https://vgamuseum.info/index.php/component/k2 … ltek-rtg3105ieh
with a few minor differences:
*resistor r5 empty
*both c16 and c17 populated
*14.3khz crystal in place of fccid sticker
and if you have a 3106, would you try swapping in a hicolor ramdac and see if it has any support? vgadoc listed some hicolor modes for realtek cards.

I agree with this. Except I wasn't talking about the 8900C - I was specifically talking about the 8900S, which was released between the 8800CS and 8900B.
UniVBE detects this as a 8900B, but the performance is exactly the same as a 9000i-3, which are the worst overall cards that I have - slower than a 8900B with 512KB of RAM.

i really didn't know about the existence of 8900s before, but from the photo we can see it has 16bit ram width and seemingly true 16bit isa bus just like the 9000, so its reasonable to perform similarly to the 9000. still, they have the advantage of 16bit isa bus over those with 8 isa data pins.

False, at least when it comes to the 9000i-3, RTG3105E and OTI067. My 9000i-3 card doesn't have jumpers to set it to 8-bit ISA mode like most 8900CL/D/9000 cards, and it refuses to work in an 8-bit ISA slot.

my mistake here, i know that the trident9000 and oak067 has true 16bit isa bus. what i meant is that 16bit video ram(at no higher than 40mhz) would become a noticeable bottleneck at higher resolutions.
still i hesitate to believe that the realtek3105e has any true 16bit isa bus, traces from isa high 8bit data fingers seem to disappear under the 72ls00 chip and i can't find where they lead to. its better performance probably owes to higher ram clock(48mhz crystal vs 40mhz of 3105).

the driver worked with both my 3105i and 3105ieh, with following results:
*3105ieh is about 10-15% faster than 3105i in most tests(dos or win9x), in resolutions up to 640*480*8bit.
*at 800*600*8bit mode its different, 3105i runs at 56hz, while 3105ieh runs at 43i hz that my lcd can't handle. i managed to get a x-vesa result though, and 3105ieh is much slower than 3105i in this mode. not sure if its related to refresh rate.
*the 3105i is a generic lay out, while the 3105ieh is not, which has 16pieces of outrageously slow 120ns 1*256kbit dram chips, so slow that screen keeps messing up at 640*480*8bit mode. no idea on actual ram clock though.
*there is a "interlaced/noninterlaced" jumper on both cards. on 3105i i can't see any difference on performance or screen refresh rate, while 3105ieh only boots with this jumper closed(noninterlaced), .