Okay, checked the RAM, probability of 70ns raised to high, high, extreme, since they've all got -7 on... Probability of 10mhz bus dropped to super low, as I finally took a look in the Advanced Chipset Setup and it's 7.15 according to that .... oh dear, everything looks a bit pedestrian there....

Looks like no surprises...
- Oaks are known to be slow, and OTI-037 is the oldest
- Tridents 8900CL and 8900D - WHEN OPTIMALLY CONFIGURED - hit the ISA limit
- the 9000 series and pre-CL 8900 series are slower
- Tseng ET4000AX is also known to hit the ISA limit, but again - WHEN OPTIMALLY CONFIGURED - this particular card doesn't use the chip to the max

Yup, no surprises, results pretty much left my gob unsmacked, my timbers unshivered etc.

Multiple scandisks and other tools could only partially unscrew that DOS install on the 40MB though, but in patched up, floppy assisted boot mode, is everyone ready for ??? ...

Nancy Drew, Indiana Jones, Harry Potter, Crash Corrigan, Bit Wrangler and the Lost Trident of Onondaga

Legend has it, that a young male, Three Tooth the Lonesome, 7th son of the leader, was driven off from the pride, and roamed far away from the cosy basement of retro things, far away into the forbidden lands of the upstairs modern computer bench. He was seeking a pride of his own. This was not to be, and ostracised by the larger, more powerful VGA cards in his new environment, he cowered alone in a corner. Until one day the call for gathering went out, the grand assembly of all Tridentkind. He heard it's notes faintly on the breeze in his upland home, and whimsically stuck his unadorned head out, wondering if he should return. The lord of all Tridents looked upon his gathered peoples and realised there was one missing, Three Tooth the Lonesome, and thus sent an emissary to the far off lands, to recall #7 to participate in the grand moot....

Yah so I remembered where there was another one...

It is pretty much identical to RAMless Roddy, #5 and has the FCCID HNG890CL-BBD1TIA1 though currently lacks a bracket. I was thinking of putting RAM in #5 and running that as well, until I put them together and saw the match, then there didn't seem to be a point. Both have 7237 Rev E2 printed on the board. #5 is 3 weeks younger according to main chip. I doubt there were major changes to that in between. However, we ended up with two runs anyway, first run was slow, so I was wondering "Is this in 8 bit mode" well, Jumper 8 was unset, even though 0ws and FIFO were set, so set it for another run....

1  Card           Chipset  LandMark TopBench  Snooper 3Dbench Doom18S
2                            6.0    gmem                1.0   classic
3                           Chr/ms  microsec  Chr/sec   FPS  gametics
4
5Tri7    "7237"    8900D     900.22    987    42,986   11.9     8,661
6  "  J8 enable      "      1800.44    420    44,847   17.5     6,994

So #7 Three Tooth the lonesome, returns from exile to the adulation of his pride as he wins the games and gains glory... or it's inside the margin of error and level with the other two. I think I gotta give #6's RAM to #5 though, since he can probably do a bit more with it. Re-examined #6, no jumpers that look like they are equivalent to 1, 2 or 8 on other cards, need the real manual I guess.

I basically drowned the Kelvin 64 in contact cleaner, and gave it a last ditch effort... got some marginal improvement, but did a lot more of the same it did before and I think the HDD is re-pooched. There is nothing to set on it... though got a theory about maybe it's 14.318 oscillator is drifting and because of the 7.15, half of that bus speed, it might be clashing going in and out of synch. If I get drive patched up again tomorrow I might set bus to 8mhz or something and test.

edit:forgot pic...

I had but one more testable card, there's an ATI VGA 16 but it is corroded and needs detail work, the one left ready to go is an 8 bit WDC, which they rudely call the VGA basic or something, seems to be this... https://theretroweb.com/expansioncards/s/west … l-pvga1a-rev-x3 and I thought, ya know what, I got some 8 bit results for TVGA9000 and 8900 in there, the OTI runs 8 bit and so is the Cardinal apparently, may as well run this to compare 8 bit and to keep Mr Oak company at the bottom...

1  Card           Chipset  LandMark TopBench  Snooper 3Dbench Doom18S
2                            6.0    gmem                1.0   classic
3                           Chr/ms  microsec  Chr/sec   FPS  realtics
4
5WDC 8 BIT        PVGA1A1    512.80   1260    40,348    8.0    12,178

... and he murdered my boy, poor old Oak, Oakley Dokely, still in last place. I was actually quite impressed, especially since Doom score just snuck in underneath three of the Tridents, I guess two were clones, two unique tridents then, whupped 'em like "Basic this you bee-atches!". Oh yeah, nobody called me out, but realised I have been writing gametics for Doom when I mean realtics, gametics are the standard 2134 of that demo of course. I think I am wrapped up now, just wanted to make sure I was using my least performant Trident as a scratch monkey/guinea pig for RAM testing in case something went wrong, Oh, a Columbo: "Just one more thing...."

Sherlock Holmes Bit Wrangler and the Case of the Perfidious Orchid ...

Call me Ishmael stubborn, but I went back to the Orchid Kelvin 64 again... Set the board to clk2/6 about 6.67Mhz, and jimmied, jiggled and jemmied the connector until pic was stable unless I breathed on it. Like that, I could get a boot, but HDD not reading right, things would not load... so went back to setup, put it to clk2/5 8Mhz, and same thing. Both were behaving a hell of a lot better than 7.15 though so I wondered if there was something to my drifting clock/beat frequency theory... so give it the old vulcan nerve pinch again and back to setup for clk2/3... and it booted, and it did not refuse to load things at the prompt, but actually loaded things and crashed... hmmm progress I guess.. anyway, still could not achieve a single score... somehow it seems to specifically hate the HDD, but also maybe the I/O card, or maybe the chipset, like the inverse of a love triangle I guess, a hate triangle, jerk. Will try this again, another time, another post, another motherboard.

BitWrangler wrote on 2024-10-27, 22:29:

... and he murdered my boy, poor old Oak, Oakley Dokely, still in last place. I was actually quite impressed, especially since Doom score just snuck in underneath three of the Tridents, I guess two were clones, two unique tridents then, whupped 'em like "Basic this you bee-atches!".

Not surprsing, considering "Video Cacheable.. disabled".
The 8-Bit bus also slows down access to VGA BIOS, not just the VGA chip.

Back in the day, in real life, on an XT or AT, users had the choice to optionally load a "RAM BIOS" from disk.
It was a TSR version of VGA BIOS that ran in Conventional Memory (or in UMBs).
That was an alternative to enabling Shadow Memory in Setup.

That being said, if we want to see various VGA chips crawl then this configuration is fine, of course. 😉

Well multiply everything by 40% then, or by another 60% if you want to fool yourself that Pentium scores would be relevant, or divide by two for a slow 386sx, doesn't matter, the relative ranking will be the same. Besides which it's a bit weird to pick fault with the 8 bit scores for lack of BIOS caching, because XT machines can't usually do that, so what would be the point of knowing it's faster?

Still been messing with this system, and when I tried an additional HDD and got some weird stack overflow errors trying to get a directory listing, I had had enough of the I/O card, and the clunky ST351 which had taken a lot of soft damage to it's filesystem and was more patch than install. Since I am trying to get this read to do some seriousish stuff with I had to swap out I/O and go to a single WD Caviar 420MB drive. Yay, it's pretty slick now, HDD access tripled. I checked doom score on the AAmazing Trident 9000 and it miiight have improved by 100ish tics or might be run to run difference. That's about the only bench there that should be held back much by hdd...

So I thought, hang on a minute, if it's running so nice now, maybe I got rid of two sides of the hate triangle, so you guessed it, back in goes the Orchid... VGA connector still dodgy of course... So changes, "new" HDD, "new" i/o card but it's got the same UMC chip as the other one, 82c blah blah forgot. Checked settings were on 7.15 bus, BIOS caching disabled for comparison with above, but not tabling it, coz I'm lazy and it's theoretically different hardware...
Doom realtics, 5,334
LM6.0 vid 2841.16
Topbench vmem 320
3D Bench 21.7
Snooper 59,248

and for the whiners, balls out 13.3333Mhz ISA caches enabled... though I am wondering if Kelvin 64 does it's own BIOS caching because quick spurt with cache on at 7.15 said it was within a percent or two.
Doom 4,567
LM6.0 4428.11
Topbench 230
3Dbench 25.6
Snooper 63,330

Now since that was matching with DX50s in TOPBench then I think it's about as quick as you'll see out of a 486SX at 40, though I could spend hours of cache tweaking and memory swapping to get the last couple of percent out of it. Or stick a pentium sink on top and turn it up to 50 I guess.

Anyhoo, interesting thing in TOPBench, one of the systems it was near was trixter's, yes that trixter, Toshiba T4700CP with wait for it, a PVGA1A in it, which is stated as "fixed" yeah, proper fixed if he's getting numbers like that from it, so I guess I gotta go dig out what exactly he's done to that, see if my WD "basic" can really lay down the smack....

But yeah, supposed to be finding a normal with this bench rig for doing shit...

Nice work and thank you for sharing, BitWrangler.

If I had a Realtek based VGA card, like an RTG3105 based one, I would have shipped it to you (had you wanted it) just to add that as a datapoint.

EDIT: RTG3103 is apparently worse

I have one (a 310x) somewhere, “found” it when I was still using a Trident 9000i. It was slower, quirkier and had worse image quality., so the Trident went back in.

Today I have a whole gaggle of Tridents, I tend to like them.

These tests are epic, thank you.

darry wrote on 2024-10-31, 06:57:

Nice work and thank you for sharing, BitWrangler.

If I had a Realtek based VGA card, like an RTG3105 based one, I would have shipped it to you (had you wanted it) just to add that as a datapoint.

EDIT: RTG3103 is apparently worse

Really, you'd put that in the hands of a monster? Heh, I mean thanks for the thought.

But what I was leading into is, I have an RTG3105 on the casualty list. The graphics gumstick PT-505S. What I seem to have done to it some 20+ years ago, was remove the sockets and try to install two matching SOJ RAM chips with an overhot soldering iron and too high a temperature solder... in overconfident neanderthal mode. This as you see was something of a complete and utter failure. Though in my defence, the manufacturers did absolutely save 0.2 of a penny on the thickness of copper cladding, and the part was probably worth in the region of zero to zip at the time. Upon reexamination of the damage, I might not have to patch too much, only three of the lines on the line of bottom right DIP pins go through underneath, two on top of that, and I think only 2 up two down on the left one... so make all the lines good that "go somewhere" and I might reanimate it. You will note that the 16bit part of the slot only has pads to a presumed "bus termination" resistor pack that was never installed, plus ground on the back. Therefore the only signalling is happening on the 8 bit section.

BitWrangler wrote on 2024-10-30, 18:56:

Besides which it's a bit weird to pick fault with the 8 bit scores for lack of BIOS caching, because XT machines can't usually do that, so what would be the point of knowing it's faster?

Early 286 PCs such as Model 5170 didn't have shadow memory, either. 😟
That's why there was an RAM BIOS utility.. It copied a VGA BIOS from disk into DOS memory (24 to 32KB). Like those VESA VBE TSRs still did in the 90s.

Considering that PC BIOS alone is accessed very often via 8-Bit bus does make the situation even more important, I think.
It's one of the main reasons why XTs are so slow, I think. Access to ROM code hogs the bus.
A cozy flash-based or RAM-based BIOS chip with 80ns or 120ns is way quicker than your typical 27xxx EPROM at 250ns.

BitWrangler wrote on 2024-10-31, 17:00:

darry wrote on 2024-10-31, 06:57:

Nice work and thank you for sharing, BitWrangler.

If I had a Realtek based VGA card, like an RTG3105 based one, I would have shipped it to you (had you wanted it) just to add that as a datapoint.

EDIT: RTG3103 is apparently worse

Really, you'd put that in the hands of a monster? Heh, I mean thanks for the thought.

But what I was leading into is, I have an RTG3105 on the casualty list. The graphics gumstick PT-505S. What I seem to have done to it some 20+ years ago, was remove the sockets and try to install two matching SOJ RAM chips with an overhot soldering iron and too high a temperature solder... in overconfident neanderthal mode. This as you see was something of a complete and utter failure. Though in my defence, the manufacturers did absolutely save 0.2 of a penny on the thickness of copper cladding, and the part was probably worth in the region of zero to zip at the time. Upon reexamination of the damage, I might not have to patch too much, only three of the lines on the line of bottom right DIP pins go through underneath, two on top of that, and I think only 2 up two down on the left one... so make all the lines good that "go somewhere" and I might reanimate it. You will note that the 16bit part of the slot only has pads to a presumed "bus termination" resistor pack that was never installed, plus ground on the back. Therefore the only signalling is happening on the 8 bit section.

Hey, I'm the trusting type.

Hope you can get that gumstick one working but, if not, you hopefully can source another one.

Since it was only a dozen places down the "stuff I'm meaning to look at list" anyway, and I might have plans for it based on it's one most obvious quality, that it's a shortarse, I decided to do more with the PT-505S RT3105. Plus it seemed like the right day for raising the dead or summoning ancient evil or whatever.

The Three Stooges, Bugs Bunny, Bit Wrangler in The Tiny Pine-y Caper...

First off I had to tabulate all the pins that go to and from the DIP holes and if they went to anything important. Having done that, I couldn't believe the answer so I went over it again. Yah in all that mess, there was only one pin needing to be reconnected to bring it back. So after lunch, got to it, found I'm a bit shaky from the shoulders today, if this needed many more wires I might have had to bail. Picked out the smallest wire I could find easy, though it looks so massive on there I guess I should have gone digging for the kynar stuff. Anyway, tinned it and tagged it. Stuff it in the motherboard, fire it up, aaaargh, scrambly screen.... okay, breathe... power off, reseat, give it a good glaring at, I have a soldering iron here, do you really wanna try me again... press it down good, power up.... "Realtek BIOS ... 256 k" yussss, POST, prompt... looks like it worked!

Hmmm, nothing picking it up as a Realtek... Oh yah, saw a VGA ID tool in the graphics workshop for DOS folder, yup, Realtek BIOS IDed, and I think it's just pulling those modes from BIOS and the bottom two won't work without 512K which I doubt this particular card has a hope in hell of seeing now. Then at the bottom it thinks it's an ET3000, huh? Is that just 'coz it's a simple framebuffer VGA with zero features, or did Realtek rip off the ET3000 ??? Anyhoo, let's get a 2nd opinion, ah, SpeedSys sees it, I didn't really notice that little VideoRAM test there, everything kinda small on it's display, anyhoo.. ah yeah, I think I said it was a WD Caviar, apparently I have a Quantum Trailblazer in it 🤣 yeah probably got the WD caviar on the brain because I have a 7.5GB one sitting on the bench where I'm working so keep seeing it, derp. It actually made me look though, yah, actually a Quantum there.

Okay so the tests, done first on 7.15Mhz as above with BIOS caching disabled as above, then with 13.333Mhz ISA with BIOS caching on "balls out" mode underneath...

1  Card           Chipset  LandMark TopBench  Snooper 3Dbench Doom18S
2                            6.0    gmem                1.0   classic
3                           Chr/ms  microsec  Chr/sec   FPS  gametics
4
5PT-505S          RT3105     774.05   1007    51,656   10.8     8,858
6 " 13.3M ISA BIOS cached "  882.44    906    57,693   12.1     7,880

Well that's a weird hodgepodge of being all over the place... 3rd off bottom for LM6.0 and Topbench mem access speed, then gets higher than original set of cards, second only to Kelvin 64 on the Snooper test, 3rd from last again on 3Dbench, then gets a middle of the pack, ahead of the slower Tridents DOOM score... wild.... I don't think the cache on or off makes much difference to it, it's BIOS is inside the ASIC, if it can't talk to it fast in there IDK what would improve it realistically. Anywayyyyy, I guess it's overall kinda middle of pack and maybe worth a consider for an 8 bit card, given that you could get one a lot cheaper than a 8900D/CS. I am kind of glad to see that though, if I use it for a low profile situation at least I'll feel it's not worst choice, just because it's the only choice, it's not actually worst of the worst.... and also that's likely to be pitched at late 80s to 92ish era so would be fine with lower perf.

BitWrangler wrote on 2024-10-31, 23:11:

Since it was only a dozen places down the "stuff I'm meaning to look at list" anyway, and I might have plans for it based on it' […]
Show full quote

Since it was only a dozen places down the "stuff I'm meaning to look at list" anyway, and I might have plans for it based on it's one most obvious quality, that it's a shortarse, I decided to do more with the PT-505S RT3105. Plus it seemed like the right day for raising the dead or summoning ancient evil or whatever.

The Three Stooges, Bugs Bunny, Bit Wrangler in The Tiny Pine-y Caper...

First off I had to tabulate all the pins that go to and from the DIP holes and if they went to anything important. Having done that, I couldn't believe the answer so I went over it again. Yah in all that mess, there was only one pin needing to be reconnected to bring it back. So after lunch, got to it, found I'm a bit shaky from the shoulders today, if this needed many more wires I might have had to bail. Picked out the smallest wire I could find easy, though it looks so massive on there I guess I should have gone digging for the kynar stuff. Anyway, tinned it and tagged it. Stuff it in the motherboard, fire it up, aaaargh, scrambly screen.... okay, breathe... power off, reseat, give it a good glaring at, I have a soldering iron here, do you really wanna try me again... press it down good, power up.... "Realtek BIOS ... 256 k" yussss, POST, prompt... looks like it worked!

Hmmm, nothing picking it up as a Realtek... Oh yah, saw a VGA ID tool in the graphics workshop for DOS folder, yup, Realtek BIOS IDed, and I think it's just pulling those modes from BIOS and the bottom two won't work without 512K which I doubt this particular card has a hope in hell of seeing now. Then at the bottom it thinks it's an ET3000, huh? Is that just 'coz it's a simple framebuffer VGA with zero features, or did Realtek rip off the ET3000 ??? Anyhoo, let's get a 2nd opinion, ah, SpeedSys sees it, I didn't really notice that little VideoRAM test there, everything kinda small on it's display, anyhoo.. ah yeah, I think I said it was a WD Caviar, apparently I have a Quantum Trailblazer in it 🤣 yeah probably got the WD caviar on the brain because I have a 7.5GB one sitting on the bench where I'm working so keep seeing it, derp. It actually made me look though, yah, actually a Quantum there.

Okay so the tests, done first on 7.15Mhz as above with BIOS caching disabled as above, then with 13.333Mhz ISA with BIOS caching on "balls out" mode underneath...

Copy code to clipboard

1  Card           Chipset  LandMark TopBench  Snooper 3Dbench Doom18S
2                            6.0    gmem                1.0   classic
3                           Chr/ms  microsec  Chr/sec   FPS  gametics
4
5PT-505S          RT3105     774.05   1007    51,656   10.8     8,858
6 " 13.3M ISA BIOS cached "  882.44    906    57,693   12.1     7,880

Well that's a weird hodgepodge of being all over the place... 3rd off bottom for LM6.0 and Topbench mem access speed, then gets higher than original set of cards, second only to Kelvin 64 on the Snooper test, 3rd from last again on 3Dbench, then gets a middle of the pack, ahead of the slower Tridents DOOM score... wild.... I don't think the cache on or off makes much difference to it, it's BIOS is inside the ASIC, if it can't talk to it fast in there IDK what would improve it realistically. Anywayyyyy, I guess it's overall kinda middle of pack and maybe worth a consider for an 8 bit card, given that you could get one a lot cheaper than a 8900D/CS. I am kind of glad to see that though, if I use it for a low profile situation at least I'll feel it's not worst choice, just because it's the only choice, it's not actually worst of the worst.... and also that's likely to be pitched at late 80s to 92ish era so would be fine with lower perf.

It's alive!

That's beautiful!

I'm drunk!

Shine on, you crazy diamond, necromancer of crappy ISA VGA cards!

Yup, I'm still drunk!

i have no objection on this, just a suggestion: the oak037 maybe slow, but if you can get a tseng et3000, it should be equally slow or even slower.
still its hard for me to believe that the oak037 supports 1mb of ram, can you post a picture of it?
the trident 8900cl still has some debate: some experiments seem to indicate that its as slow as the older 8900c, while others suggest that its as fast as the newer 8900d.

The ATI EGA Wonder 800 is also worth a try.
It doesn’t have support for MCGA (mode 13h), but VGA (mode 11h and 12h).

The ATI VIP would be interessanting, too. It's predating ATI's use of custom VGA cores. It rather uses a Chips P82C441.

Both are early designs from 1987.

noshutdown wrote on 2024-11-01, 05:20:

i have no objection on this, just a suggestion: the oak037 maybe slow, but if you can get a tseng et3000, it should be equally slow or even slower.
still its hard for me to believe that the oak037 supports 1mb of ram, can you post a picture of it?

You are right, I had 128x8 RAMs on the brain, it has 4x64, for only 256MB total, same as the WDC. Sloppy first thing in the morning pic tho. Some guy needed PAL details for a different card with OTI-037, but this one appears to do it all in 74 series logic, so they may be useful to study to know what the PAL is replacing.

I think there are ppl who have tested 8900CL in 16 bit mode with zero wait states and FIFO enabled, and people who haven't.

Other notes...

keropi tested some Realtek implementations and also seems to have found the BIOS caching doesn't seem to move the needle much. Re: 8-bit VGA BIOS Performance in 16-bit Systems

Performance notes: Some cards are getting pretty much near their limit on this board/CPU combo, they stick at around these numbers whatever more powerful CPU you try to push them uphill with. Testing with Pentium and higher CPU shows this on Thandor's site and VGAmuseum. Others however scale a bit more until they are also running out of potential.

There was a machine I forgot about had an ISA card in, checked it, nothing very different... it's an identical PCB to Trident #1 and had a 9000B on it instead of the A, but since you can barely slide a sheet of paper between the A on #1 and the C on #3 also on very similar board, I did not see any point in testing it. Oh most exciting difference was that the bottom set of jumpers were not populated, so maybe avoid the one with the B if you want it for 8 bit.

So, funny story... I was going through disks, looking for any utils which would let me poke about in VGA RAM and test it, and I found the TVGAUTIL file on actually a TGUI 9680 disk, disk 2 of 3 of UX 6.5 set, and some interesting stuff in there. It sayyyys TVGA right there don't it, maybe these utils work for all of Trident's cards... So I've got the #3 Card, AAmazing 9000C in there, and I'm looking at random stuff, and it appears to be working as much as it will let me do on the no feature 256k econo model, i.e. I'm not gonna set SVGA high color modes or anything. But anyway I've gone to the SVM util and set my video mode to 640x480 x16c graphics from normal VGA text I think, and it dropped from 72hz to 60 hz... I don't know at all if this is relevant but bear with me...

Then I see the SETBOARD util and run that with no parameters to see what it says, all sorts of settings for the PCI cards that I don't expect to work, then I see at the bottom, DRAM clock speeds, and think, "Wouldn't it be funny if that sped up the DRAM clock on this clunker." so I typed SETBOARD DCLK50 and pressed return... and it returned to the prompt.. hokay, so did anything happen...

LM6.0 2026.89
TopB 380
Snooper 53,771
3DBench 18.5
Doom18s 7,853

Holy hell yes something happened, I have no clue what actual DRAM speed this ASIC might actually be running at, but it flipped a bit and sped the damn thing up!!!

Hoo boy this is exciting, reboot and try for 58 (Next setting, other two are 64 and 75) annnd going straight to SETBOARD, it doesn't return to prompt, hmmm, feel the RAM chips, well they are a little warm, and they are marked -12, 120ns, when the manual for this card says use 100ns slowest, so guess we were doing good at "50".... anyway, reboot out of the lock, try to set to 50 again... same thing, lockup... hmmmph... is this conditional though? First time I went to SVM first and changed the video mode, so let's do that again, set 640x480x16, done, SETBOARD DCLK50 and we're back to the prompt and fast again, wooohooo..... though this is as far I have got, I don't know if it just gave it long enough for the chips to cool down, the extra messing around, or whether it really needs the SETVM thing first, maybe it unlocks settings or something.

Anyhoo, lot more testing of this to be done, like on a card with faster DRAM for instance, or on one of the 8900CL/D to see if that also gets faster.

Edit: bounced over to machine with floppy and attached TVGAUTILS from my disk that I used...

Edit II: was looking up some chipset info to figure out what is going on, the way DOSdays told it, it could be that DRAM clock programming was a new feature on the 9000C over the B and A, but sometimes marketing info puts the "all new" spin on every revision and repeats what was "new" with introduction. Had a look at the datasheet, but it doesn't get spelled out super obviously, i.e. need a deep dive into the workings so you know that's what they're doing when EA goes to register 1A or something like that.