simon_e_hall wrote on 2026-03-24, 16:56:

AT Bus Clock: 14.318/2, PCLK2/3 ~ PCLK2/6, PCLK2/8, PCKLK2/10 and PCLK2/12 (Think this was here before though)

This sounds the same - but I swear it was called ISA Bus Clock, is there two?

Only asking because I played with overclocking the ISA bus because it helps Doom immensely but unfortunately with the stock bios it also makes the system seriously unstable (Doom completes benchmarks but Win3.11 will crash).

BinaryDemon wrote on 2026-03-25, 11:12:

simon_e_hall wrote on 2026-03-24, 16:56:

AT Bus Clock: 14.318/2, PCLK2/3 ~ PCLK2/6, PCLK2/8, PCKLK2/10 and PCLK2/12 (Think this was here before though)

This sounds the same - but I swear it was called ISA Bus Clock, is there two?

Only asking because I played with overclocking the ISA bus because it helps Doom immensely but unfortunately with the stock bios it also makes the system seriously unstable (Doom completes benchmarks but Win3.11 will crash).

Copy and pasted from a AMI BIOS manual:

AT BUS Clock: ISA Bus Clock timing selction. When Auto Configuration is Disabled, then 14.318/2, PCLK2/3.... ....is available on ISA Bus.

So assess, is the same thing AMI just using a different name.

simon_e_hall wrote on 2024-08-28, 16:06:

I have no idea WHY either, just does not make sense. Already tried something similar, had another system look at the CHS and tr […]
Show full quote

BinaryDemon wrote on 2024-08-28, 12:56:

simon_e_hall wrote on 2024-08-28, 11:21:

Been doing a bit of fault finding regarding my weird issue of no Ad Lib sound with CF cards, plugged in the original 2gb and that worked, so thought it maybe an issue with certain manufactures, so created a number of drives on 2gb all the way up to 8gb, and the issue only seems to happen to Transcend cards, which is the majority of CF cards I use! But again, cannot figure out why, interesting little issue though.

I still can't imagine WHY it would make a difference but maybe look beyond the bios AUTO values and see what Transend is using for stuff like Heads, Cylinders, Sectors. There is probably a DOS utility which could help if BIOS doesn't save them switching from auto to manual.

I have no idea WHY either, just does not make sense. Already tried something similar, had another system look at the CHS and transferred them across, but there is no difference between what auto detects and the other system detects. Thought maybe it was because I was transferring an image of the drive (VHD to CF) and maybe I had used a similar but wrong image, but that was not the case.

Even plugged a floppy controller into the ISA adapter and formatted the drive the traditional method but same issue.

Just going to chalk this one up to an oddity and not use Transend cards with it.

A while ago I was having issues with Transcend cards over 2gb stopping the OPL3 from working, sometimes not detected at all and sometimes detected, and not generating any sound. So, while I had the Pocket taken apart for better access to the BIOS chip, thought I would re-visit that issue and get the scope involved, turns out did not need the scope.

To get a baseline of the fault, I made a DOS booting 8GB Transcend card, with a few games that do Adlib and the Adlib jukebox to test, and same issue as before no sound, but as a last ditch effort I changed the PIO mode in the BIOS to 1 (from the deault 4) booted and the damn thing has sound, tried a few different Transcend cards at with PIO 1 and working, some still report they are in PIO 4 on the boot screen, but the BIOS still shows them at PIO 1.

So some sort of progress with what is causing the issue, but I have stopped using Transcend cards in the Pocket a while ago and have been using Lexar cards, which do not have this issue.

(Or I have broken something inside when taking it out the case!)

BinaryDemon wrote on 2026-03-25, 11:12:

simon_e_hall wrote on 2026-03-24, 16:56:

AT Bus Clock: 14.318/2, PCLK2/3 ~ PCLK2/6, PCLK2/8, PCKLK2/10 and PCLK2/12 (Think this was here before though)

This sounds the same - but I swear it was called ISA Bus Clock, is there two?

Only asking because I played with overclocking the ISA bus because it helps Doom immensely but unfortunately with the stock bios it also makes the system seriously unstable (Doom completes benchmarks but Win3.11 will crash).

The reason Windows is crashing is that there are some timing issues with the ALi SoC which was described in more detail on supplemental documentation alongside the datasheets, but has since been lost. Doing some investigating on my own, it is the PIT implementation not running consistently with different AT bus speeds over long periods. This can skew the system clock as well!

My personal opinion is that it was expected that programmers handle the AT bus speed changes in the software written bespoke for the SoC, instead of relying on behavior not unlike consumer board BIOSes and chipsets doing clock "boosts" automatically based on behavior seen on the bus.

For those who may have picked up on this, chipsets made for budget 386SX machines, such as the SCAMP chipsets, had this automatic overclock feature as a selling point for OEMs.

Since this 386 SoC was never meant to be in a "retail consumer software" situation, it had no need for parlor tricks like this. The downside is that you'll have to do that work yourself in your code. Perhaps this was to save power, as the datasheet discusses the benefits of its many power saving features (to a ridiculous degree imho).

Quick update on my Pocket 386 is that it's dead on the bench. I'll need to go and troubleshoot what's wrong but I'd expect it's something we've covered on the forums here. I've also got a spare board to test against so I'll report back once I've got it done. Who knows, maybe I'll finally make another video and do an accompanying writeup on it.

Before it went all untimely to the workbench on me I'd been using it at least once a month for gaming and as a portable tag-along for any i386 related shennanigans. I really like Win95 on there with a clock slowing tool for running late DOS/ Windows 3.11 and early Win95 games.

The boot time is fine; I recall it being a half minute. If anyone thinks that's slow they should go grab a C64 with a floppy drive!

Oh, and if anyone wants the ISA -16 breakout and doesn't want to CAD it up themselves let me know. With enough prodding I could be convinced to grab a board and make at least one more of them; or send it out kitted up. They're fiddly to manually modify the connectors to all fit.

Wow, I'm sorry to hear such news. I've always admired and learned from your posts. For instance, I did not know one could get a 30-sec+ boot into Win95. Mine took minutes and while Word 97 (tho its install was glacial) ran fairly well, any other program took, again, minutes to load, even to open a directory to browse its files (i.e. HanDBase). So I've been using Win3.11, with polpo's gpio wifi adapter, speedy, but internet things have been the usual challenge, tho I can ftp to my mothership.

I found a fairly fast cf for the P386, but I'm nowhere near 30-sec+.

Fingers crossed that you reanimate the device.

Inhibit wrote on 2026-04-23, 16:37:

Quick update on my Pocket 386 is that it's dead on the bench. I'll need to go and troubleshoot what's wrong but I'd expect it's […]
Show full quote

Quick update on my Pocket 386 is that it's dead on the bench. I'll need to go and troubleshoot what's wrong but I'd expect it's something we've covered on the forums here. I've also got a spare board to test against so I'll report back once I've got it done. Who knows, maybe I'll finally make another video and do an accompanying writeup on it.

Before it went all untimely to the workbench on me I'd been using it at least once a month for gaming and as a portable tag-along for any i386 related shennanigans. I really like Win95 on there with a clock slowing tool for running late DOS/ Windows 3.11 and early Win95 games.

The boot time is fine; I recall it being a half minute. If anyone thinks that's slow they should go grab a C64 with a floppy drive!

Oh, and if anyone wants the ISA -16 breakout and doesn't want to CAD it up themselves let me know. With enough prodding I could be convinced to grab a board and make at least one more of them; or send it out kitted up. They're fiddly to manually modify the connectors to all fit.

Oh no!
Hope its an easy fix, just in case this helps:
The one I bought over a year ago now has failed quite a few times (and also another that I bought to replace it did this the first time I plugged it in to charge!), every time its the IP2312 battery charge IC - exactly how this manifests seems to vary - depends which or both of its power fet switches short out!
Sometimes it shorts the +5V to the battery - which then shuts down... sometimes it shorts the battery and +5V to ground (battery also shuts down!) I think...
Either way it usually gets hot!!
What I think is going on is (particularly when its off, but charging the battery) the main +5V regulator stability is borderline; Ive seen it oscillate at about 30kHz, with the peaks going over 6V, I suspect that this is sometimes getting too high for the IP2312 to take, its absolute max is 6.5V with max recommended working voltage of 5.5V... Putting a larger package 10uF ceramic cap on the input to the IP2312 (instead of the 1uF C26) along wth the 10uF cap + 2R series resistor - as detailed in the datasheet (which implies its to deal with long wires/inductive supplies being plugged in, though ultimately any source with a rising output impedance with frequency, like a regulator! can look inductive...) seems to stop this, though I think the root cause may actually be the main +5V regulator stability - Ive been meaning to measure the 470uF caps ESR, I suspect they may not be low ESR...
If so then putting low ESR may solve a variety of issues Ive seen...

The only other issue Ive found so far is when the IP2312 went the last time, it seemed to take out the 8051/STC8G2K32S2 adc input that measures the battery voltage - it seems the input protect diodes on it have shorted to +5V or similar, which also caused the 8051/STC8G2K32S2 to disable the parallel port(perhaps some interaction between io pins when one has been damaged) ! lifting the ADC input pin on the 8051/STC8G2K32S2 solved that... this one still has one remaining issue, it draws a little too much power so the battery only lasts about 45 mins... also (with my +5V monitor chip) it re-sets quite often, particularity when some sounds are playing - it may be the power amp chip, though really to be sure it needs a new 8051....
I have done another mod here on my newer, fully working machine, according to the datasheet that micro should really have a series 100R resistor on that input, to cover situations where the input goes high before the +5V is stable... this doesnt happen much on a working pocket386 - only time I can think of is for a brief period when the regulator is powering up, so I think that's why they get away with it most of the time.... Anyway while i was at it I put an RC low pass filter (I think it was 1K and 10N) on the input to remove switching noise, now get lovely stable battery voltage readings in the OSD - this mod has a problem atm though - when you plugin the 12V and the battery switches of, the input to the filter is open cct and its output drops slowly and this confuses the software into thinking the battery is low ( i think if its in the range of something like 2 to 3V it gives you the on screen low batt warning, less than that it assumes the battery is disconnected & therefore charging...). My fix for this, which I haven't got round to testing, would be a resistor on the input to the filter to ground, prob about 10 to 30K.... to discharge the filter cap more quickly when the battery is disconnected...

I think boot time for Win 95 is OK & loading word is OK as well- as I said in a previous post, using an early (pre i.e 4 I think) version of Win95 is key, later than that has a bigger memory footprint. Also don't load the USB memory driver (unless you need it), as it forces slower access to the CF card....
Perhaps that explains some of why others have found it unacceptable?
IIRC 30 sec boot time for PCs back in the day was pretty normal 😀...
Id like a 16 bit ISA to play with! more than happy to assemble myself & cover costs... Im in the UK though?

30 sescs booting win95? Who am I kidding?? 😁 I just checked mine, about 50 secs to get to a log in prompt... & if you let it do the usual disk thrashing at log in and settle down... then its about 20 secs to open up word (the first time, it caches enough to get it down to around 12 secs the second time)
Odd, I could have sworn it was faster then that... still, ive been mucking about installing all sorts, perhaps it did used to be faster with a fresh install???

jakfish wrote on 2026-04-23, 19:37:

I found a fairly fast cf for the P386, but I'm nowhere near 30-sec+.

Fingers crossed that you reanimate the device.

Yea, I'm actually looking forward to hacking on it. I'm not nearly as familar with what's in there as Jules and with the notes here it'll make an interesting cleanup exercise.

And a few minutes for a boot time seems like a heck of a variance. I'd +/- 10 seconds on mine down to positive thinking but I'll have to actually time it when it's back up and running.

I don't know if it was the stock Win95 install either. I thought so but now I'm questioning that. When I bought it I also had a chuck bin full of CFs and spent quite a while just installing different things on there and messing about.

But I don't recall noticing a difference, just quite a few non-booting attempts to install different OSs due to the wonky DM&P/ALi SoC implementation.

I did another test on mine, I had an image of a fresh Win95 install ( Version 4.00 950B, Windows 95 OSR2 - but before they added usb support & IE4)
On a Sandisk extreme II 2GB (20MB/S - the pocket386 cant exploit the full transfer rates, but it still seems a bit quicker than slower cards)
it took mine 42 seconds from pressing the on switch to getting a log in prompt, TBF several seconds of that was switch on delay (from the PSU monitor chip I fitted) + BIOS POST etc
The loading win95 was around 35 secs...
My other one seems slower to boot up (on a sandisk extreme III 8GB, 60MB/s) but I had installed an unofficial service pack (who knows what thats done!) when I was setting up the serial WIFI card in it....
Sometimes win95 with networking setup seems to take longer on bootup as it looks for/sets up networking - maybe thats part of it...

With 9x there can be a longish 10-20 second stall where it waits for a DHCP response if networking is on and no static address set.

For low end hardware, the Win 95A floppy edition tends to be the fastest.

@Inhibit @Jules_nerd @BitWrangler Many thanks for weighing in on W95B speed. I rearranged my startup and went back to my 95 setup and it booted faster than I thought--it just seemed like minutes, and certainly the lag of program startups remains considerable.

Apropos of wasting my sanity, I have both Polpo's gpio wifi adapter and a Tindy serail-to-wifi adapter, and with the gpio, I'm on-line, via M. Brutman's fine work, in both DOS and Win311. No-go in 95, tho.

I installed DUN in 95, set up a modem dial-up, but when trying to connect, I get an error that my dial-up is conflict with another already-running dial-up. I'm assuming that's because I'm calling M. Brutman's connection in autoexec.bat.

I could REM the Brutman commands, but have to wonder if Win95's Communications package (its 5 files had to be installed in Add Programs) is enough to pull off an internet connection.

Has anybody gotten on-line with a P386/Win95B/serial-to-wifi union?

@BitWrangler, good call on the WIN95A (the 14 floppy) version 😀, I put that on the same 2GB sandisk extreme (just for those wanting to try it, this win95 version is FAT16 only...), it boots to log in only slightly faster, around 39 secs or so, but there is a lot less disk thrashing after log in & you get a usable start menu much faster & it feels more 'snappy', word in office 97 loads up in about 10 seconds (with all the office stuff taken out of the start up folder)... TBH I think thats pretty decent really, esp for a 386sx... of course we are all used to faster boot times etc these days, but I think that kind of speed was pretty common back n the day (albiet 486/pentiums on later versions of win95...)...

@jackfish, yes Ive got online with an internal wifi module soldered to the (V1.2) serial interface PCB, it was a while ago now so Im not sure how i did it! I can have a delve, I think I did a post on here that mentions win95 after I did it? I seem to remember a cisco dial up adapter driver or something... It is very, very slow. Whilst the serial wifi module worked can get close (ish) to 10KB/sec for file transfers from SMB shares on a win98 machine, it drops right down to 1 to 1.5KB or so when web browsing as the processor struggles to keep up with both rendering the pages & graphics & getting the data from the serial interface all at the same time... Brutman was kind enough to offer some advice, it turns out that MTU of around 500 seems about the best...
I may have another go and see if I can get it working on this fresh WIN95A install, but a bit busy for a couple of days here...

Have to admit, every single time I get notified about new posts on this thread, I get excited in case it is about someone releasing a pocket"486" with 16MB of RAM ... find that would make a much better 95 "pocket".

386, specially SX, even with a for cheap SX huge amount of 8MB RAM is more a W3.11 machine.
Problem is that 486 means 32 bit bus = more complicated PCB and less a n easy project for fun machine for builder as 386 Pocket is.

@jackfish Ive just checked my win95B machine with the Wifi (esp12) module. I dont think you need any of the brutman stuff - thats more for DOS & 3.11(with a TCP/IP stack like trumpet)
Ive just got dial up installed and I used a cisco .inf file i found on the net, i couldn't find a suitable dial up adapter in the standard win95 install - but I think others may have & I cant pretend mine is the ideal solution, it works, but goes through the pantomime of 'dialing' a phone number etc that takes a while....

javispedro1 wrote on 2026-04-25, 11:24:

Have to admit, every single time I get notified about new posts on this thread, I get excited in case it is about someone releasing a pocket"486" with 16MB of RAM ... find that would make a much better 95 "pocket".

Next known step up in currently available SOCs is more of a 586, extra fast enhanced 486, typically using the Vortex86 core. Which would probably be alright for Quake and DN3D but could be "too fast" for 486 stuff than needs a little more speed than this 386. Yet while having the integer oomph of low end pentium, wouldn't have the instructions for the early pentium required stuff. So would be a bit frustrating that it blew through the classic 486 envelope and out the other side. There might be a viable market for something like that though when Toshiba Librettos go from rare to unobtainium.

There are FPGA 486 cores with chipset and VGA, but FPGAs can be a bit thirsty in the power department. So small devices with them would need to be half battery or more.

BitWrangler wrote on 2026-04-25, 13:45:

Next known step up in currently available SOCs is more of a 586, extra fast enhanced 486, typically using the Vortex86 core.

I'd love a Vortex86 laptop. And they've gotten even faster with the die shrinks if I recall. Although to be fair I haven't ordered one and that may be a different core. I am curious what a neutered 800mhz kinda-586 looks like!

I always have to remember that the <number>86 numbering just referred to marketing for the Intel chips and everyone else is just a smorgasboard of functions that may or may not be equivelent. Came up again with the recent legacy processor's code leaving mainline linux (although in reality support ended some time ago).

Well the Via C3 chips with Winchip based core and half speed FPU would be somewhat close I'd think. Winchip was kinda hardware 486 instructions and emulation of everything less used.

@Jules_nerd Many thanks for looking into this. Weirdly, Win95 identified (supposedly) Polpo's gpio wifi adapter, but Win95B is missing some network files (ftp.exe, etc) that I think are stored in the Winsock 2.0 .cab. I found the .cab but in the meantime, Win95 freezes opening my usb d: drive past a list of files. Same drive works fine in Win311. I need to look into this further, since the usb is impt to my setup.

Other than the small hassle of .doc conversion b/w Word 6.0 and Word 97, Win311 seems more stable. With WS_FTP, I can swap files and with 3.11 stock program, I can telnet to lynx.scramworks.net, the last of the BBSs that allows lynx browsing. I got a 16-bit Pegasus to check email, but very slowly and missing a utf-8 character set to properly present smart quotes, etc. 16-bit Outlook Express 4.0 does have utf-8 but you have to spend significant minutes opening the program and downloading an email.

Word 6.0, with a add-on converter, will open a 32-bit .doc but it won't save it as such. My guess is the converter is deliberately dumbed down, as MS wanted folks to move on the new Word.