Cy6x86. Some fun memories with it.
My neighbor bought a preassembled PC with top of the line Cyrix 6x86 around 1998-1999.
Remember how disappointed he was with it. Everything stuttered.
I spent quite a bit of time on it trying to squeeze the most out of it.
It was a good experience and a reminder that Intel hardware is worth the extra $.
But i liked the unpretentious nature of the Cyrix cpu - good solution for lots of use cases.

EDIT: fixed syntax errors

Cyrix 6x86 always felt like a great "windows accelerator" to me, the integer side really lapped up 9x code, think it was still stuffed with 16 bit remnants. On the desktop you would think you had something twice as fast as the competition. Up until MII I managed to run all of them at the clockspeed of the PR rating with a decent heatsink on. After MII, I tried.. but mostly only got halfway there. Come to think of it, the old PR was only 25% higher normally... the MII (When it was marketing name not from core designation) PR was +40% in some cases. The way I thought of it back in the 90s was, I can get a cyrix with a base clock speed 1 or 2 speed notches higher than an intel at same price, which means about the same FPU power for same money, but MUCH more integer. It's not pointed out that heavily in modern comparisons, but you weren't comparing a P100 against a 100Mhz PR120+ back in the day, it was more like you could choose a P100 or a P166+. Anyway, not having their own fab, they were probably not able to tune as well as AMD and Intel for best use of the process, which left them lagging later. Seemed to get ahead with their last gasp of the non-mmx models then got a few months, then over a year behind by the time Via bought them.

The recent chat about Trident video cards here and few more conversations in private prompted me to share a list that i got long ago from somewhere.
Ended up testing most of the ISA/VLB and some of the very late PCI models before the 3D series. Marked the ones i liked best.
This was not a targeted effort, just over time enough Trident cards passed through here and were casually checked.
Many of the test results in previous posts, such as this one show only the preferred cards from each crop/brand/chipset to reduce clutter.
There are also other results scattered throughout the thread that should probably be linked here at some point.

Desktop, ISA

- TVGA 8200LX (1987)
- TVGA 8800 (1988) - first S/VGA compatible chipset (ISA), 512KB framebuffer
--- 8800BR (512KB, 128KB banks only)
--- 8800CS (512KB, 64KB banks + old/new mode support)
- TVGA 8900 - high-color (65,536) display-mode support, 1MB framebuffer
--- 8900B (up to 1MB)
--- 8900C (up to 1MB, SVGA, ISA)
--- 8900CL (up to 1MB 2MB, SVGA, ISA/VLB. Slightly faster than earlier 8900 cards) <- EXCELLENT CARD
--- 8900D (up to 1MB 2MB, 8900CL w/ bugs corrected) <- EXCELLENT CARD
--- 8900DR (SVGA, ISA, max 1MB)
- TVGA 9000 - first integrated (VGA+RAMDAC) VGA chipset
--- 9000 (low component version)
--- 9000B (1992)
--- 9000C (1992) - External RAMDAC
--- 9000i - (rev. a/b/c, 512KB, 9000 w/ onchip 15/16bit DAC + clock gen)
--- 9000i-1 (1994) - appeared on Trident's VC512TM ISA video cards
--- 9100B - Slightly faster 9000

Desktop, VLB

- TVGA 92xx, TVGA 938x, TVGA 94xx - first Windows accelerators
--- 9200CXr
--- 9380
--- 9385
--- 9388
--- 9400CXi (max 2MB, truecolor, onchip 24bit DAC + clock gen)
--- 9420DGi (PCI, 9400 w/ 2D acceleration )
--- 9430DGi (PCI, 9420 w/ hardware cursor)
- TGUI 9440 (1994) - first performance competitive Windows 2D-accelerator (2MB PCI/VLB)
--- 9440AGi (PCI, 9430 w/ 16bit DAC interface + programmable clock) <- EXCELLENT CARD
--- 9440-1 <- EXCELLENT CARD

(We touched on 9440AGI and 9440-1 in previous posts. Basically the situation with them is the same as with Cirrus Logic GD-5426 and 5428. In other words - no practical difference.)

Desktop, PCI

--- 9440-3 - Unknown changes, differently designed Trident logo
--- 9460
--- TVGA9470

- TGUI 966x - similar to 9440, 64-bit data path
--- 9660
--- 9660XGi
- 968x - motion video accelerator (zoom + YUV->RGB, DirectDraw overlay)
--- 9680 - (64bit, 1-4MB, 9660 w/ video acceleration)
--- 9680XGi
--- 9682 (9680 with video in)
--- 9683
--- ProVidia 9685 (TV video out, UMA) <- EXCELLENT CARD

------ My interests end here. No idea about the stuff below. Adding it for completeness. ------

- 3DImage9750, 3DImage9850 - first Windows 3D-accelerators (4-8MB PCI/AGP)
--- 3DImage 9750 (64bit, PCI/AGP, TV out)
--- 3DImage 9753
--- 9753WAVE (9753 + 32voice wavetable, 0.35ym, 208PQFP)
--- 3DImage 9754
--- 3DImage 9750DVD (9750 with DVD decoder)
--- 9783
--- 9785
--- 3DImage 9850 (Faster 9750 core, AGP2X, AGP only)
--- 3DImage 985DVD (DVD Acceleration)
--- 3DImage 9850+ (Faster)
--- 3DImage SME-GMA (Superscalar, Memory Enhanced, Graphics Mode Adaptability. Laptop graphics chip with 6MB embedded DRAM, smooth scaling for non-native resolutions and hardware MPEG2 decoding for DVD playback. Based on 9850+ Core.)

- Blade3D 9880 (1999) - first performance competitive Windows 3D-accelerators (8MB PCI/AGP)
- Blade3D Turbo 9880T - 135Mhz clock speed, up from 110Mhz on the Blade 3D vanilla.
- Blade T16 9950 - little known
- Blade T64 9970 - dual-pixel quad-texture engine, AGP4X, 32MB of memory, 64-bit data path
- Blade XP 9980 Same as T64 except 128-bit data path, 200Mhz vs. 166Mhz
- XP4 - DirectX 8 chip.
- XP4E - AGP8x support.
- XP8 (cancelled) - DirectX 9 chip, marketed for under $100US.
- XP10 (cancelled) - PCI Express controller.

Mobile

- Cyber9320
- Cyber9382
- Cyber9385
- Cyber9397 and Cyber9397DVD
- 9525DVD
- CyberBlade
CyberBlade Ai1
CyberBlade e4-128
CyberBlade i1
CyberBlade i7
- Blade XP
- XP4
XP4m16/XP4m32 - embedded memory.
- XP8 (cancelled) - DirectX 9 chip.

Integrated

- ALi CyberALADDiN-T
- ALi CyberALADDiN-P4 (CyberBLADE XP2)
- ? (codename Napa2T)
- ? (codename Napa2-P4)
- ? (codename Napa2-Banias)

@bitwrangler

I admire Cyrix for what they did back then, but the other side of the coin is that no kid at the time was bragging about having a Cyrix CPU.

pshipkov wrote on 2022-10-01, 06:50:

--- 8900CL (up to 2MB, SVGA, ISA/VLB. Slightly faster than earlier 8900 cards) <- EXCELLENT CARD
--- 8900D (up to 2MB, 8900CL w/ bugs corrected) <- EXCELLENT CARD

I can also attest to those being excellent. In my experience they were significantly faster than the previous 8900s when the FIFO and zero wait are enabled

Wait...the 8900s can accept 2MB? Does such a card exist? Does the extra memory actually do anything?
Anyone ever seen a 9100? Not me.

Something interesting about the 9000i. My friend had this one in his PC. Despite the card having the high colour RAMDAC, he could never get high colour modes to work. I suspect that only 320x240 can do it, as the card is limited to 512kb DRAM, and I doubt that's enough to display 640x480 in 15/16 bit.

Updated the above post with 2 more "exceptional" tags. Had to remove them yesterday for some reason.

8900D is a bugfix variant of CL. Many D cards come with 24-bit DAC, while the CL come with 8 or 16-bit ones.
The description about CL/D supporting up to 2Mb of RAM is incorrect. Fixed it.

The 9000 series are based on cost-reduction design which results in slower cards.
The story about your friend checks-out.

Never saw a 9100 chip myself.

While on the subject of 2Mb ISA cards:
I have this Diamond Speedstar Pro based on Cirrus Logic GD-5429 (shown here) that has solder pads for second megabyte of RAM.
Glued the second megabyte and tried different configs with the SMD components but so far only 1Mb is recognized.
If somebody has an idea how to go about it let me know..

pshipkov wrote on 2022-10-01, 06:50:

Desktop, ISA

--- 8900CL (up to 1MB 2MB, SVGA, ISA/VLB. Slightly faster than earlier 8900 cards) <- EXCELLENT CARD
--- 8900D (up to 1MB 2MB, 8900CL w/ bugs corrected) <- EXCELLENT CARD

maxtherabbit wrote on 2022-10-01, 13:37:

I can also attest to those being excellent. In my experience they were significantly faster than the previous 8900s when the FIFO and zero wait are enabled

I got the Trident 8900CL for my ISA only 486 board through advice given here at Vogons, based on it being cheaper to get that many other cards and with good performance. I had no idea it was "excellent" though, happy to hear that. My lack of knowledge forces me to ask, what do you mean by enabling "the FIFO and zero wait"? Is that something I should do as well to boost speed? In any case if it's necessary my BIOS is very barebones I have to say

andre_6 wrote on 2022-10-02, 15:26:

pshipkov wrote on 2022-10-01, 06:50:

Desktop, ISA

--- 8900CL (up to 1MB 2MB, SVGA, ISA/VLB. Slightly faster than earlier 8900 cards) <- EXCELLENT CARD
--- 8900D (up to 1MB 2MB, 8900CL w/ bugs corrected) <- EXCELLENT CARD

maxtherabbit wrote on 2022-10-01, 13:37:

I can also attest to those being excellent. In my experience they were significantly faster than the previous 8900s when the FIFO and zero wait are enabled

I got the Trident 8900CL for my ISA only 486 board through advice given here at Vogons, based on it being cheaper to get that many other cards and with good performance. I had no idea it was "excellent" though, happy to hear that. My lack of knowledge forces me to ask, what do you mean by enabling "the FIFO and zero wait"? Is that something I should do as well to boost speed? In any case if it's necessary my BIOS is very barebones I have to say

Which exact video card is it, can you post a pic? The zero wait and FIFO are controlled by jumpers

maxtherabbit wrote on 2022-10-02, 17:46:

andre_6 wrote on 2022-10-02, 15:26:

pshipkov wrote on 2022-10-01, 06:50:

Desktop, ISA

--- 8900CL (up to 1MB 2MB, SVGA, ISA/VLB. Slightly faster than earlier 8900 cards) <- EXCELLENT CARD
--- 8900D (up to 1MB 2MB, 8900CL w/ bugs corrected) <- EXCELLENT CARD

maxtherabbit wrote on 2022-10-01, 13:37:

I can also attest to those being excellent. In my experience they were significantly faster than the previous 8900s when the FIFO and zero wait are enabled

I got the Trident 8900CL for my ISA only 486 board through advice given here at Vogons, based on it being cheaper to get that many other cards and with good performance. I had no idea it was "excellent" though, happy to hear that. My lack of knowledge forces me to ask, what do you mean by enabling "the FIFO and zero wait"? Is that something I should do as well to boost speed? In any case if it's necessary my BIOS is very barebones I have to say

Which exact video card is it, can you post a pic? The zero wait and FIFO are controlled by jumpers

I can't post a pic of my exact card at the moment, sorry, I'm unsure if it's an 8900CL-B or 8900CL-C, but jumpers wise they have the same layout:
https://www.vgamuseum.info/images/vlask/tride … a8900cl-cfb.jpg - 8900CL-C
http://www.vgamuseum.info/images/palcal/tride … cl-b_top_hq.jpg - 8900CL-B

Its Ebay ad is long gone, this is the main order's pic, by the PCB lighter green tint looks like an 8900CL-C:
https://i.ebayimg.com/images/g/cqoAAOSwU9RgrNiT/s-l140.jpg

Sorry I couldn't be more helpful at this time

Edit: found one, see below please, thanks

You picked a card with 8-bit RAMDAC which means maximum 256 colors on screen. Not ideal, but not the end of the world really.

Look at the jumpers on the card - some of them control "wait states" of the video card. It is probably JP1. You can check this by running some of the simple DOS interactive graphics tests we often talk about around here.

pshipkov wrote on 2022-10-02, 18:48:

You picked a card with 8-bit RAMDAC which means maximum 256 colors on screen. Not ideal, but not the end of the world really.

Look at the jumpers on the card - some of them control "wait states" of the video card. It is probably JP1. You can check this by running some of the simple DOS interactive graphics tests we often talk about around here.

Didn't think such differences would exist in 8900CL model variations, were you able to see that through my awfully low resolution EBay pic, my actual card? Wow

maxtherabbit wrote on 2022-10-02, 17:46:

Which exact video card is it, can you post a pic? The zero wait and FIFO are controlled by jumpers

Edit: had a picture somewhere after all, here it is. Is this the "8-bit RAMDAC" version after all?

I looked at enough Trident ISA cards to see the pattern from your low-res image.
The DIP-28 RAMDACs are 90% 8-bit, or less.
In your particular case the RAMDAC is actually 6-bit, but that still results to maximum of 256 colors on screen.
Many CL models come with 8-bit DACs. Some with 16-bit.
The D variants come with Trident based 24-bit RAMDACs. They look square and are easy to distinguish from the elongated DIOP-28 ones.

pshipkov wrote on 2022-10-02, 21:15:

I looked at enough Trident ISA cards to see the pattern from your low-res image. The DIP-28 RAMDACs are 90% 8-bit, or less. In y […]
Show full quote

I looked at enough Trident ISA cards to see the pattern from your low-res image.
The DIP-28 RAMDACs are 90% 8-bit, or less.
In your particular case the RAMDAC is actually 6-bit, but that still results to maximum of 256 colors on screen.
Many CL models come with 8-bit DACs. Some with 16-bit.
The D variants come with Trident based 24-bit RAMDACs. They look square and are easy to distinguish from the elongated DIOP-28 ones.

That's impressive in any case.

Such a shame, I guess people forgot to mention that I had to be on the lookout for that when they advised me the card. It's a low to mid-level 486 ISA only board with a DX-2 66, and 16MB so I would never play Doom in high detail on it on a large window anyway. Could you please point me out in the photo where I should apply the jumper? In beginner lingo, how would I notice a difference in performance, does one jumper setting apply the FIFO and zero state changes that maxtherabbit was talking about? Thanks in advance

andre_6 wrote on 2022-10-02, 19:21:

Didn't think such differences would exist in 8900CL model variations, were you able to see that through my awfully low resolutio […]
Show full quote

pshipkov wrote on 2022-10-02, 18:48:

You picked a card with 8-bit RAMDAC which means maximum 256 colors on screen. Not ideal, but not the end of the world really.

Look at the jumpers on the card - some of them control "wait states" of the video card. It is probably JP1. You can check this by running some of the simple DOS interactive graphics tests we often talk about around here.

Didn't think such differences would exist in 8900CL model variations, were you able to see that through my awfully low resolution EBay pic, my actual card? Wow

maxtherabbit wrote on 2022-10-02, 17:46:

Which exact video card is it, can you post a pic? The zero wait and FIFO are controlled by jumpers

Edit: had a picture somewhere after all, here it is. Is this the "8-bit RAMDAC" version after all?

J1 at the top of the card, and J2 at the bottom right are the two relevant jumpers.

Different manufacturers use different values for these, i.e. whether we want the jumper on or off. So you should experiment with all 4 possible combinations and see which yields the best speed.

maxtherabbit wrote on 2022-10-02, 22:30:

andre_6 wrote on 2022-10-02, 19:21:

Didn't think such differences would exist in 8900CL model variations, were you able to see that through my awfully low resolutio […]
Show full quote

pshipkov wrote on 2022-10-02, 18:48:

You picked a card with 8-bit RAMDAC which means maximum 256 colors on screen. Not ideal, but not the end of the world really.

Look at the jumpers on the card - some of them control "wait states" of the video card. It is probably JP1. You can check this by running some of the simple DOS interactive graphics tests we often talk about around here.

Didn't think such differences would exist in 8900CL model variations, were you able to see that through my awfully low resolution EBay pic, my actual card? Wow

maxtherabbit wrote on 2022-10-02, 17:46:

Which exact video card is it, can you post a pic? The zero wait and FIFO are controlled by jumpers

Edit: had a picture somewhere after all, here it is. Is this the "8-bit RAMDAC" version after all?

J1 at the top of the card, and J2 at the bottom right are the two relevant jumpers.

Different manufacturers use different values for these, i.e. whether we want the jumper on or off. So you should experiment with all 4 possible combinations and see which yields the best speed.

Thank you so much, out of curiosity, would you recommend a particular DOS graphics test that pshipkov mentioned? It would help me with some metrics beyond visual impression

search online for Phil' benchmark suite or something like that.

you can do it for the fun of it, but you wont see much performance difference, if any.
your computer is hard cpu limited.
also, good chance the card is set by default to the optimal config.

pshipkov wrote on 2022-10-03, 00:07:

search online for Phil' benchmark suite or something like that. […]
Show full quote

search online for Phil' benchmark suite or something like that.

you can do it for the fun of it, but you wont see much performance difference, if any.
your computer is hard cpu limited.
also, good chance the card is set by default to the optimal config.

Thank you both for your help, it's always nice to learn a thing or two, might try it anyway

@pshipkov
Thank you for the kind words. Definitely have more planned for the Batman's Revenge in the future!
Not sure about DC's licensing; it's possible, but the actual board name is "Premiere PCI."
It's mostly referred to as Batman, but not sure why. Same with other Intel codenames--Mainspike, Marryville, etc.
Not sure where they come from, if more internally-designated or externally so. Interesting nonetheless!

Batman's Revenge is really well-built. Not like lots of other Socket 4 boards that are just Socket 3 boards w/ a Pentium grafted on.
It's stable, fast enough, and pretty easy to use. Definitely not the worst, but like you said, very rigid and not ideal.
@AnonymousCoward hit the nail on the head. Not even the DX4-100 existed on market at the time.
DX2-66 is certainly no match. P66 is arguably faster/superior to the much later P75 as well. Pretty impressive stuff for the time.

Ran some 3D studio and got the following results:

• [66Mhz]: 124 seconds
• [60Mhz]: 137 seconds

Pretty solid clock-for-clock, I'd say. The Pentium FPU is truly a beast.
Integer performance is also cream of the crop.
I can re-run on an SiS50x board from later and we can do some comparisons.
Not quite 5x86-160 level, but certainly better than what I'd guess a P75 would do.

FDIV bug is an interesting one. I think @BitWrangler hit the nail on the head there. For scientific apps it was a problem.
I'd imagine most major research institutions and universities were quite displeased having to send in their CPUs for replacement.

Great information regarding the Trident boards--thank you very much for that!
Always wondered about the differences. I am curious in the 9685.
I can attest the 9880 is very good. It's not the fastest, nor the most special, but it's decent for 1999 3D graphics.
The 9440AGI/-1, 8900CL/D, and 9685 seem to make up affordable, easy to find good all-arounders for 2D graphics. Nice cards, honestly.
@maxtherabbit hit the nail on the head; zero wait and FIFO are the key to it all.
----
@BitWrangler
Well said with respect to the first gen of P5s. Definitely interesting chips and really solid performance for what they are.
I think performance w/r/t MP3s, websurfing, multitasking, solid Win95 experiences are all great examples of why the future was P5.
There's a lot of charm in this platform. It's a shame that S4 died quickly; S5 is great and all, but S4 feels like a hybrid between a 486 and 586 in way that just makes you smile.

I like your take on the 6x86 as well. A windows accelerator is a great way to put it; these things were beasts at 16-bit code and truly great for ALU-intensive tasks. Any FPU deficit as it came to gaming really would've been overcome by a Voodoo 1/V1000/etc. anyway, and true power users weren't the key market of Cyrix (from what I can tell). I do think that the PR thing was massive bs and that it was just all around deceptive, terrible business practices. It's frustrating, looking back at it. But it's also ingenious.

----
@AnonymousCoward
Agreed on all fronts. I think the P5 was an unparalleled beast; great for scientific apps and really interesting in terms of flexibility for professionals. Really interesting too in that it in many ways fit the 486 mold of things; the boards were not too different from their contemporaries and the feature set was not so functionally different that it'd confuse a new user. I think that's pretty appealing, all things considered.

I think Intel's feats were really only paralleled like you said by the Cyrix 6x86-P200+. I think the K6-233 was equally awesome but its thunder was stolen by the PII's release. It's wild to think that the Cyrix 6x86-P200+ was the first CPU to really "dent" Intel's stronghold on having the fastest chips in the early Pentium days. The unfortunate thing about the P200+ is that it feels like it never really had any good platforms that were mature enough to handle it well. 75MHz FSB was pretty hard to handle on a lot of consumer boards. Thankfully there were enough good ones out there, but it seems like having good mainstream boards capable of such took longer than the time for which the 6x86-P200+ was available. It seems like a relatively "rare" CPU these days, for whatever that's worth. They're not hard to find, per se, but it seems that not too many were made and really only those who know why they matter want them, otherwise, I see 6x86L-PR200s much more often.

320x240 at high color seems weird for the 9000i, but it makes sense given the cheapness of the card. Kinda wild, honestly. What's even the use of high color at such a resolution?

----
@andre_6
The PhilsComputerLab benchmark suite is excellent, as noted by pshipkov. It's very easy to use.
I'd recommend running options 2), 5), 6), a), b), c), and whatever else suits your interests.
Those are what we tend to use the most in this thread.

I'm curious to see your results! Please do share them, if you don't mind.
These past few posts have piqued my interest a bit in Trident hardware.

WJG, thank you for testing Diamond's v2.09 BIOS on your Trio64 -P card and confirming that you also experience the soft-reset problem. Could you remind me what motherboard you tested it on and have you tried it on an Asus VL/I-486SV2GX4? This is the motherboard I tried it on and experience the soft-reset problem, however pshipkov tried it on the same board but did not experience the soft-reset issue. I find this most curious. However, using the Number9 BIOS definately resolves the soft-reset issue and supports the VESA modes of the Diamond 2.09 BIOS?