Seems like now's the perfect time to pop back into this conversation!

@feipoa and @pshipkov

All numbers pshipkov shared are mine from a few pages ago.
I tested everything on a Nice/Lion SuperEISA with the following configuration:

• BIOS version 1.21 was on the board at all points throughout testing
• All BIOS timings were tightest/most optimized
• A 40MHz DIP-14 crystal oscillator was used for clock generation
• I/O recovery was changed from 3 (lowest) to 11 (highest) for the Trio32 and Trio64 w/ N9 BIOS
• 64MB of 60ns parity 30-pin DRAM was used for all testing
• an Am5x86 ADZ was used for all testing at 160MHz
• L1 cache was in WT mode due to chipset limitations
• L2 cache was in WB mode (1024k on all tests)

Importantly, this board is pretty primitive. The chipset is from 1992, and the VLB implementation is not the best, relative to that of, say, an SiS471-based board. By this, I mean to say that the board cannot adjust between Synchronize/Transparent or T2/T3 with respect to VLB settings and timings. I have found that there are erroneous/strange results in testing that do not equivocate with those from other boards, for example, the S3 Vision868 widens the gap between itself and the Trio64 on my SiS471-based MTech R407e, providing WinTune2 scores in the ~14.7 kPixels/s range, as compared to the Trio64's ~12.9 kPixels/s score. Again, odd, but notable, is that the Mach64 is pretty much 1-for-1 outperforming the Mach32 on the SiS471-based b0ard, but on the SuperEISA, this is not the case. I posit this is due in part to an "incompatibility" with the board, after having scoured usenet archives and noting comments about lousy EISA configurations with the Mach64 and other such issues. The ARK1000VL is also not that much faster than its peers on the SuperEISA, if at all. On the SiS471-based board and even on OPTi895-based boards, this is not the case, as has been touted numerous times on Vogons.

On the SuperEISA, and probably due to the reasons above--namely, the SiS406/411 chipset--it seems that DOOM hits a hard wall at around ~57.7fps. This is not the case with the SiS471-board aforementioned. It becomes clear upon seeing some numbers that the faster VLB cards are in some way hindered by the VLB implementation of the SiS406/411 or that of the SuperEISA itself. On the MTech SiS471 board, I saw the Vision868 pull marginally ahead of the Trio64 in most regards, besting it in DOOM at 63.1 fps to 62.9 fps. A Mach64 Graphics Pro Turbo w/ 4MB VRAM pulled 55.2 fps, same as the older Mach32 w/ 2MB VRAM. The ARK1000VL there posted 68.0 fps, for reference.

I actually have a Diamond S3 968 VLB. It is in pieces.
It had bad memory modules, and in my haste to repair it, I lifted a few pads. I need to have someone else take a look at this because I am clearly incapable of fixing it properly myself. I would like to add that and see where things fall; clearly the Diamond 868 exceeds expectations, and that makes sense, given its VGA core's usage in the Trio64V+ and the overall slightly faster performance of said Trio64V+ vs the standard Trio64.

It's interesting your SPEA replica is faster. Have you ever tried swapping the BIOSes on your 968s and re-running the results?

I have noted that the 964 was slower than its siblings in DOS. The whole DRAM/VRAM speed difference thing doesn't seem to show truth in my testing, at least with both the ATi Mach32 Graphics Ultra Pro (VRAM) and Graphics Ultra Plus (same card but DRAM), the latter of which was supposed to be marginally faster as per PC Mag. I found this to be the case only in DOOM at 0.1 fps, so I consider the whole thing a wash. I think the S3 928 might be a better candidate for testing, but I only have an EISA/ISA version, and no DRAM counterpart.

I'm still waiting on a miracle to find a Trio64V+ on VLB to add to the roundup of the S3 Vision/Trio family. That would be interesting.

Like @pshipkov suggested, I will be re-testing the 864 with a Diamond BIOS, although I suspect not much will improve from the Paradise BIOS and its aggressive timings. I would be happily proven wrong though; that Paradise card has 60ns EDO modules on it. That could get interesting, given the Trio64 and 868 only have 70ns modules. Same with the 964.

If you want to see a few results laid out with the Trio64, ARK, and Vision cards, here's what I've got:

1ARK1000VL (1ws enabled)
2Wolf3D: 123.6 fps
3PC Player: 26.8 (10.3 fps @ 640x480)
4Doom: 57.7 fps
5Quake 1: 17.9 fps
6WinTune 2: 7634 KPixels/second
7
8MiroCrystal 20SV (S3 Vision964 w/ Diamond Stealth64 VRAM H BIOS)
9Wolf3D: 114.5 fps
10PC Player: 26.2 fps (10.1 fps @ 640x480)
11Doom: 56.0 fps
12Quake 1: 17.7 fps
13WinTune 2: 13525 KPixels/second
14
15S3 Trio64 (Stealth64 v2.02)
16Wolf3D: 120.3 fps
17PC Player: 26.7 fps (9.7 fps @ 640x480)
18Doom: 57.0 fps
19Quake 1: 17.9 fps
20WinTune 2: 12495 KPixels/second
21
22Diamond Stealth 64 DRAM (S3 Vision868) (Diamdond Stealth Video v1.01 BIOS)
23Wolf3D: 122.1 fps
24PC Player: 26.8 fps (9.8 fps @ 640x480)
25Doom: 57.1 fps
26Quake 1: 17.9 fps
27WinTune 2: 13347 KPixels/second
28
29Paradise Bahamas 64 (S3 Vision864)
30Wolf3D: 121.8fps
31PC Player: 26.8 fps (9.7 fps @ 640x480)
32Doom: 57.0 fps
33Quake 1: 17.9 fps
34WinTune 2: 11955 KPixels/second

----
@pshipkov

Really liking that PIO-3. I now see why it was loved by RedHill computers.
Voodoo 3 support is a nice surprise.
As for the flat curve in SpeedSys--I have noticed this more predominantly on Intel chipsets and with Intel CPUs. The Pentium shows the same on many 430xx boards, as does the Intel DX2/DX4 on 486 boards. I am curious if this is a glitch in SpeedSys, or if there's more to things here. Seems like 180MHz doesn't do too much on this board. Is this due to the relaxed timings or wait states for chipset I/O?
Seems like a middle-of-the-road board for the POD100. Not too bad, but definitely versatile.

Interesting that this board is so much better than the VIP-IO-2, based on the same chipset. Strange stuff.
VLB bridging on that board must really kill performance.

Either way, happy to read that this was trouble-free. Seems like a really nice assembly.
Do you know if PCI is bridged on this board? It seems to be pretty fast, but the -505 part is *I think* a bridge.

FIC made good stuff. The PA-2007 is easily my favorite Socket 7 board of all time.
The VA-502 is also quite good.
Same with the VIP-IO if you don't care about performance and just want a DX2/4 with PCI/VLB. It works with many cards without hassle.

Weird but interesting fact for prosperity: FIC is/was part of-or related to-Formosa Plastics. I believe VIA at one point was too.
Hence, when Intel sued VIA for the controversy surrounding their P4 chipset that resulted from reverse-engineering the P4's bus, they actually sued FIC as well for distributing such boards. Weird stuff.

Also, regarding the M919.
Never owned one and probably won't get one unless it's like @feipoa said, ~$20 would be my cap personally.
I think I've seen enough about it from your guys' tests.
That being said, it's an alright board overall.
I do think it's worth mentioning that the recent price spikes are probably due in part to a certain YouTuber many of us know.
His Quantex 486 has that board, and he made a few "blerbs" about it. He even made one about the replacement cache sticks.
Neat boards, in the sense that they're just so strange.
But nothing strange enough for me and my weird tastes, I guess 🤣

I recall synchronize/transparent having a modest impact on performance and that I needed to use the slower setting of the two (synchronize, I think) for my Diamond S3 968 card in my motherboard. I left it on synchronize when testing the replica SPEA Mercury P64V. I should really check to see if the the SPEA can work on the transparent setting. Unfortunately, I don't have a ARK VL1000 to compare with, which appears to form a baseline for VLB cards. I might consider trading VL1000 for a Trio64V+ replica as I have two of those.

Cool, you found a Mach64 with 4 MB. My card is 2 MB only. But it is not a great DOS performer compared to the alternatives.

Real bummer about your Diamond S3 968 card. I had replaced the chipset and all the RAM on my card without issue, but I had plenty of practice before hand. I wouldn't have done it without prior experience. I have some cards with pulled up pads. They are repairable, but it won't look as good.

Unfortunately, I don't own any of the 868, 864, or 964 cards. I'd like to see how the SPEA Mercury P64V compares here, but based on existing results, I think it will at least meet the Trio64 in DOOM. Is it is possible that the SPEA BIOS has 0 ws enabled while Diamond BIOS does not?

Yes, I put the DIAMOND BIOS into the SPEA replica card, but it showed the same slow result as the DIAMOND BIOS in my Diamond S3 968. However, I'm not sure if the two are really compatible. I tried to boot my system with SPEA replica w/DIAMOND BIOS and DIAMOND drivers, but Windows showed some distorted image and hung. These were just some quick tests for now, I've too many other repairs in the works to dig deeper now.

Also interesting might be trying the SPEA BIOS in the Diamond, but then I'd loose the MPEG decoder card's functionality, which is a cool perk of using the Diamond.

It's interesting that you note the modest performance difference. I found the difference, at least with the Mach64 VRAM, to be somewhere in the ~2-5% range. It seemed vastly to improve DOS benchmarks, and slightly helped in WinTune2. I suppose one interesting tidbit could be that the whole lot of S3 cards I've got does not work with Transparent, but seems fine on Synchronize. Intriguingly, the Mach32 DRAM and VRAM both boot and POST in Transparent, but immediately face corruption and require a reboot to change settings to synchronize. Unfortunately, my solution with the R407e is a bit ham-fisted at best. It is my "fastest" board, for standardization purposes here, but the stock BIOS is a mess and, while it works with that of the VL/I-SV2GX4, it's not stable with either and 1024k L2 installed. It will run DOS benchmarks, but will not complete 3D rendering tests. I plan on adding a write-up on it in a little bit, as it's a quirky thing and was sold with fake cache initially (a la PCChips), but seems to work well otherwise, but I digress.

The SPEA working on transparent could be interesting--if that's the case, it may pull into the league of the faster 868/Trio64/864-based cards. It seems marginally faster than the 964, but my 964 is a Miro whose BIOS I replaced with a Diamond BIOS. The Miro BIOS was MUCH slower. In fact, that was the problem with another 868-based card I have.

That Mach64 was a weird find, and I really adore it despite it underwhelming me. I bought it NIB, sealed through an mis-marked auction on fleabay. I paid more than I should've, but still got a fair deal. The 2MB add-on module was harder to source, but I found one through a nice fellow. Funny enough, I found another on a cheap Mach64 PCI. It's definitely not the best in DOS, but the image quality is superb, and the drivers are decent.

The 968 really bums me out. I have done some SMD soldering and figured it'd be easy. Just 8 chips and done. This experience is a lesson. I don't mind it not looking as good, but I definitely want to get it working for prosperity's sake. It actually came with a weird, seemingly-dead aftermarket 2MB add-on card. The card had memory modules dated to 1993 and 80ns VRAM. Strange stuff. My card is the model with 8 chips and it seems that finding an appropriate add-on card might be a chore these days.

I guarantee your thought is right. The SPEA Mercury P64V seems quite impressive. I think 0ws is a possibility. I recall in Madao's posts where the cards were designed he mentioned that wait states are related to the BIOS of these cards. This makes sense, as evidenced by the Miro->Diamond swap and resultant speed improvement suggests something similar. Diamond may have sacrificed 0ws for something else; their BIOSes are--from what I have seen--weird, in that they are generally faster but also less compatible with motherboards of the era. This is best seen with the SpeedStar64 ISA. I believe it's marginally faster than the STB Nitro64 in some regards; I have both, although my SpeedStar64 died over time. Swapping the BIOSes reveals a slight speed difference, but only in writes--the STB is faster in reads.

Interesting result with the Diamond BIOS still being slower. I bet you're right about the incompatibility. Drivers definitely could be half the battle. The other half might be the MPEG decoder you mentioned. I am very curious about the 968's VGA core. I understand it's the same as the 868's, right? In that regard, it should perform about on par with the Trio64V+ in DOS, although it also seems that the marginal VRAM/DRAM difference could put its siblings ahead.

When we are at the top tear performers, I'd consider 2-4 fps in DOOM to be a modest improvement. yeah, the Mach64 worked fine with Transparent on the VLI/I-SV2GX4. If you have one with working 4 MB, that's a keeper. Not much point in a Mach 64 with 2 MB when there's plenty faster 2 MB cards. My card doesn't have the 2 MB add-on headers. I do have a 4 MB ISA Mach64, which is just as unimpressive in DOS. Running 1280x1024x16bit on an ISA card on a 386 or early 486 would be painfully slow, so I keep the card as a novelty.

Your Diamond S3 968 has eight soldered RAM chips? I thought they all had 4 soldered chips for 2 MB, then a header for another 2 MB. Does your card have the less desirable IBM RAMDAC rather than the TI?

I made a comparison with various VLB cards on an Am5x86 @160MHz on SV2GX4, also with the S3 968 from Madao („by Matt“) with Spea BIOS (it has a slightly higher Memory clock i think) and the Diamond Stealth 64 Video VRAM.

The whole test description is here: https://dosreloaded.de/forum/index.php? ... enchmarks/

Very nice compilation. Do you have the Trio64V+ and S3 868 to include as well?

This was done with the Synchronize setting?

From your list, it looks like the Trident 9440AG, ARK1000, and Trio64 are fairly well matched in DOOM, but the SPEA still falls behind the Trio64 by 3.7 fps.

I have Trio64V+ too, but no S3 868. The S3 DRAM cards are a little Bit faster in DOS than the S3 VRAM cards. And the Trident 9440 is a very descent card.

The Tests were all done in transparent mode.

Transparent mode? That is curious. I was unable to get Windows 3.1 working in transparent mode with the Diamond S3 968 card at 40 MHz, L2-WB-1024K, 64 MB, L1-WB and tightest memory settings. Do you recall your system specs?

Since our S3 968 numbers match up, either your system is really in Synchronize mode, or mine is really in transparent mode.

My config was exactly that

But I didn‘t ran it in Windows while this test, only in DOS. I think the transparent Mode has no impact to some of the benchmarks - if i remember correct. Made this Test two years ago.

@40MHz the Diamond card sometimes has problems to boot up with the correct monitor frequency (out of range), so that i have no picture in BIOS mode and post screen. This effect is independent of transparent/syncronize mode.

Great info in the last several posts.

Thanks for chiming in PC-Engineer. Going to link your post to the directory of this thread.
I remember looking at your benchmarks before.
Noticing that i get slightly better perf with Ark1000VL on the same mobo and the settings, but slightly lower with Diamond S3 Trio64 than your Spea Mirage P64 VL.
Can you share its BIOS please ? Wonder if that will make a difference if i slap it on the Diamond card here.

Also, can you add numbers for S3 Trio64V+ based on Matt's design ?
It will be interesting to see where it fits.

Read the rest of your notes in the German forum - nodded at each one.
A small comment - CL GD-5434 ISA is the top dog in ... well ... ISA VGA/SVGA.

The 3 of you are the closest group to be able to provide an end-to-end overview of this S3 VLB business, despite each one is missing a card or two.

The missing link for me is how S3 Vision 868 (which according to WGS is faster than S3 Trio64) and S3 Trio64V+ stack-up.

Here's the SPEA BIOS for the Mercury P64V. Madao/Matt said it is the original BIOS, no hacking was done. If you test it on your Diamond card, also see if it works in Windows.

Thanks for the approval.

I have a few ISA cards (OAK, Trident, CL) here besides the Trio64V+ that I would like to add to the test. The GD5434 ISA and VLB are unfortunately also missing in my portfolio. That the ISA variant is the fastest of all ISA cards, I believe gladly.

The fact that your values are minimally higher with the ARK may be due to my boot configuration (mouse driver).

I will upload the BIOS file of the Spea Trio64 here later.

EDIT: BIOS File - Mirage P64 Trio64 attached

pshipkov wrote on 2022-12-17, 05:11:

but slightly lower with Diamond S3 Trio64 than your Spea Mirage P64 VL.
Can you share its BIOS please ? Wonder if that will make a difference if i slap it on the Diamond card here.

Looks like I misunderstood something. I thought Madao's design for the Trio64V+ used the STB Powergraph BIOS. Did he also do testing with a SPEA Mirage P64V? I know there is a modified STB Powergraph for 0 WS, but I don't think it was stable in Windows. I'd have to re-check. If there is an inbetween, and that is the Mirage, I'd also be interested.

EDIT: I thought you guys were hot for the Number 9 BIOS on the Trio64 VLB card. It was faster than the Diamond BIOS? Do I recall incorrectly?

Thx for the bios feipoa.

My head is spinning a bit from all the confusing card names and which bioeses they have been tested with.
So i need to review all the posts so far before say anything.

But one thing is clear to me already - vision 868 seems to be the unicorn in the s3 camp, judging by WJG's findings and overlapping them with pc-engineer and feipoa data. Too bad he does not have yet asus vli board to normalize to the rest. But the fact remains, i think.
So the outstanding question is how trio64v+ and vision868 compare.
Feels like it is going to be between these two.
WJG, if you are struggling with your 968 i would like to offer assistence. I have the equipment and the experience (i know i know - i dont sound modest) to potentially help you.
This motivates me as well to assemble the trio64v+ pcb i got from madao a while ago.

The Number 9 BIOS on the Trio64 VLB card didn't work well in Windows.

Not sure what BIOSes Madao tested with. I know about STB bios with lowered wait states.

Unfortunately neither of the two provided BIOSes worked on the Diamond S3 Trio64 VLB card here.

PC-Engineer wrote on 2022-12-17, 08:10:

EDIT: BIOS File - Mirage P64 Trio64 attached
MirageP64_501.zip

PC-Engineer, I take it you are using this BIOS on the intended card only - SPEA Mirage P64 ?

@pshipkov, yeah, Madao, to the best of my knowledge, used two BIOSes on the Trio64V+ replica: STB Powergraph 64V original, and STB Powergraph 64V 0ws. I also had my head wrapped a bit and confused the SPEA Mirage as a 64V+ card.

I have the BIOS from an original Spea Mirage P64 VLB with Trio64 without [V] or [+]. You can drive the Trio64 with EDO and FPM - for each you need another config in BIOS. I think the Spea has EDO-RAM.

Small detour - one more 486 motherboard with 1024Kb level 2 cache.

Chaintech 486SOM M102 (SiS 85C486, 85C497)

Condition is like new - crisp.
Very satisfying.

No EDO memory support despite BIOS sticker has year 1998 written on it.
Level 1 and 2 caches in write-back mode for best performance.
Very easy CPU jumpers.
Clock generator supports 60/66MHz, but no jumpers available for that. May try a simple tweak one day.
And that's pretty much all. No major peculiarities. Setup is straightforward.

--- Am5x86 @160MHz (4x40)

All BIOS settings on max.
The on-board IDE controller is not bad at all.

SpeedSys:

benchmark results

Not bad overall.
Best results in Windows accellerated GUI among all tested boards so far.
Lacks quite a bit in Wolf3D and Doom.

--- Am5x86 @200MHz (4x50)

BOOT to DOS never completes no matter what.
Tried hard to overcome that but without success.

--- P24T @100MHz (POD100, 2.5x40)

Inherently unstable with level 2 cache enabled.
This is addressed in the next version of the series - the SPM model.

---

This motherboard works pretty well in standard configurations up to 160MHz.
Not so much past that.
The POD100 issues are a red flag.
There are better options for sure.

Sad to hear the POD doesn't do well in the board, as that is the only 486 board I own ...

The next (SPM) version of the motherboard addresses the P24T compatibility problems.