Dorunkāku wrote on 2024-09-24, 13:26:

The Cypress CYM74BM550 looks like this:

CYM74B550
Async SRAM, not pipelined burst = will actually slow computer down 😮

rasz_pl wrote on 2024-11-08, 02:38:

Dorunkāku wrote on 2024-09-24, 13:26:

The Cypress CYM74BM550 looks like this:

CYM74B550
Async SRAM, not pipelined burst = will actually slow computer down 😮

Slow it down more than having nothing but the L1 in the CPU? 😉

For what it's worth, the MiTAC board in my Compaq has room for a jumper block to set which type of cache is installed and the factory has hard wired it to Async.

Carcenomy wrote on 2023-08-08, 06:00:

Long time reader, first time poster. […]
Show full quote

Long time reader, first time poster.

I have the same exact model,Compaq Presario 7222.
The things that puzzles me the most is 1- is there a way to change boot sequence?the whole Bios thing setting "device setup" on This model looks odd to me and different than anything I have seen.
2-how did you install your voodoo card? I tried , but windows doesn't like the card saying that it's disabled by hardware?! Gives me a protection error when I try to enable the driver in device manager. Not sure if there is a physical jumper i need to disable or if I need to do something in Bios
Any help is appreciated 👏

fadi.alsaidi@gmail.c wrote on 2025-01-21, 22:10:

Long time reader, first time poster. […]
Show full quote

Long time reader, first time poster.

I have the same exact model,Compaq Presario 7222.
The things that puzzles me the most is 1- is there a way to change boot sequence?the whole Bios thing setting "device setup" on This model looks odd to me and different than anything I have seen.
2-how did you install your voodoo card? I tried , but windows doesn't like the card saying that it's disabled by hardware?! Gives me a protection error when I try to enable the driver in device manager. Not sure if there is a physical jumper i need to disable or if I need to do something in Bios
Any help is appreciated 👏

Hey man, happy to help!
1. No control of boot sequence - it's always Floppy > Hard Disk. The only option to adjust that per se is the ability to disable floppy booting from Security Management in Computer Setup. However, if your hard disk has been formatted you may not have the setup partition so will need to use the original boot floppies (images are still around online for this if not let me know I'll put them up!) to either set up the partition again or let you make adjustments.
2. Plonked it in, no adjustments needed. What model of Voodoo are you using? I do know specifically the Orchid Righteous 3D is picky about drivers.

izuokan wrote on 2023-12-19, 18:04:

Hello! […]
Show full quote

Hello!

Thanks for your response.

I will go for an overdrive.

Have you tried this one? 237716-001 with mitsubishi sticker.

Greetings from Chile

Hey Izuokan, a development!

Every so often I go for a fossick on eBay just in case a unicorn pops up in the wild. Turns out, two did! I've grabbed one, the gent has one more... https://www.ebay.com/itm/277648501995

The IDT 7MPV6214. It's a 256K async module, nothing super exciting, but it is the right part for the job - by default, there's a jumper block on the PL5600G mainboard for selecting cache type and it's hardwired from factory in Async mode. So that module, will drop directly into a Presario 7200 with the MiTAC PL5600G and just work in theory.
And looks like modifying it to become a 512K async module would be relatively trivial - the pads for the additional 256K and decoupling caps are all there ready to go and it has enough tag and dirty capacity out of the box for 512K. Neat!

Once I've got mine in hand I'll do some quick checks to confirm it's all happy days, then try it in my machine. If all is well, I may go down the route of reverse engineering it so any other poor schmuck with an OPTi Viper who has no cache can get themselves out of a pickle without trying to find esoteric old Cypress or IDT cache modules.

Carcenomy wrote on 2025-01-15, 11:27:

rasz_pl wrote on 2024-11-08, 02:38:

CYM74B550
Async SRAM, not pipelined burst = will actually slow computer down 😮

Slow it down more than having nothing but the L1 in the CPU? 😉

Now that you got yourself one it will be easy to verify this theory. Async slows down every chipset that knows how to handle EDO properly and can run EDO at tight timings aka Intel chipsets https://dependency-injection.com/intel-430fx- … riton-l2-cache/ 2% slower in Doom but 8% faster in Quake
With a fail chipset that barely supports EDO like Sis 501 situation is much different https://dependency-injection.com/2mb-cache-benchmarks/ But then again sis 501 with 2MB of Async cache is still 10% slower than 430FX with no L2 at all and just EDO ram 😀
Im very interested in seeing Viper results, but judging by this https://dependency-injection.com/early-pentium-chipsets/ Im afraid its even worse than SIS 501, in Quake fitted with 256KB async its 20% slower than 430FX with no cache at all and good 40% slower when 430FX has proper cache.

rasz_pl wrote on 2026-02-13, 14:28:

Now that you got yourself one it will be easy to verify this theory. Async slows down every chipset that knows how to handle EDO […]
Show full quote

Carcenomy wrote on 2025-01-15, 11:27:

rasz_pl wrote on 2024-11-08, 02:38:

CYM74B550
Async SRAM, not pipelined burst = will actually slow computer down 😮

Slow it down more than having nothing but the L1 in the CPU? 😉

Now that you got yourself one it will be easy to verify this theory. Async slows down every chipset that knows how to handle EDO properly and can run EDO at tight timings aka Intel chipsets https://dependency-injection.com/intel-430fx- … riton-l2-cache/ 2% slower in Doom but 8% faster in Quake
With a fail chipset that barely supports EDO like Sis 501 situation is much different https://dependency-injection.com/2mb-cache-benchmarks/ But then again sis 501 with 2MB of Async cache is still 10% slower than 430FX with no L2 at all and just EDO ram 😀
Im very interested in seeing Viper results, but judging by this https://dependency-injection.com/early-pentium-chipsets/ Im afraid its even worse than SIS 501, in Quake fitted with 256KB async its 20% slower than 430FX with no cache at all and good 40% slower when 430FX has proper cache.

Well my friend, can only but try! I'll report back with some benches once it arrives, will compare it against my other similarly configured machine - a dual 200MMX 430HX. The HX should maul it, the question is by how much 😂

rasz_pl wrote on 2026-02-13, 14:28:

Now that you got yourself one it will be easy to verify this theory. Async slows down every chipset that knows how to handle EDO […]
Show full quote

Carcenomy wrote on 2025-01-15, 11:27:

rasz_pl wrote on 2024-11-08, 02:38:

CYM74B550
Async SRAM, not pipelined burst = will actually slow computer down 😮

Slow it down more than having nothing but the L1 in the CPU? 😉

Now that you got yourself one it will be easy to verify this theory. Async slows down every chipset that knows how to handle EDO properly and can run EDO at tight timings aka Intel chipsets https://dependency-injection.com/intel-430fx- … riton-l2-cache/ 2% slower in Doom but 8% faster in Quake
With a fail chipset that barely supports EDO like Sis 501 situation is much different https://dependency-injection.com/2mb-cache-benchmarks/ But then again sis 501 with 2MB of Async cache is still 10% slower than 430FX with no L2 at all and just EDO ram 😀
Im very interested in seeing Viper results, but judging by this https://dependency-injection.com/early-pentium-chipsets/ Im afraid its even worse than SIS 501, in Quake fitted with 256KB async its 20% slower than 430FX with no cache at all and good 40% slower when 430FX has proper cache.

I can give some actual stats now!
Without the cache fitted, Speedsys reported 352.48MB/s L1 and 82.33MB/s memory throughput, Quake achieved 28.6FPS.
With 256K async fitted, Speedsys reported 352.70MB/s L1, 119.10MB/s L2 and 81.44MB/s memory throughput, Quake achieved 30.3FPS

None of these numbers are going to break world records but the system certainly feels a bit more spritely in practical use. The biggest limiting factor now is the integrated S3 Trio64V+, its memory bandwidth is ~8.5MB/s which is about 2/3 what a retail Trio64V+ can achieve which does explain why it always felt so incredibly crippled in high res DOS modes.

Carcenomy wrote on 2026-03-03, 20:37:

I can give some actual stats now!
Without the cache fitted, Speedsys reported 352.48MB/s L1 and 82.33MB/s memory throughput, Quake achieved 28.6FPS.
With 256K async fitted, Speedsys reported 352.70MB/s L1, 119.10MB/s L2 and 81.44MB/s memory throughput, Quake achieved 30.3FPS

On one hand great that it worked and actually made computer faster. On the other it means OPTi Viper is one of those chipsets with bad EDO implementation 🙁
30fps is below Intel chipset + Pentium 133MHz and no L2 cache at all score.

Carcenomy wrote on 2026-03-03, 20:37:

The biggest limiting factor now is the integrated S3 Trio64V+

Trio64V+ is great Pentium era graphics chip

Carcenomy wrote on 2026-03-03, 20:37:

its memory bandwidth is ~8.5MB/s

Its (internal) memory bandwidth would be somewhere closer to 40-60MHz * 8 bytes = ~300MB/s
Where do you get the 8.5MB/s? Vidspeed? That is _very_ low, lower than low end VLB cards. What happens when you plug another PCI VGA, does it take over or is there a conflict?

If its really due to low clock on the VGA chip then maybe MCLK.EXE from https://www.geocities.ws/liaor2/myutil/myutil.html https://www.geocities.ws/liaor2/myutil/mclk.html might be able to bump the clock.

rasz_pl wrote on 2026-03-04, 04:45:

On one hand great that it worked and actually made computer faster. On the other it means OPTi Viper is one of those chipsets wi […]
Show full quote

Carcenomy wrote on 2026-03-03, 20:37:

I can give some actual stats now!
Without the cache fitted, Speedsys reported 352.48MB/s L1 and 82.33MB/s memory throughput, Quake achieved 28.6FPS.
With 256K async fitted, Speedsys reported 352.70MB/s L1, 119.10MB/s L2 and 81.44MB/s memory throughput, Quake achieved 30.3FPS

On one hand great that it worked and actually made computer faster. On the other it means OPTi Viper is one of those chipsets with bad EDO implementation 🙁
30fps is below Intel chipset + Pentium 133MHz and no L2 cache at all score.

Carcenomy wrote on 2026-03-03, 20:37:

The biggest limiting factor now is the integrated S3 Trio64V+

Trio64V+ is great Pentium era graphics chip

Carcenomy wrote on 2026-03-03, 20:37:

its memory bandwidth is ~8.5MB/s

Its (internal) memory bandwidth would be somewhere closer to 40-60MHz * 8 bytes = ~300MB/s
Where do you get the 8.5MB/s? Vidspeed? That is _very_ low, lower than low end VLB cards. What happens when you plug another PCI VGA, does it take over or is there a conflict?

If its really due to low clock on the VGA chip then maybe MCLK.EXE from https://www.geocities.ws/liaor2/myutil/myutil.html https://www.geocities.ws/liaor2/myutil/mclk.html might be able to bump the clock.

Oh we're under no illusions it's a good chipset but it wouldn't be the same machine if I yanked the board out and found another Compaq LPX to stuff in it, so gains are gains we'll take 'em! 😀

Sysspeed is reporting the 8.5MB/s. With a discrete Trio64V+ its getting 13.5MB/s. For now I've disabled the onboard Trio64V+ and installed a discrete Trio64V2/DX which has woken up some high res DOS titles a bit.

Still far from perfect but it's progress. Would love to get the integrated Trio running a bit nicer but no software tools seem to be able to affect it and after looking at the VBIOS best I can, there's some shenanigans going on inside the Compaq BIOS that make changes difficult. I can see where the bits can be adjusted for the Trio and compared against a retail BIOS, made some minor changes based on the differences... instant no boot, no matter how tiny or subtle the change. I suspect that'll be a dead end unless by some miracle I came by the Compaq tools from the mid 90s or talked to a Mitac engineer from the era who worked on this project. I feel like there's a higher chance of Satan making a downpayment on a snowmobile before either of those things happen.

MCLK.EXE is innefective?

rasz_pl wrote on 2026-03-05, 05:17:

MCLK.EXE is innefective?

Unfortunately, yeah it doesn't help.

I'm guessing something to do with how Mitac implemented the Trio on the board. I've not put too much thought into it recently, it seems pretty happy with the V2 for now. But will have a peek at it again another day, unless someone else has some great suggestions!

Can you upload your systems BIOS?
The fix could be as easy as copying and pasting a S3 Trio64V+ BIOS over the first 32KB of your systems BIOS.
Also if you don't have the 2 optinal video memory chips installed the Trio64 is crippled by a 32-bit memory bus.

Dorunkāku wrote on 2026-03-18, 10:49:

Can you upload your systems BIOS?
The fix could be as easy as copying and pasting a S3 Trio64V+ BIOS over the first 32KB of your systems BIOS.
Also if you don't have the 2 optinal video memory chips installed the Trio64 is crippled by a 32-bit memory bus.

Yeah it had the extra chips fitted. I've swapped them onto the V2 in the meantime.

Stock BIOS attached. Take a peek, it's a bit of a messy thing.
I have tried injecting the full retail 64V+ BIOS into that 32K, that prevents the machine starting.
I also tried just editing some of the initialization bits to less conservative settings, that also prevents the machine starting.
But I'm an absolute novice when it comes to BIOS editing, especially on esoterica like this thing.

I saw this too late...

Carcenomy wrote on 2026-02-13, 10:17:

Hey Izuokan, a development! […]
Show full quote

izuokan wrote on 2023-12-19, 18:04:

Hello! […]
Show full quote

Hello!

Thanks for your response.

I will go for an overdrive.

Have you tried this one? 237716-001 with mitsubishi sticker.

Greetings from Chile

Hey Izuokan, a development!

Every so often I go for a fossick on eBay just in case a unicorn pops up in the wild. Turns out, two did! I've grabbed one, the gent has one more... https://www.ebay.com/itm/277648501995

The IDT 7MPV6214. It's a 256K async module, nothing super exciting, but it is the right part for the job - by default, there's a jumper block on the PL5600G mainboard for selecting cache type and it's hardwired from factory in Async mode. So that module, will drop directly into a Presario 7200 with the MiTAC PL5600G and just work in theory.
And looks like modifying it to become a 512K async module would be relatively trivial - the pads for the additional 256K and decoupling caps are all there ready to go and it has enough tag and dirty capacity out of the box for 512K. Neat!

Once I've got mine in hand I'll do some quick checks to confirm it's all happy days, then try it in my machine. If all is well, I may go down the route of reverse engineering it so any other poor schmuck with an OPTi Viper who has no cache can get themselves out of a pickle without trying to find esoteric old Cypress or IDT cache modules.

If your board supports pipeline burst sync cache you could go for the 256KB Module offered on ebay as 'AC41669 CACHE Module'. That should even be faster than async cache.

It is mentioned in this document (page 92 or 100th page) as being Opti Viper compatible.