8mb of memory is enough.
If you edit the Config.sys file and load all devices HIGH then you can make it run using only 3mb RAM
This Free’s up conventual memory for games and programs.

This is a DOS/Win3x computer.
For Win95 you really should use a Pentium CPU class computer.

Intel486dx33 wrote on 2023-09-18, 19:00:

8mb of memory is enough. If you edit the Config.sys file and load all devices HIGH then you can make it run using only 3mb RAM […]
Show full quote

8mb of memory is enough.
If you edit the Config.sys file and load all devices HIGH then you can make it run using only 3mb RAM
This Free’s up conventual memory for games and programs.

This is a DOS/Win3x computer.
For Win95 you really should use a Pentium computer.

That's my plan. I don't think I'll even install Windows 3.1 on it, but we'll see what happens.

I actually had a surprisingly productive evening. The generic VLB multi io card worked out the gate. My Gotek didn't. I had to make sure either the JC jumper was closed, or the interface was set to ibmpc. That problem solved, I moved onto getting DOS 6.22 installed on the compact flash card.

This also caused a problem, where it appeared to install fine, but then the drive refused to boot into anything. I did a bunch of basic, random stuff to try to narrow things down. Took out some extra memory, turned off all the shadowing and the caches. None of it helped. Then I shuffled a compact flash card from another system which booted just fine, and then tried the new one in the other system where it also failed to boot. So that narrowed it down to the card itself. But, I've had weird issues like this before with drives that had unknown prior states, and after a quick FDISK /MBR it was booting properly into DOS 6.22. Didn't even have to reinstall.

The ES1868 wasn't detected at all at first. I moved it to another ISA slot where it worked fine... for a while. I think it's very sensitive to exactly how it's inserted into the socket, so getting it properly mounted instead of loose on the bench will probably help a lot. I'm using a driver package from DOS Days, and I actually hooked up a CD-ROM to it's IDE port with the help of this old blog.

I ran a few benchmarks, and it seems to be underperforming a bit. So I dropped the "Cache Write Option" down to 0 W.S. (YOLO) and seem to be getting 3DBench and Doom FPS closer to what I expected from other system benchmarks. About 41 FPS in 3DBench and 22.5 FPS/3311 realticks in Doom. I think my initial 3DBench result was closer to 32 than 42. No ill effects so far.

There are a smattering of BIOS settings I'm very unfamiliar with however.

I have no clue what "Hard Disk Type 47 RAM Area" could possibly be, nor "Memory Remapping", "F Segment Shadow RAM", "E Segment Shadow RAM" or any of the disabled "Non-Cacheable BlockX" settings.

Namrok wrote on 2023-09-19, 02:41:

I have no clue what "Hard Disk Type 47 RAM Area" could possibly be, nor "Memory Remapping", "F Segment Shadow RAM", "E Segment Shadow RAM" or any of the disabled "Non-Cacheable BlockX" settings.

I have a copy of Programmer's Guide to the AMIBIOS and have looked at a bunch of these so here goes.

Hard Disk Type 47 RAM Area. The BIOS has to set up a Fixed Disk Parameter Table that holds the C/H/S values among other things. When you use a preset disk type, it also has a bunch of pre-made FDPTs in ROM and sets up a pointer to the ROM. With user-defined types, the table has to be in RAM. The choices are to steal some of the BIOS stack area/ROM BASIC interrupt vectors and use it to store the FDPT, or to steal a KB of conventional memory. The option is disregarded if shadowing is enabled and it can just patch the FDPT into the shadow RAM. The "steal a KB" option is obviously not preferred but I believe it may have been required for OS/2 compatibility for its side effect of reserving some space for 286-to-386 loadall translation (being so random, that's definitely something I remember from the book).

Memory remapping is for memory-scarce systems such as 2MB or 4MB systems and often the chipset can't do it with any more than that. Of the 384KB of RAM you have between 640K and 1024K, 128K is used for shadowing (C000 for video BIOS/SCSI BIOS/etc., F000 for system BIOS) and the other 256K is lost. Memory remapping recovers that 256K by moving it to the top of the address space. It can cause issues if that 256K is beyond your cacheable area, and I had another case where a VLB video driver incorrectly found the top of memory and tried to map the linear framebuffer there. Since we tend to use these systems with plenty of RAM you may want this off.

F segment shadow RAM is copying the system BIOS from slow and narrow ROM to fast and wide (32-bit) RAM, always keep enabled. Same with C000-C7FF shadowing aka "video BIOS shadowing"

E segment would be the same, but only if you have option ROMs there or if the BIOS spills over into there. You would not see this on ISA/VLB systems but possibly an EISA system would have BIOS there.

I can't think of a use of the non-cacheable blocks. It would make sense if the system was otherwise erroneously caching RAM on a VLB card (such as video framebuffer or network card shared memory buffer or something) but as far as I can tell all the chipsets are already smart enough to not cache off-motherboard RAM. Perhaps the non-cacheable blocks are needed when you have both SIMM RAM as well as memory on some card in the same location (such as VGA linear framebuffer--which would be disabled by default and therefore wouldn't be detected during POST) similar to the "memory hole at 15MB" option that some chipsets have. You would want all those disabled.

It's likely that your bios has hidden options that you can see using the program AMISETUP. In particular underneath that autoconfig option there is usually an option for burst cache read timing which is a big deal for 486 cache/memory performance (2-1-1-1/3-1-1-1/2-2-2-2/3-2-2-2). I've found the autoconfig is often pessimistic since now we all have 60ns or 70ns SIMMs and 15ns cache SRAM after upgrading these old systems. You can run the CACHECHK program to quickly check on your cache/memory setup. This is what it looks like on a fully-tuned 486DX2-66 system I was just working on, so if your values are drastically slower you know something isn't tuned right.

Well, these results are disheartening.

The reported RAM read time is nearly 4x yours. However that may be skewed by access times past 512 KB sizes being enormous compared to all the rest, which while slower, seem competitive. I'm just running 2 banks of 30-pin SIMMs.

Definitely some additional wait states and such as hidden settings (use amisetup)

Yeah, I messed around with AMIBIOS some. Dropped memory timings to 3-1-1-1 from 3-2-2-2 for a start. Also dropped, I think it was "Cache Read Option" to 1 W.S. from 2 W.S. The biggest change I noticed was turning on FastPage. Numbers are looking a lot better from where they were. Still off from yours. I did pick up a single extra FPS in the Doom timedemo.

Edit: Alright, getting closer. Close enough that I'm happy enough with it.

I believe the settings I finally arrived at were

• Cache Read Option: 2-1-1-1
• Cache Write Option: 0 W.S.
• DRAM Type: FastPage
• DRAM Wait States: 0 W.S.
• Fast Gate A20 Option: Enabled

Raw memory throughput aside, it's putting me right where I'd expect from Phil's Ultimate VGA Benchmark Database. 2973 Realticks in the Doom benchmark, which is solidly middle of the pack for AMD DX2 results on that sheet.

Edit2: Checking this 486 Chipset Race I found, the performance of my PC Chips, SARC chipset motherboard is probably about right. In fact, I wonder if I was able to use such aggressive memory timings and get away with it largely because the chipset/motherboard itself is such a laggard.

Looks good. I think your SARC chipset is actually a relabeled UM491 which is the same I used in the test above. However there is another piece. Notice the empty SRAM socket on your board. That is presumably for dirty RAM and left empty to save money. The cache performs better with that fitted for reasons you can read elsewhere on vogons by looking up "always dirty." Those chips (32Kx8 SRAM 15ns or 20ns skinny DIP) you should be able to get for under $10 on ebay. That's the final piece to get those 55-ish timings for RAM access down into the 30s. You're lucky, as I had to do a hardware modification on mine to add one which I did for practice as I am trying to get better at fine soldering.

Hmmmm, this is all new ground for me, but I think it already has a TAG chip? I honestly have no clue what the empty socket is, and it looks like it could be one. However, empty socket or no, the cache configuration here shows two banks of 4 chips, with a 9th tag chip, and my board has 9 chips installed. That empty 10th socket is simply not listed on any of the sources for information about this board I have. Although I do see a photo of what is allegedly the same board here with that 10th socket filled, so I suppose it can't hurt to try.

Cache performance should be fine as long as you don't enable write back.

Anonymous Coward wrote on 2023-09-20, 02:44:

Cache performance should be fine as long as you don't enable write back.

Unless someone has run into convincing evidence otherwise, I believe both the UM481 and UM491 run L2 cache as either write-back or disabled. There is no write-through on these chipsets

Weird. We’ll if that’s the case then finding that chip should be a priority. Cache that’s always dirty isn’t much better than going without.

Perhaps this is a socket for a dirty tag ram or the configuration with 2x 8K tag chips.

I think I'm going to table the cache for now. I don't want to get side tracked tilting at windmills.

With where it's at now, whatever problems it may still have aren't showing up in games or benchmarks near as I can tell. I found another set of benchmark results specifically for CACHECHK on DX2's. It appears that my results of 68.8 MB/s for L1, 42.6 MB/s for L2 and 19 MB/s for RAM is, if not bang on, even above average by these results.

One weird problem I am still having is that whenever I plug in the PC Speaker to JP11 it complains about the keyboard being locked, and I can't progress past that. I do see keylock is on the adjacent JP10, but I don't have any specific pinouts for either.

At risk of jinxing myself, I'm happy with how easy this build came together. Nothing I bought was defective, dead, or incompatible near as I can tell. Which is good because I really didn't have spares to go rigorously testing parts in known good configurations. It's time to move onto the case.

I want to make a wooden case, roughly in the style of an old TV. Back when TV's looked more like furniture. I want to mount an LCD screen and speakers directly into it, and have panels covering the CD-ROM and GOTEK when I'm not using them. I have a cheapo case I don't care about which I'll probably carve up with my dremel for drive cages and slot brackets.

I'm still working out exactly what design elements of these old TVs I want to incorporate into the case. I'm not the most amazing woodworker, so some compromises will likely have to be made. It will however be made with red oak, likely with an Early American Classic stain. Because the heart wants what the heart wants, and even though my wife hates it something about that combination speaks to me. Probably just a seal coat of Shellac since I want it to be fairly matte in appearance.

I think my plan is to start coming up with design sketches and milling my rough lumber this weekend.

Hello,

I've enjoyed reading about this 486 build. I have found memories of my own 486 back in the early 90's as the first PC I bought myself. Did a lot and learned a lot on that machine.

But I'll be following this thread as I'm looking forward to seeing how your proposed tv style case turns out!

Best wishes,

Tony.

Well, I sketched out what I think I'm gonna try to do. I found myself primarily inspired by the second picture I posted. Going to do similar style grills for the speakers, only moved below the monitor instead of to it's sides. I think I appreciate a taller, leaner profile for an all in one desktop.

I really don't want to have to glue up panels for the two sides, bottom or top, so if I can make everything fit in a case that's about 9" deep, with the motherboard mounted vertically on the back side, I think I'm good. Might do a mockup with cardboard or something just to see if my idea of having the motherboard hinge outwork to be installed and maintained is remotely feasible. I also need to actually remove the drive cage of my spare case, and see what it's exact dimensions might be, and how I might mount it in a removable fashion.

I think I've also settled on a Kinter amp with 4" speakers. The dimensions look like it could possibly fit inside a drive bay?

Well, fiddled around with this system some more. Connected the CD Audio cable to my ES1868. Then got an old serial mouse connected and CTMOUSE put on. Installed WarCraft II to make sure it all works, and it's.... weird.

The CD audio works, so yay. But the sound effects are very crackly. Doom 2 doesn't have that problem at all.

I tried a few different digital sound settings in WarCraft, including backing all the way down to Sound Blaster mode, to no effect. Also tried turning off the CD Audio and the FM Audio, just in case, also no change. Then I tried changing around the DMA and IRQ settings, which definitely only made the problem worse in that WarCraft II began crashing. Doom 2 was fine. On a lark I tried changing the AT Clock Select from CPUCLK/4 to 7.44 Mhz, thinking maybe the bus was too fast for the ISA card, but it just caused the system to fail to POST. So I walked that back. When the BIOS was back to a virgin state I tried to game again just to see if any of the optimizations I'd made were the cause, and nope. Not really sure what's going on here. Weirdly the movies don't have that issue, or the menus. But the scenario briefings and gameplay sound effects do. Who knows! We'll see how other games work I guess.

On a wild guess, I started some CD audio playing, then launched Doom 2, wondering if some sort of crappy mixing of sources was causing the audio crackling, and nope. I did try Master of Orion as well, and the audio seemed fine there too. I donno, maybe I'll try a different ISA slot or PSU.

I saw some conflicting information about what sort of voltage the external CMOS battery needs. The first source I found said 6V so I had been doing that, but another said 3V so I backed down to that. Doesn't seem to make a difference, so I'm gonna stick with the lower voltage and hope the 6V battery pack didn't do any harm.

I did also turn on IDE BLOCK MODE which made a huge different in programs loading, so that's good.

Probably going to be rained in tomorrow, so hopefully I can spend some time in my shop dismantling the spare case and getting some boards milled.

I don't think you did any harm with 6V. On these boards it usually goes through two diodes anyway which would drop it below 5V which is the voltage it receives when powered on. You may find that 3V is enough to hold the settings when powered off, but that the clock freezes in place until turned back on. When I attempt to power these boards using a single CR2032 that's the symptom I get.

I don't know anything about the crackling problem but I do wonder if you have seen UNISOUND right here on the forum. The ESS configuration utility doesn't annoy me the way the DOS Creative software does but UNISOUND is an option too.

I would not worry about motherboard cache. As long as you have it working with 64kb cache that's fine.
adding more cache will gain you about a 3% boost in performance. I have performed allot benchmarks on different 486 computers to confirm.

The best performance boost is to upgrade the CPU.
Intel overdrive CPU

jakethompson1 wrote on 2023-09-23, 04:45:

I don't think you did any harm with 6V. On these boards it usually goes through two diodes anyway which would drop it below 5V which is the voltage it receives when powered on. You may find that 3V is enough to hold the settings when powered off, but that the clock freezes in place until turned back on. When I attempt to power these boards using a single CR2032 that's the symptom I get.

I don't know anything about the crackling problem but I do wonder if you have seen UNISOUND right here on the forum. The ESS configuration utility doesn't annoy me the way the DOS Creative software does but UNISOUND is an option too.

Good to know about the CMOS voltage.

You know, I tried UNISOUND. But it messed up the IDE settings of the CD-ROM I have connected to the ES1868. And then of course WarCraft II requires a CD to get into the single player. I know there are ways to load the CD drivers after running UNISOUND, but I haven't messed around with them yet. I may just move the CD-ROM to be slave on the VLB IDE controller. Who knows, that could be part of my issue too!

EDIT: Yeah no, none of that mattered. Switched the CD-ROM over to the VLB IDE instead of the ES1868 IDE, no change. Used UNISOUND instead of the ESS drivers, no change. So strange.

Intel486dx33 wrote on 2023-09-23, 09:33:

I would not worry about motherboard cache. As long as you have it working with 64kb cache that's fine. adding more cache will ga […]
Show full quote

I would not worry about motherboard cache. As long as you have it working with 64kb cache that's fine.
adding more cache will gain you about a 3% boost in performance. I have performed allot benchmarks on different 486 computers to confirm.

The best performance boost is to upgrade the CPU.
Intel overdrive CPU

I'm not really worried about pushing this motherboard to the utmost limits of 486's. I specifically wanted a DX2. I mostly just don't want it to be a horribly gimped DX2. The default BIOS settings had it about 25-33% slower. Seems to be about where it's supposed to be now.