Currently the scope of the thread is on VLI SV2GX4, POD100, 1024K L2 and stability with maximum amount of RAM when L1/L2 caches are in WB mode.
The question is - with 64Mb RAM and DRAM WRITE WS set to 0 - is the instability coming from the 60ns RAM modules, or something inherent to mobo+POD.

Video cards are not the problem. Tried 5-10 ones. Makes no difference.
Tried LFB, paged, diff quake 1 settings. Makes no difference.
Disabled L2 cache entirely. No difference.
It is RAM related entirely.

One people told me, he has difficult with 64MB RAM and Trio64V+ VLB.

Their system run stable (Gigabyte with SIS471 ), if he use only 16MB RAM.

I think and can say why: Adress decoder from STB Powergraph P64V (same as my relica ) respond at 0-8MB and it is mirrored at 128/256/384MB (and more memory location). Possible make this address decoder a problem, if you use huge 64MB RAM. I have here a early prototype with better address decoder (It respond at 0-8MB and 2GB )

The issue is clearly not video adapter related.
If DRAM WRITE WS is 1 - 64Mb are just fine.
Need to comb more carefully through the stack of memory sticks and see if there is a magical combination.

Discovered that dropping Hardware Acceleration down resolves the screen corruption issues I was seeing in Windows DirectX games. Not sure what kind of impact this would have on those games or with Windows in general? Makes me wonder if there is a better driver out there.

At minimum it disables HW accelerated GDI, possible other things as well.
Don't know from the top of my head the extent of HW coverage in these old adapters and their drivers.

An interesting benchmark built into Terminal Velocity.

pshipkov wrote on 2021-11-11, 16:18:

Currently the scope of the thread is on VLI SV2GX4, POD100, 1024K L2 and stability with maximum amount of RAM when L1/L2 caches are in WB mode.
The question is - with 64Mb RAM and DRAM WRITE WS set to 0 - is the instability coming from the 60ns RAM modules, or something inherent to mobo+POD.

To confirm, if we set the L2 cache to WT mode all is good? Or if we swap the POD for X5-160, all is fine?

@Feipoa
This is strictly POD + L2 WB mode related.
If jumper sets L2 in WT mode all is good.
This issue is not present with 486 cpus (including X5-160).

@chadti99

6 digits after the decimal point !
This is what i call precision.
It can bring a whole new dimension to the retro perf hair splitting.

I discovered by mistake once that the shortest possible IDE ribbon to disks gives better signal gain for these old things.
Custon ribbon kit was what i used when accidently breaking a cable plug.
Though I wouldn't have it with the disk right on the motherboard; short as possible to limit slck is best to acheive maximum throughput
usually improving benchmarks.

pshipkov wrote on 2021-11-11, 22:34:

@Feipoa This is strictly POD + L2 WB mode related. If jumper sets L2 in WT mode all is good. This issue is not present with 486 […]
Show full quote

@Feipoa
This is strictly POD + L2 WB mode related.
If jumper sets L2 in WT mode all is good.
This issue is not present with 486 cpus (including X5-160).

Thank you for this clarity. I had not tried this particular combination. Is it just the 1024K size that is the issue? 512K double-banked write-back is good?

pshipkov wrote on 2021-11-11, 22:41:

6 digits after the decimal point !
This is what i call precision.
It can bring a whole new dimension to the retro perf hair splitting.

OffTopic: Hej, what's about a hydrogen maser clock? It's accuracy is 15 digits after the decimal point.

pshipkov wrote on 2021-11-11, 22:41:

@chadti99 […]
Show full quote

@chadti99

6 digits after the decimal point !
This is what i call precision.
It can bring a whole new dimension to the retro perf hair splitting.

Haha indeed, I still want my missing frames though 🤣!

So what’s the standard app for testing 2d video performance in 95/98? Curious how some of my cards stack up and what impact adjusting the “Hardware Acceleration” option has.

It is not l2 cache related. At 40mhz base frequency it is well within its capabilities. Also, spent time moving chips around - no difference.
But swapping memory sticks has effect. Some are better, some worse.
Didn't try 512k, but I won't like the setup that way. Won't accept it as an option.

Latest arrival, Dos performance isn’t the best as others have reported but it’s still good. I’m sure requiring the “synchronized” setting in the bios versus “transparent” doesn’t help. Seeing the higher resolutions and color settings available in Windows is nice and it maintains a sharp picture up to 1280x1024, though not as sharp as a Voodoo3 for example, but that’s a 220MHz vs 300+MHz RAMDAC so understandable. Not that a Voodoo3 would function here 🤣. I should def try an mpeg movie for kicks as I think the 968 has mpeg acceleration features.

Wow, how much did that set you back in this day and age?

For MPEG, you will need the MPEG-1 add-on card, shown here: MPEG-1 decoder add-on board for [S3 Vision 968] Diamond Stealth 64 VRAM VLB

Are you willing to dump the EEPROM? As you can see from my image in the other link, my version is only 2.03.

feipoa wrote on 2021-11-15, 22:06:

Wow, how much did that set you back in this day and age?

For MPEG, you will need the MPEG-1 add-on card, shown here: MPEG-1 decoder add-on board for [S3 Vision 968] Diamond Stealth 64 VRAM VLB

Are you willing to dump the EEPROM? As you can see from my image in the other link, my version is only 2.03.

I dont mind discussing it in a PM but not as bad as a Mach64 VLB would of set me back 🤣. People are asking some insane prices for those.

I can dump the EEPROM for sure, thought I could use CPU-Z but it only spits out a text file which I’m sure someone smart enough could make work. Anyone know a good way to dump the EEPROM with the card in the pc?

Also on the driver version I’m on, 4.02.268, I’ve lost the ability to adjust the refresh rate beyond default or optimum, no in-between. I have my monitor driver installed to be sure. Any idea what causes this?

I've run into similar issues with the refresh rate options disappearing like that, but with the MVP-2000 MPEG drivers. There's a way to create your own monitor inf files to set your target refresh, however I suggest you try driver version *.325 and *.241 first. I'm pretty sure *.325 didn't have this issue.

I'll tell you what I paid if you tell me yours. I paid somewhere in the $18-24 range for the Diamond-based S3 968 w/2MB upgrade card and for the MVP-2000 decoder card, I $99. I'm sure I could have bargained it down, but when I've done this in the past, someone else went in and bought it.