Grzyb wrote on 2023-10-27, 18:58:
That's dated Sept 21, 1992, and it doesn't contain VGA Wonder drivers, only the ULTRA drivers.
But indeed, they are identical (with the exception of corrupt files) to the ULTRA part of the Sept 2, 1992 release.
Woops, my bad - I had linked to the wrong folder from the MPC Wizard 2.0 CD - the correct folder is WIZSETUP\ATI_W31 but it's pre-extracted on the CD. Either way, the drivers I'm referring to are in the ZIP files I attached earlier.
Anyway, good work with all the tests and benchmarks!
Waiting for more... […]
Show full quote
Anyway, good work with all the tests and benchmarks!
Waiting for more...
As of yet, it looks like there are only two good accelerators for 286 machines with ISA bus:
- S3 911
- ATI mach8
Thank you! I've received the following on Friday just in time for some weekend testing/tinkering:
- 1MB Oak OTI087
- 1MB S3 801 (STB PowerGraph X-24)
- 1MB S3 924 (2theMax GUI6000)
- 2MB S3 928 (Diamond Stealth PRO, BIOS 1.10)
- Weitek 5286
The S3 928 didn't even POST on a 286, so that's immediately ruled out. Have a feeling that the Diamond BIOS has 386 instructions, but I can't say that I want to drop more money on another ISA S3 928 that's not a Diamond card any time soon to confirm that theory. Works on my 386DX-40, trying S3 911/924 drivers resulted in a failure so as far as I can tell the chipset has no chance of ever working on a 286, so it seems like a dead-end anyway.
The eBay lister mentioned the OTI087 was an OTI087X with only 512KB - I should have paid closer attention to the chipset (I noticed that the RAM was actually 1MB), as a documented difference between the OTI087 and OTI087X is that the latter has a hardware mouse cursor. My interest in the OTI087(X) was due to a recommendation from viti95, and it absolutely kills in DOS. Pretty sure the bad reputation comes from being on the Diamond Viper cards, as it easily saturates the ISA bus. Good framebuffer performance, better than the Tsengs and Tridents and unlike the OTI-077 keeps a persistent performance across all 256-colour modes. Also has the only functional 24-bit colour drivers that I've come across, but the RAMDAC on the card is only capable of 16-bit. It should be noted that the OTI087 and OTI087X have separate Windows 3.11 drivers, and you cannot use the OTI087X drivers with the OTI087.
The S3 924 went exactly as expected - a noticeable performance boost compared to the S3 911, 0-10% depending on the specific feature. I'm guessing parts where the performance was the same was because of CPU bottleneck, will need to confirm on a 486. At 800x600x256/1024x768x256/640x480x16b it is easily the best card for Win 3.11 on a 286. Sadly, unlike my S3 911 (Orchid Fahrenheit 1280) it will not change from 56hz on 800x600x256, while 1024x768x256 and the 640x480 15/16bit modes are interlaced. Swapping the BIOS and RAMDAC from my Orchid Fahrenheit (which has DOS utilities available for changing refresh rates) did not result in any change - the Orchid Fahrenheit does have dip switches on the I/O panel that the S3 924 card lacks.
The next two cards are where it gets more interesting:
The Weitek 5286 caught my attention because of its VGADOC listing - the card I received is an evaluation board (several are available from the same eBay seller) that has sockets for 1MB of ZIP-28 VRAM (empty), 512KB of ZIP-40 DRAM (included) and unpopulated DIP-20 DRAM solder pads for 1MB. Three drivers are known to exist, two are available on Weitek's site and the third are for a W5186 card named WinMax. The only reference to a consumer W5286 card that I'm aware of is the ASL Lightning SVGA for which no photos exist and certainly no drivers have been uploaded that I can see. None of the DOS utilities for any version of the drivers will allow modeset changes - they expect a W5086/W5186. 16-colour and 15-bit colour modes work, though performance of the 15-bit modes are in line with most other cards that have 15/16b support. 256-colour modes is where it gets truly bizarre - it has very glitchy output, this also happens on a 386 as well. I soldered DIP-20 sockets and successfully confirmed that 1MB was working fine, but even after the upgrade from 512KB DRAM to 1MB DRAM it still has glitchy output on 256-colour mode. What gives?
Turns out the earlier 1.10 drivers from Weitek's site has some very detailed documentation (RN110P.TXT) that states that the 800x600x256 and 1024x768x256 modes need VRAM, while 640x480x256 can use either DRAM or VRAM - curiously, it also states that the 15-bit modes also need VRAM. I have some ZIP-28 80NS VRAM coming from the US, so I'll know when that arrives if it really depends on VRAM for having functional 256-colour modes, or if the drivers are actually expecting a W5086/W5186 for whatever it is trying to do.
The S3 801... whew. I initially started with the 1.30 drivers (archived off the floppies of another S3 801 card that had died quickly after receiving it, which also had really weird modifications done to it with regards to the feature connector and even more bizarrely displayed VGA text mode as a 1024x768 framebuffer) but couldn't get past the startup logo. 1.31 (from MPC Wizard 3.0 CD, pre-extracted) and 1.33 drivers (incomplete copy found from Actix GraphicsENGINE drivers that are missing 15/16-bit modes also from MPC Wizard 3.0 CD, pre-extracted) also failed to get past the startup logo. I tried the STB PowerGraph X-24 drivers dated December 1992 that are found on the MPC Wizard 2.0 CD, and it was able to get to the desktop! Except it only drew the program manager window (and open folders) itself, while leaving everything inside it blank and eventually crashing back to DOS.
Similar with the PCI mach32/GD5434 tests, I decided to take a stab at using the known working S3 911/924 drivers for any hope. v1.6 and v1.7 both failed to get past the startup logo, but v1.4 and v1.5 worked! I have a full set of 800x600x256 benchmark results with the v1.5 drivers - relative performance compared to the 924 shows even further improvement in line drawing, but the v1.4 and v1.5 drivers have bad polygon fill scores (same with 911/924) compared to later drivers. I decided to dig around further and tried the STB WIND/X drivers, which finally unlocked the best potential of the 801 on the 286 - I only took a Winbench 3.11 screenshot at 800x600x256. Did some further poking around with drivers, and suddenly I get nothing but massive glitches and constant crashes. Deleting and reinstalling Windows twice has resulted in no improvement. Testing all S3 911/924 drivers on a 386 in Standard Mode (including v1.71 and v1.74, which are not 286-compatible) shows perfect stability. I am truly baffled as to why it's happening. I'm guessing there's a specific order of things that I need to do, but reproducing it has failed to get a working desktop.
As for how vendor-specific the STB WIND/X drivers are - testing the drivers with my S3 911 and 924 cards showed transparency issues with icons, but was otherwise fine. The STD WIND/X drivers also include an incomplete copy of generic v1.51 drivers - all 16-colour modes except 1280x960 and 640x480x16-bit are missing, but everything else is intact though the OEMSETUP.INF file never references them. It looks like the 640x480x15-bit driver is the same from the v1.5 drivers. I recreated the OEMSETUP.INF using the v1.5 file as a template.
Still, I have more data that I need to throw in the spreadsheet, along with filling in data from other cards, I still need get time to do basic testing of other drivers. I'm going to refrain from including partial S3 801 results for now. I also have an ATI VGA Wonder 1024D and a Cirrus Logic GD5402 arriving at some point - they're more of a curiousity for me (especially as the ATI Graphics Ultra has really bad 800x600x256 performance with the VGA Wonder drivers) and were cheap enough.
- Avance Logic ALG2101 - is this even an accelerator? Can't find the datasheet... anyway, not faster then good framebuffers
Absolutely positive on this one, it is one of four cards that gets an overall WindSock score of triple digits. A closer inspection of the Winbench results will show where its strengths and weaknesses are compared to simply looking at the Graphics WINMARK score - it has strong BitBLT and line drawing performance but it does not do font caching at all and polygon fills aren't the greatest, which I'm certain is what is weighing the overall score down. Curiously, it's the only accelerator that loses performance when using higher resolutions with 256 colours.
- Cirrus Logic GD5429 - must use 5422/5424 drivers, so no acceleration
At 1024x768x256 it performs 15% better overall compared to the GD5422, it has the best 800x600x16b results and the third-best 640x480x16b results far behind the S3 911/924. It has font caching (not as good as S3 and ATI) and BitBLT is definitely functioning to some extent.
- Tseng Labs ET4000/W32i - surprisingly slow indeed
My understanding is that having 2MB of RAM will improve performance - I have two ET4000/W32i cards, one with 1MB expandable to 2MB and a 16-bit RAMDAC, and another at fixed 1MB with a 6-bit RAMDAC. The expandable card arrived in a rough shape with some of the DRAM sockets being physically damaged and faded solder joints - I replaced the DRAM sockets that were in a bad physical shape and was able to successfully detect 2MB, but I was getting visible glitches. Might have something to do with adding 70ns chips to existing 45ns chips, might need to replace the other DRAM sockets. Unfortunately when doing a full retest with the February 1995 drivers after doing a test with the December 1994 drivers, that card had started to suddenly show heavy visual corruption. Hopefully I can repair it.
- Western Digital WD90C31 - about the same speed as good framebuffers
This one is really odd - it has some insane scores (MS PATCOPY and MS PATINVERT) that surpass the S3 911/924 and mach8 by a significant margin, has good BitBLT performance, has good horizontal/vertical line drawing but crap diagonal drawing, but no font caching at all and polygon scores are also bad. Both this and the ALG2101 are why I'm posting full benchmark results rather than the summary - summaries never tell a proper tale of where strengths and weaknesses lie.
Meanwhile, I've tried ATI Graphics Ultra in Real mode of Windows 3.0, on a 386+, using the 8514/A driver - and it worked.
Now, considering the card can be configured for 8-bit bus I/O, there's a chance for it to run even with an 8088...
Note to self to try it some day... PC/XT running Windows at 1024 x 768 x 256 with acceleration would be magical...
This is good to know - I haven't really paid much attention to Win 3.0 drivers, but I have noticed that both ATI and S3 have Win 3.0 drivers. Hunting down OS/2 1.x drivers would be interesting as well.
