A quick search lead me to an upgrade thread on these that mentioned the gate array chip only being able to work with fsb frequency of 16, 25 or 33 mhz, so that is likely your problem. Also, it sounded like these don't have a VRM for the CPU and just use 5 volts so use caution if that is the case, if you say that you are using a 5x86 it is likely running significantly overvolted unless it is a QFP chip on an interposer with a voltage regulator

As for the onboard graphics it doesn't seem like there is any way to speed that up, likely it is on the ISA bus which would be no surprise if it was meant to be paired with a slow SX cpu anyway

The WD90C22-26, and similar WD90C31/33 VGA controllers are on the local bus on laptops, so they're like a VLB card.

MPEG players will run slow because the CPU is doing the MPEG decoding. VGA cards couldn't do it until much later. There may be some external PCMCIA 16 bit MPEG decoder available such as the Canopus PowerWindow T64V.

The standard 256KB video memory may upgradable to 512KB/1MB. If there's no space on the motherboard it may be possible to stack the memory vertically (Ie. identical DRAMs soldered on top of the original). All pins except RAS for the upper row should be connected to all pins on the lower row. The upper row should share the same RAS, and connect to the second RAS line on the WD90C23.

If the WD90C23 has a changeable XTAL, you can try up to 45Mhz, or 50Mhz. Higher clock requires faster video DRAM.

There's some datasheets here for the WD90C22 and WD90C24, but not the 23. Which ever matches the number of pins closer is likely compatible. https://bitsavers.org/components/westernDigital/_dataSheets/

MikeSG wrote on 2025-11-29, 13:57:

The WD90C22-26, and similar WD90C31/33 VGA controllers are on the local bus on laptops, so they're like a VLB card. […]
Show full quote

The WD90C22-26, and similar WD90C31/33 VGA controllers are on the local bus on laptops, so they're like a VLB card.

MPEG players will run slow because the CPU is doing the MPEG decoding. VGA cards couldn't do it until much later. There may be some external PCMCIA 16 bit MPEG decoder available such as the Canopus PowerWindow T64V.

The standard 256KB video memory may upgradable to 512KB/1MB. If there's no space on the motherboard it may be possible to stack the memory vertically (Ie. identical DRAMs soldered on top of the original). All pins except RAS for the upper row should be connected to all pins on the lower row. The upper row should share the same RAS, and connect to the second RAS line on the WD90C23.

If the WD90C23 has a changeable XTAL, you can try up to 45Mhz, or 50Mhz. Higher clock requires faster video DRAM.

There's some datasheets here for the WD90C22 and WD90C24, but not the 23. Which ever matches the number of pins closer is likely compatible. https://bitsavers.org/components/westernDigital/_dataSheets/

Thanks for this - yes, I have looked at possibly stacking the onboard DRAM chips before - but the WD90c22 datasheet seems to indicate that I would need to change what is stored in some registers to have 4x 64k - 16-bit chips (the current board only has 2 and there are no blank pads to add new chips) the relevant section from the datasheet is attached.

I do have extra identical chips, but I wasn't sure I could just piggyback and solder them like you mention. Are you sure I can do that and then hook up the RAS line of the top row to the second RAS line on the WD90C23? (soldering-wise I could do it, but I just don't want to go through the hassle if it likely won't work). It would actually be amazing to get 512k onto this machine however, as there are several windows games that I could get to run that wouldn't otherwise run on 256K of video memory.

Regarding the XTAL / crystal oscillators - there are 3 on the system board that are hooked up to the WD90C23 (at least according to the manual):

- 25.175 - this is very likely to be for producing 640x480 resolution - VCLK0
- 28.322 - this is very likely for text mode / higher resolutions - VCLK1
- 44.900 - this seems like its used for other resolutions - VCLK2, though very small chance it's the MCLK signal?

It does say here (https://dosdays.co.uk/topics/Manufacturers/pa … al.php#Crystals) for this card that there should be a 36Mhz MCLK signal or a 42 Mhz Crystal for MCLK, and it's likely the 44.9 Mhz is used for a display mode - but I can't seem to work out what may be generating the MCLK signal on the Toshiba T4400. There is a component listed in the T4400SX maintenance manual (https://www.minuszerodegrees.net/manuals/Tosh … ce%20Manual.pdf named "variable frequency oscillator" and that is timed with a 24 Mhz crystal - maybe this is where MCLK comes from (maybe 24x2=48Mhz?).

In the manual it says the variable frequency oscillator chip (which is timed by a 24.00 Mhz crystal) feeds the large Super Intergration T9900 chip, so that could be generating a bunch of other signals. I don't think I can mess with the 24Mhz clock for it though, as it says that the same 24Mhz oscillator is used to time the floppy disk controller, though I do have a 25Mhz crystal that I could try... it is tempting, and it seems like it is possible to overclock a typical floppy disk controller searching on Vogons.

I had mentioned the playback of an MPEG file just as that was what I did last when I was benchmarking, but perhaps that was a poor example. I know the video bus is what is holding the system back, as in almost all benchmarks, I get pretty low scores for video throughput (wintune gives me 1.9 MP/s).

Do you think the piggybacking would work for the video RAM? and do you think I should give it a go at swapping the 24Mhz crystal to 25MHz to overclock the variable frequency oscillator?

bertrammatrix wrote on 2025-11-28, 17:13:

A quick search lead me to an upgrade thread on these that mentioned the gate array chip only being able to work with fsb frequency of 16, 25 or 33 mhz, so that is likely your problem. Also, it sounded like these don't have a VRM for the CPU and just use 5 volts so use caution if that is the case, if you say that you are using a 5x86 it is likely running significantly overvolted unless it is a QFP chip on an interposer with a voltage regulator

As for the onboard graphics it doesn't seem like there is any way to speed that up, likely it is on the ISA bus which would be no surprise if it was meant to be paired with a slow SX cpu anyway

Yep using a slim-profile drop in CPU upgrade from Kingston that has the onboard voltage regulator that drops the CPU voltage down to around 3.5V (if I remember correctly) . The post on overclocking the T4400 comes from https://www.euronet.nl/users/ernstoud/upgrade.html - and involves 1) changing the CPU crystal oscillator from 25Mhz to 33.333 Mhz and then removing the W2 soldered jumper.

What I found is that you can still swap the CPU's crystal oscillator to anything you want with the jumper remaining in (25Mhz 'mode'). Removing the jumper just seems to maintain system stability (I'm guessing through adding wait states or something similar) but at a huge cost to any speed resulting from overclocking. What I ended up with after testing various crystals with various combinations of the W2 jumper is leaving the W2 jumper in (as if it's configured for 25Mhz) but acutually using a 30 Mhz crystal. This seemed to give a stable system (with the 5x86 at 3X multiplier only).

At 4x multiplier, there is very little benefit for some reason (I think the CPU ends up being "too fast" for the rest of the system, like the system bus etc., and somehow increases instability as well). Benchmarks in DosBench do show the CPU is configured correctly, running at 133 and even running overclocked at 166 Mhz but performance is actually degraded or unstable in applications / complex benchmarks / windows. 3X at 90 Mhz seems to be the sweetspot for now with a 30Mhz crystal and the W2 jumper remaining in for '25mhz' mode in terms of performance and stability.

The fastest I got the computer to sort of work was at 40 Mhz with the W2 jumper removed (33 Mhz 'mode') and a 4x 5x86, benchmarks were all significantly higher (MIPS, memory speed and a bit faster on graphics) BUT the system could only manage to run dos games and Windows 3.11 apps - Windows 95 would BSOD if I tried to run Wintune 95 or other complex programs / games.

uscleo wrote on 2025-12-01, 07:24:

Thanks for this - yes, I have looked at possibly stacking the onboard DRAM chips before - but the WD90c22 datasheet seems to indicate that I would need to change what is stored in some registers to have 4x 64k - 16-bit chips (the current board only has 2 and there are no blank pads to add new chips) the relevant section from the datasheet is attached.

It looks like register PR1 is set via the BIOS. So the BIOS will need to count and display the RAM correctly... or alternatively write a DOS program to unlock PR1, set it, then test the value.

If your display is greyscale and fixed to 640x480, I'm not sure what more DRAM will add, unless programs physically won't open wih 256KB.

uscleo wrote on 2025-12-01, 07:24:

I do have extra identical chips, but I wasn't sure I could just piggyback and solder them like you mention. Are you sure I can do that and then hook up the RAS line of the top row to the second RAS line on the WD90C23? (soldering-wise I could do it, but I just don't want to go through the hassle if it likely won't work). It would actually be amazing to get 512k onto this machine however, as there are several windows games that I could get to run that wouldn't otherwise run on 256K of video memory.

After seeing the note about MA8, I think it should be connected to this. The RAS line on the top chips connected to MA8 on the WD90C22. It's not the lowest chance of working. Should be >50%.. but not sure if it'll improve anything.

uscleo wrote on 2025-12-01, 07:24:

Regarding the XTAL / crystal oscillators - there are 3 on the system board that are hooked up to the WD90C23 (at least according […]
Show full quote

Regarding the XTAL / crystal oscillators - there are 3 on the system board that are hooked up to the WD90C23 (at least according to the manual):

- 25.175 - this is very likely to be for producing 640x480 resolution - VCLK0
- 28.322 - this is very likely for text mode / higher resolutions - VCLK1
- 44.900 - this seems like its used for other resolutions - VCLK2, though very small chance it's the MCLK signal?

44.9MHz is the MCLK (memory clock) for <=100ns DRAMs. Later cards go up to 50MHz for <=80ns DRAMs. Not sure if this card supports more than 45Mhz.
25.175MHz is for 640 HSYNC (VCLK1) I believe and shouldn't be changed.
28.322MHz is for 720 HSYNC (VCLK0) I believe and isn't used.

uscleo wrote on 2025-12-01, 07:24:

I had mentioned the playback of an MPEG file just as that was what I did last when I was benchmarking, but perhaps that was a poor example. I know the video bus is what is holding the system back, as in almost all benchmarks, I get pretty low scores for video throughput (wintune gives me 1.9 MP/s).

Do you think the piggybacking would work for the video RAM? and do you think I should give it a go at swapping the 24Mhz crystal to 25MHz to overclock the variable frequency oscillator?

There's just no way to make MPEG decoding faster on a laptop except by an external video card. Later, more modern VGA cards can decode MPEG, and MPEG decoder cards can decode it... it's primarily a CPU bottleneck.

Images that can be displayed in the colour & resolution available, can't go faster except via a faster motherboard FSB, lower RAM wait states, faster video MCLK...