Yes, all of them except for Bsquare. I'm trying to get the Bsquare driver from Bsquare, but they are having a hard time finding it.

Stay away from the Jungo USB driver, at least version 8.02. It gave me BSOD upon installation and now I cannot even boot into NT 4.0 (BSOD at boot). Good thing I tested it out on my junk box.

The last idea I have is to try the Xircom portstation, which are a bunch of modules, like modem, parallel port, PS/2 ports, ethernet, that connect to a USB line. They have NT4.0 drivers. So I figure I can take out all the modules except for the PS/2 module. I do not have high hopes, but I get surprised sometimes.

I just rediscovered that not all Biostar MB8433-UUD motherboards work as well as mine. I'm using a v3.0 board with 2-1-2 and 0/0 cache and ram speed settings, respectively. It works fine on a Cyrix 5x86-120 and Cyrix 5x86-133 and has worked great for years.

I have another v3.0 board with a slightly newer chipset date, and it will choke on any cache setting other than 3-2-2. 0/0 for RAM is fine though. Both boards have identical 512 KB cache chips.

The same rings true for a v2.0 and v1.0 board I have. The cache isn't much faster than the RAM at a 3-2-2 setting. It seems I can increase the cache speed to 2-1-1 if I use double-banked 256 KB cache instead of the single-banked 512 KB cache I've been using. However in WB cache mode, I'd be limited to 32 MB RAM with 256 KB cache and such a configuration isn't practical in a GUI operating system.

Has anyone else noticed this similar oddity?

I got Firefox 2.0.0.23 running under NT4. It is manually streamlined for low memory usage and good response times. (see the ff2 video on my ftp)
The main points are:
- disable flash/java and javascript if possible (most pages like forums work also without javascript fine)
- no plugins / add ons (Adblock Plus f.e. reduces performance by some orders of magnitude)
- disable rendering while loading (rendering takes longer than loading on low cpu system which is a contradiction to browsers default settings. Same behaviour as early browsers.)
- flush memory use for non visible components instantly. (Basically memory is only used for the actually shown content.)

If you are you interested in the specific settings I could save a copy.

If memory serves me, 2.0.0.16 was the latest Firefox I was able to get installed on NT4.0 and a 486-class processor. Firefox works fine for me, but I usually use IE6 as it feels faster. I followed your recommendations, but am unfamiliar with your final point, "flush memory use for non visible components instantly." Perhaps you can elaborate?

Are you using 2.0.0.23 on a 486?

feipoa wrote:

If memory serves me, 2.0.0.16 was the latest Firefox I was able to get installed on NT4.0 and a 486-class processor. Firefox works fine for me, but I usually use IE6 as it feels faster. I followed your recommendations, but am unfamiliar with your final point, "flush memory use for non visible components instantly." Perhaps you can elaborate?

Are you using 2.0.0.23 on a 486?

I have currently no access to this retro machine. So the following is from memory:

This is the time how long it takes until firefox starts rendering after loading started. If you set this to a time that is higher than the page loading time firefox will not render while loading. This is much preferred for slow CPU systems.
nglayout.initialpaint.delay (in ms)
I would also increase this:
http://kb.mozillazine.org/Content.notify.interval
and check this
http://kb.mozillazine.org/Content.switch.threshold

Browser memory cache size, set this manually since the automatic detection for low mem system is somewhat off reality:
browser. cache. memory. capacity

Pages cached in memory for fast forward backward navigation:
browser. sessionhistory. max_total_viewers
set this to 0

Set this to true to let windows trim the firefox memory usage on minimize, check with taskmanager.
config. trim_on_minimize

I am currently not sure, if there was another setting for opened background Tabs (such way that only the active Tab is kept in memory)
Please check also
http://kb.mozillazine.org/About:config_Entries#Advanced.
http://kb.mozillazine.org/Category:Tweaking_preferences

My FF runs on a Pentium 166MMX. So it can aswell be that the last version compatible with a 486 CPU is 2.0.0.16.

For those who are not familiar with this modification, in Firefox's URL window, type About:config. This will bring up the configuration editor.

Most of the below settings do not exist, so they will need to be created (right click, create...). Where a number value is desired, create an integer. Where a true/false is desired, create a boolean.

I'm going to start-out with a test of the following settings,

nglayout.initialpaint.delay = 2000 ms (previous value not existent)
content.notify.ontimer = true (previous value not existent)
content.notify.interval = 1000000 us (previous value not existent)
content.interrupt.parsing = true (previous value not existent)
content.switch.threshold = 100000 us (previous value not existent)
browser.cache.memory.enable = true (previous value already true)
browser.cache.memory.capacity = 4096 (previous value not existent)
browser.sessionhistory.max_total_viewers = 0 (previous value -1)
config.trim_on_minimize = true (previous value not existent)

@elianda, when your machine is available, please let me know what settings you decided were the most optimal for your (rather fast) Pentium 166MMX. I'm using a significantly slower Cyrix 5x86-133 (~Pentium100).

EDIT: Well, I don't really know how fast Firefox was before the mods (I'd need to get out a pocket timer), but it still isn't quite as fast as IE6. Has anyone else played with these settings? If so, what was found to be most optimal on a 486? Thanks.

Well, I figured I'd add some test result images to what I consider to be the world's fastest 486. This thread follows largely from this post,
So you want a Cyrix 5x86-133?

The IBM 5x86C-133 is running on a 66 MHz front-side bus and a 2x cpu multiplier. Vcc is 3.85 V.

Proof is in the photos. Although the photos are many, I cut the size way down so that 30 photos only takes 1.5 MB.

The remainder of the test results from the U4BC.

5 image limit? Perhaps a total KB limit per posting would be better, with a variable image limit? These images are way small.

The last of the DOS benchies.

Now for Windows 98SE tests.

A few more Windows 98SE benchmark results.

I thought I'd share some images of the two cases which these beasts reside in. The cases aren't quite identical twins, but the hardware is the same (see page 1 for specs). Case photo moved to page-1 post. Left case at 66x2 right case at 33x4.

The right case has a new Sony SATA DVD-RW drive which gets connected to the SCSI 50-pin narrow cable through a SATA-to-SCSI converter card. It seems to work fine and I installed WinNT4.0 and W2K using it.

The left case also has a Sony SATA DVD-RW drive, but connected to the onboard IDE header in PIO-4 mode. The left case contains a VIA PCMCIA external card readre, while the right case also contains an external CF card reader connected to the onboard IDE port. The IDE/CF adapter works fine in PIO-4 mode and an 8 GB CF card. The BIOS has no issue with the 8 GB CF card. I don't use the CF card much except to test questionable software that I don't want to mess up my SCSI harddisk installation. The 8 GB CF card contains a clone of my Win98SE/WinNT4.0/W2K SCSI media.

The CF/IDE card reader also works on the SCSI bus with an IDE-to-SCSI converter, but it doesn't work when my SCSI scanner is attached (which it currently is).

Stable BIOS settings for IBM 5x86C-133 at 66 MHz x 2 and 3.85 Volts.
EDIT: Latest stability tests reveailed that a 2 ws DRAM READ was required for long-term stabiliy, with the ideal voltage being in the 3.73-3.75 V range.

I am now using the 2 x 66 MHz FSB system as my daily computer (upgraded from the 33 MHz x 4) to test the long-term stability of this upclock. I have discovered that everything is fine with the case off, but with the case on, the system crashes after 15 minutes. The base of the CPU is quite warm, so the cause is thermal.

I beleive the U320 SCSI drive is the heat culprit, which is right next to the CPU. These harddrives get awefully hot. I've added a front case fan (to a 486!) which really seems to help. The system hasn't crashed yet with the case on.

I'm going to let Subspace run in spectator mode for several hours to confirm things are ok. I've taken the CPU voltage down to 3.80V to reduce heat. The 2 x 66 MHz feels a lot more chipper than the 4 x 33 MHz system. I should be at or above P100 levels now. Using Win NT 4.0, it feels a lot more like a P133.

Some subsystem component photos were added to the start thread along with my customised Biostar MB-8433UUD user manual.

Due to the difficulty in members finding the latest BIOS for this board, I have added it here. It is version UUD960520s (20 May 1996), AWARD 4.51PG, and is the latest known BIOS for this board. It is only to be used with the 5V SST style of EEPROM (SST PH29EE010 150-3CF). Happy flashing!

@feipoa

I must say, that Cyrix 5x86 is phenomenal!

As soon as you get that TNT card working you'll most definitely have a beast of a 486 (hope to be right there with you eventually!)

Now that I've confirmed my DIP-32 sockets are working, this M919 has L2 cache! (amazing)... I just need to find some 10 or 12ns chips and, while I'm at it, upgrade from 256K L2 to either 512K or 1024K (ideally 1024K since I have 128MB RAM). All the places online with these chips are overseas, which I'm not 100% comfortable with. Did you purchase yours from a merchant in North America?

Attached is my latest Speedsys 4.78 result, with the L2.

numeriK, nice results. You are on your way to 486 greatness. The key for me was to have a long-term stable system. How stable is your 200 MHz system in Windows? Your system might be more stable at 180 MHz. You may also find that when you increase the amount of cache, that the cache timings have to be slowed down. This is not always true.

I sourced my 10 ns cache from China. If you find any N. American distributors, I'd be surprised if the cache was not originally from the same factory(s) in China. I used seekic.com to find the 10 ns cache. I started with sending out several dozen inquiries for prices, datecode, and availability. Then I narrowed down the cheapest overall supplier for the sum of items on my list. On this order, I had a lot more than cache I was ordering - BIOS chips, counters, PLL's, flip-flops, etc. You can setup Excel to sort your suppliers by cheapest total cost. I then bargained down a few items I thought were too expensive. There is a lot of competition for sales from Chinese IC vendors, so bargaining tends to work. I am not sure what the minimum total order is, but suppliers were willing to send 20 cache DIPS.

If you just want cache, I recall the lowest price being about $0.80/unit for 1024 kbit, 10 ns, DIP-32. Some vendors require escrow for payment.

You may also want to refer to this link,
Super Fast 486 Cache - 10 ns

feipoa wrote:

6) After this read, I'm motivated to update the Ultimate 486 Benchmark Comparison to include derived CPU data for AMD X5-180, AMD X5-200, and Cyrix 5x86-150.

The Cx5x86 is capable of operating at 150mhz without frying? I thought 133mhz was at the thermal limit for the design and it was hard enough for Cyrix just to produce chips that were capable of operating at that speed. I was always told that unless you're using an underrated IBM chip, overclocking experiments with the Cyrix design tend to end badly.

The Ultimate 486 Benchmark Comparison contains the derived data for the Cyrix 5x86-150.

I doubt you'll ever find a Cyrix 5x86 which is stable at 150 MHz. Maybe someone with the newer fab date 5x86-133/4x can test it at 150 MHz? I know at least 20 people on CPU-world have one of those, but nobody seems to want to test it. From this forum, Neon_WA and Tetrium have this chip.

Forum user nemesis got a speedsys score out of an IBM 5x86 at 150 MHz, but I don't think he had much luck with Windows. Forum user Anonymous Coward seems to have had success running a Cyrix 5x86-100 at 120 MHz. If that is true, then some Cyrix-branded chips are overclockable. I guess the closet thing to a stable Cyrix 5x86-150 would be a MediaGX chip. National got those up to 300 MHz. I wonder if it is possible to create some kind of MediaGX-to-socket3 interposer/converter board?