Pictures of the cards...

Gah! Just realised I posted this in general not the video specific forum. Can a mod please move it? Thanks. 😀

Not all Trident cards were slow. The 8900D was one of the faster cards, I think it used memory interleaving. And the CL GD542x series was low-end.

There are big differences between Trident cards. Some are really slow, but there are a few as fast as the best ones - like Tseng ET4000. I also have CL-5429 but I'd guess much newer and it is fast.

This is my card:
http://i39.tinypic.com/1687c7a.jpg

http://i39.tinypic.com/rlw8yf.jpg

No Trident card used memory interleaving. This came later with 2MB cards - ET4000/w32i, some S3 805i and newer CL5434, S3 864, ... but this only shows in Windows acceleration mode.

Trident cards were not actually that bad, both of your cards are pretty fast for isa but of course a VLB card will be much faster. 8900D had a FIFO buffer, the cirrus is a later and better card with 24 bit color support and acceleration.
Benchmarks usually don't really matter with such slow cpus, what matters is performance in games such as DOOM.

Moved.

Stiletto wrote:

Moved.

Thanks 😀

dondiego wrote:

Trident cards were not actually that bad, both of your cards are pretty fast for isa but of course a VLB card will be much faster. 8900D had a FIFO buffer, the cirrus is a later and better card with 24 bit color support and acceleration.
Benchmarks usually don't really matter with such slow cpus, what matters is performance in games such as DOOM.

My 8900D supports 800x600 in 15/16 bit, and 640x480 24bit.

Trident 8900D is actually a good ISA card, they just weren't really recognized because they were still being shipped with PCs even when PCI was around. You can't ask for PCI or VLB performance out of the 16-bit ISA bus, it's a bottleneck with a dx2-66.

The issues with the Trident cards usually comes down to:
1) nearly every one I've seen has disgusting video output on an LCD. Vertical banding and distortion makes seeing far off enemies very hard in games like DOOM.
2) Tragically big memory latency

What do you get on those cards if you run top bench? Does your bios have an ISA speed divider? If it does try reducing the divider, or overclocking to 40 bus. I've had luck with isa cards being able to handle a massive overclock from stock ISA spec, usually having to dial back to whatever keeps the Sound Blaster from getting garbled up. The 8900D can run massively out of spec and the performance increase from the reduced memory latency is massive, maybe the CL card will do well too. In my experience video memory needs to be reporting under 160 in top bench for a smooth experience in games like doom, this is where VLB really shines.

Edit: Ahh just notice you ran it on a pentium, I thought those scores were from a DX2-66. What are your bench results on the 486?

I'm not sure if you should be expecting more out of the cirrus logic card, I've never gotten my hands on one.

I think your problem is 80ns RAM on the CL vs 70ns on the Trident. On VLB the CLs usually came with at least 70, the good ones were 60.

edit: Actually that's better speed RAM than I've seen on most Tridents too, seen 100ns even. 80 common.

BitWrangler wrote:

I think your problem is 80ns RAM on the CL vs 70ns on the Trident. On VLB the CLs usually came with at least 70, the good ones were 60.

Thanks, I didn't even think of that. I'll sheepishly admit I was so chipset focused I didn't look at memory speed.

I forgot the 8900D had 24 bit color support (those usually came with 1 mb of ram), the 8900c was the one with only 16 bit and was very slow. But still that cirrus was a better card.
Your problem is the isa bus not ram, you could get some performance boost overclocking the bus from 8 to even 12 mhz but the hdd controller might fail over 10 mhz (it's already a 25% overclock) and you could get data corruption. I personally haven't even tried beyond 10 mhz on an isa system but on a pci one my trident 9000i worked @12.

BeginnerGuy wrote:

You can't ask for PCI or VLB performance out of the 16-bit ISA bus, it's a bottleneck with a dx2-66.

Yep, I know that and I'm not expecting amazing performance. I just thought, hey here's a (supposedly) rubbish 8900, maybe it's worth getting another card to see what a (supposedly) better one is like. 😀 In the early 90s I used to run an ET4000 - maybe I should have stuck with one of those after all!

BeginnerGuy wrote:

The issues with the Trident cards usually comes down to:
1) nearly every one I've seen has disgusting video output on an LCD. Vertical banding and distortion makes seeing far off enemies very hard in games like DOOM.

Thanks, you just reminded me of another reason I thought of changing. I switched to a CRT for that exact reason (and actually prefer it on this old hardware anyway). Oh, and the Cirrus also has brighter output than the Trident. This was especially noticeable on the Doom & Quake tests where the Trident was sometimes too dark. Not that I intend to be playing those on the 486 though anyway. 😀

BeginnerGuy wrote:

What do you get on those cards if you run top bench? Does your bios have an ISA speed divider? If it does try reducing the divider, or overclocking to 40 bus.

Yes to the ISA clock divider, no to overclocking to 40.
Topbench video memory speed on the DX2-66 for the Trident is around the 190-195 mark, and for the Cirrus around 230. Changing the ISA speed from CLK/4 (8.25MHz) to CLK/3 (11MHz) brought the Cirrus down to ~215, but had negligible impact on the Trident.

BeginnerGuy wrote:

The 8900D can run massively out of spec and the performance increase from the reduced memory latency is massive, maybe the CL card will do well too. In my experience video memory needs to be reporting under 160 in top bench for a smooth experience in games like doom, this is where VLB really shines.

Edit: Ahh just notice you ran it on a pentium, I thought those scores were from a DX2-66. What are your bench results on the 486?

Ok, I played around some more with both cards in the DX2-66, with different clock divisors of 4, 3 and even 2.5 (13.2MHz). The Trident was mostly unaffected - far greater impact on the benchmarks was had by reducing my memory wait states. This also got the Topbench video memory speed down to ~170. (I need to reassess system stability now though because last time I reduced wait states I had issues).

On the other hand changing the ISA speed did affect the Cirrus card.

DX2-66 numbers for: Superscape v1.0c | Chris's 3D benchmark | PC Player | Doom min detail | Quake timedemo

Trident CLK/4 = 30.3 | 19.6 | 7.4 | 49.8 | 5.1
Trident CLK/3 = 30.7 | 20.4 | 7.5 | 50.1 | 5.2
Cirrus CLK/4 = 27.7 | 17.9 | 7.2 | 47.6 | 5.0
Cirrus CLK/3 = 28.7 | 19.0 | xxx | 48.9 | 5.2 (forgot to test PC Player)

Still not a huge improvement with the Cirrus card, and it is mostly behind the Trident across the board regardless of the clock divisor setting. I haven't tested it with the reduced wait states (above numbers are with the same settings btw so they are comparable), but I'm sure that wouldn't change the result either.

Cheers,
Stu

dondiego wrote:

Your problem is the isa bus not ram, you could get some performance boost overclocking the bus from 8 to even 12 mhz but the hdd controller might fail over 10 mhz (it's already a 25% overclock) and you could get data corruption. I personally haven't even tried beyond 10 mhz on an isa system but on a pci one my trident 9000i worked @12.

Yep, unfortunately not it seems as per my post above.
Re the hdd controller though - I've previously experimented with CLK/4 (default/correct), CLK/3 and CLK/2.5. The system appeared to be stable without any problems on each setting. However, I didn't notice any benefit either so I went back to the safer 8.25MHz.

Memory (and cache) wait states have a greater impact though so I'll be doing some more testing there now.

Back in the day, fastest I got working was 40/3 after swapping cards around... but soundcards tend not to like being over 10Mhz in general... I had one system running at 12Mhz with sound. It was hard to go faster than 40/3 because it was a 2.5Mhz jump to the next speeds, which didn't help you weed out cards that were flaking at a mhz more at a time. I thought 10Mhz was fairly safe for most things though, older cards didn't like it, say pre-92ish... typically ones that were full height rather than level with top of bracket height or shorter. Maybe through hole components rather than surface mounted.

Seems like that setting is ignored on your board then, if i remember right going from 8 to 12 mhz on a k6-2 i got about a 40% performance increase in Doom with my 9000i.

This is an old board that only supports up to a 33MHz 486DX. My guess would be the max ISA bus speed is limited to 8.33MHz. CLK/4 should give 8.25, thus CLK/3 lets me squeeze out a tiny bit more (negligible).
A PCI machine probably lets you get more.

Recently I've found this card (mixed access time memory):

Did some tests in Quake on PIII 800EB & ASUS P3B-F machine under Windows 98 SE:
320x240 - 34.5 fps
360x480 - 15.3 fps

According to VGA museum benchies my card only slightly missed ISA bus cap in 320x240 mode. Maybe that's 80ns memory or tests under windows environment to blame. Anyway, Quake is super smooth in 320x240 mode, but 320x200 and many other modes are plagued with vertical columns on my LCD monitor.

InbetweenDays wrote:

The only good thing I can say about it is that the Trident gives me a monitor identification problem (starts in mono) on reboot whereas the 5429 doesn't.

There's a fix for that starting in mono thing:

Get a M/F VGA "gender changer" like this
https://www.amazon.com/BoNaYuanDa-Female-Coup … /qid=1520360154
(it doesn't actually change gender -- one site is M and the other is F)
Or you can use an extension cable.

Remove pin 12 on the M side of the gender changer.

Attach the gender changer to the Trident's VGA port and plug the monitor
into the gender changer.

The fix has worked for me with old Trident and Oak VGA cards -- no more unwanted starting in mono.
My brother found this fix, but can't remember where he found it.

Here's a description of the VGA pinout:
https://allpinouts.org/pinouts/connectors/com … deo/vga-15-pin/
Removing pin 12 forces color.