I would just make sure it's a mmx pentium that can be slowed with setmul.

SETMUL does allow L1 cache enable/disable for Pentium Classic. Back in 1996 I had to go to system BIOS to do that. I sure wish there would have been SETMUL back then. ICD and ICE worked only for a 486 and not for Pentium.

It was rusted from the back but otherwise in decent shape. Power button was very yellow but not rest of the plastic parts. I retrobrighted the button, grinded the rust off and then spray painted grinded areas and whole back of the case. I like when case is white also from behind.

But I didn't find just empty case. It had a very good motherboard inside and that is what I'm going to use. It also had near perfect graphics card, that I'm most likely going to use (but we'll test that). It also had decent sound card but I'm not going to use that one (one of the SB16 vibras). I just wanted the case but got a whole lot more: an excellent start for building a 1996 system.

And I'm already cheating a bit. That revision, 3.10, is from early 1997. But 430HX sure is 1996 chipset and that motherboard was originally released in 1996. So I hope nobody gets upset that there is a 1997 date code printed on bottom of the motherboard.

Intel 430TX is a 1997 chipset and not necessary the best Pentium chipset from Intel. Usually that title goes to the 430HX that I have here. HX is a professional chipset. Multi CPU support, ECC, RAM up to 512MB that can be fully cached, it is fast and stable etc. So lots of features I'm not actually going to need. In fact, for DOS system, TX and even VX are enough. But both can cache RAM up to 64MB only. No matter for DOS systems but Windows systems it does matter. I'm sure most VX systems didn't have over 64MB anyway but with TX that might not be the case. My friend had 80 MB (first 16MB that was later upgraded with 64MB DIMM), for example.

The motherboard came with two 60ns 32MB EDO SIMMs. I'm going to use those and add two more, although more than 32MB with DOS only gives problems. But my 486 has 32MB so this one must have more. I don't know if the two SIMMs are original since they did not contain any date codes. But they very well might be. And since I'm using HX chipset, then surely I need to use something to make this look better than TX. So 128 MB RAM will be a good amount. And I don't need to hunt rare and expensive 64 or 128 MB SIMMs.

The CPU under the heat sink is Pentium Classic 150 MHz. That is a bit odd here because of the 60MHz bus. Surely this motherboard would have deserved to be run at 66 MHz BUS. I'm certainly going to (but I'm testing the 75MHz with Cyrix and maybe I'll try to overclock Pentium as well). I'm thinking of using 166 MHz but I'll test 133 MHz as well. I don't have Pentium Classic 200 MHz. That would have been a 1996 CPU but I kind of don't like the idea of using a 200 MHz CPU on a 1996 system. But we'll test if the Pentium 166 MHz can be overclocked to 200 and run tests with it.

But before adding those two extra SIMMs, I need to add extra TAG RAM chip. Otherwise this motherboard will cache only up to 64MB. This revision already has 512k cache on board. It just needs the TAG RAM chip. The one soldered is 15ns so no need to use any faster that that.

But before going to CACHE, there are more issues. I need to make a battery mod for the DALLAS RTC. It just must be done but it is quite easy. Or would be but this one is soldered and I need to desolder it and add a socked. I need to be extra careful about the mod. See the other picture above. It is not trying to show the RTC but how well this motherboard is designed. Those VRM heat sinks are just perfect size so full lenght ISA cards can be used on every ISA slot. Really nice, although this motherboard only has three ISA slots. But it will be enough. So I need to be careful with the mod. I don't want it to prevent installing long cards.

aitotat wrote on 2023-03-17, 07:08:

[*]1.2 GB Seagate HDD (believe me, this was HUGE compared to my previous computer)

I think I had the same HDD back in 1996. It was from their "Medalist" line, if I'm not mistaken. And yeah, that was a huge amount of space for DOS and Win 3.1 games. I remember being able to do a full install of Fallout from the CD when that came out, and still had room to spare.

It also had a very satisfying crunchy sound while working. Kinda subdued, and not as harsh as some of the drives from the later years (e.g. those crappy Quantum LCT things). That said, I wouldn't use one today for practical reasons, but I do remember it fondly.

I hope I can do the RTC battery mod during next week and start running some benchmarks to see what is the optimum clock for the CPU. But before that there is also the TAG RAM issue. Well, actually I already have 15ns SRAM chip that can go there. According the manual 15ns or faster should be used. When looking at the pictures on retro web it can be seen that some boards have 15ns and some 12ns TAG RAM soldered. Is the 12ns better for some timings not reachable by 15ns? I don't know. But I ordered some 12ns SRAM chips from a Chinese seller. I know ordering those is unsafe and yes, that was the case. I don't know what I got but I sure noticed right away they are fake.

They all looked identical on the top (all 10). They all have 1415 printed below chip model number. That alone was enough to know they were fake. But they looked like grinded and still contained some dust and other visible hints that original markings were grinded off. Under the chips there were different kinds of markings. Different fonts etc. So whatever these are, they sure are not all the same chips. With luck they are all SRAMs but I don't believe a second they are 12ns SRAMs.

So I'm going to use the known good 15ns SRAM I already had.

Your P120 system was a pretty decent configuration for the time. While not high-end, it was good balance of power for the money.
The item to have that year was an 8X CD-ROM drive...I probably would have paired that with a cheap ESS soundcard.

I agree that some ESS soundcard would have been great in there.

I tried to find a receipt from my 1996 computer but maybe this is even better. Here is the basic configuration and options with prices. These are from august 1996 when I ordered it. I got the computer in early September, if I remember correctly. I had also saved brochure from Forefront monitors.

Lets examine a bit about those prices and options. It can be seen that I excluded 16-bit ExpertMedia wavetable sound card (I wonder what model it was? Maybe one of those Opti-based card with wavetable included?), 4-speed Mitsumi CD-ROM-drive and Windows 95. Looks like Windows 95 would have cost about the same as the sound card and CD-ROM-drive combined. Those prices are Finnish marks, by the way.

So I would have paid 6800 FIM back then. Actually a bit less since I also excluded those Forefront speakers (those would have been nice, take a look at the Forefront brochure). So maybe this cost about 6700 FIM? In todays money that would have been 1741 €.

So price of the Mitsumi 4x-CD-ROM-Drive was 250 FIM (65€) and to get 8x Hitachi instead would have cost 300 FIM more, so the full price for a 8x drive would have been 550 FIM (143€).

Price of the ExpertMedia wavetable card was 350 FIM (91€) and to upgr...downgrade that to SB16 Value Edition PnP would have cost 200 FIM more. So price of the cheapest SB16 was 550FIM (143€), same as Windows 95.

Notice that the price difference between SB32 and AWE32 is 500 FIM (143€). Sounds like a waste of money but not possible to know since exact models are unknown. You could have got SB32 + Windows95 with the price of AWE32.

I actually made some progress last night.

Looking good! No problems, except the RTC battery, of course. It does not even allow to save settings. As soon as I exit BIOS setup, default values will be loaded. But it was enough to boot from a floppy.

https://www.dosdays.co.uk/topics/1996.php

😃🤞

And as can be seen, it does not prevent using long ISA cards on any of the slots. I could't glue one of those typical battery holders to the RTC since it would have blocked one of the ISA slots. I could have glued one on the side of the RTC so that the battery holder goes over the motherboard. But it would have been risky since motherboard might have not fitted inside a case anymore. But this is a good result. I should have used a longer wire so I could have taped the battery holder on the bottom of the case. That would have made battery replacement very easy. Now, in worst case, I might have to remove ISA cards and maybe even the DALLAS as well when replacing a battery. But at least it is now on a socket.

Now what is the long card that appeared on one of the pictures above?

I wanted to build a period correct 1996 system. What exactly is period correct and what is not?
Is GoldLib period correct for a 1992/1993 build? It is a recreation of Adlib Gold and requires those old chips. But GoldLib cards are made in 2022 or 2023. What if you, for example, recap some 1996 sound card? Does it stop being period correct since it now has new capacitors? That can't be true, otherwise fixing things would somehow be a bad thing. So if fixing is OK, then what about building a new clone card? I would accept GoldLib as a period correct sound card for a 1992/1993 system.

But Orpheus II does have a problem here. That Crystal chip is released in 1997! No problem with the Interwave but that Crystal CS4237B is too new for a 1996 build. But it has the same features as CS4236B + 3D sound that nobody wants to enable anyway? Is it okay if we keep 3D sound disabled and pretend this is a CS4236B ? I think this kind of thinking is cheating.

By the way, the sound card that came with this computer is:

It has a good sound quality because of the Vibra chip. But the CT2505 Vibra does have CQM integrated in it. Then there are all the MPU-401 bugs as well as all the Vibra bugs. So you got the computer with that in 1996. A mistake that maybe had to be corrected in 1997 so a new better sound card is bought... And that way we can accept a sound chip from 1997!!! I know, I'm still cheating. But we'll look real 1996 sound card options once we get that far. And that Orpheus II might end up in the SS7 system. I don't know yet but I'm going to try it on this system first.

Now that the CMOS settings can be saved again, it is time to choose a CPU. I want to do all the CPU tests on a test bench so it is easier to access jumpers and change CPUs. I'm going to test Pentium 133, 150 and 166. Just for comparison, I think I'll test MMX 166, 200 and 233. And Cyrix 6x86L pr200+. I would have very much liked to test Pentium Classic 200 as well but I do not have it. Pentium 166 was multiplier locked (I tried) so best I can do is to test 75 * 2.5 = 187 MHz. It will most likely be faster than 200 would have been.

First thing to do is to find a fastest possible memory timings for the CPU tests. And 75 MHz bus turns out to be a bit problematic. I started memory tests with memtest86 3.5. I had the disk close. It is a good version since it works on a 386. Somewhere during the testing I decided to use memtest86+ (some 4.x) and it started to reveal problems. In fact, even the BIOS defaults for 60ns had errors. I need to do more testing. Nothing was overheating, although the RAM and chipset got warm. Cache chips didn't even get warm but maybe memtest disabled L2 cache. I was surprised how cool the overclocked CPU was. After all the testing, I removed the heatsink to try how hot the CPU was.

Next I need to test the SIMMs one pair at a time in case the other pair just does not like 75 MHz. Or maybe I should do the memory testing with 6x86 since it is meant to be used with 75 MHz bus.

I found out what the problem was. The tested 166MHz CPU is one of those that use "VRE" voltage. So 3.5V instead of more typical 3.3V. No wonder there were problems since the CPU was undervolted and overclocked at the same time. AMD and Cyrix have properly printed voltage requirements on the CPU but it is not so easy with classic Pentiums.

Now 60ns default timings work fine and I can continue finding fastest stable settings.

Memory testing continues but not without problems. Faster timings (faster than the 60ns default) produced single error with the Pentum 166@187MHz. It might have come at the second pass so quite random error. So I decided it is better to test with non-overclocked CPU. I could have used the Cyrix but that v4 memtest likely uses Pentium instructions not supported by Cyrix 6x86. So I skipped that and decided to test with Pentium MMX 233. It is not multiplier locked so I can use 75 x 3 = 225 MHz.

I also updated BIOS to the latest. I should have done that from the beginning as it might affect those timings. Motherboard had version #401A0-0109. There is 0207 with patch to support drives up to 128 GB. I decided to use that. I tried 3 different versions of awdflash but couldn't flash the chip, or more likely awdflash couldn't identify it for some reason. I had to use EPROM programmer instead. It didn't support that flash either since my programmer is older than this system I'm building. But I managed to flash it anyway as I found compatible Atmel flash settings that worked. This motherboard has Winbond flash in it.

Now I decided to test with L2 cache disabled in case it might affect the testing. Everything went really well.

DRAM Turbo Read Leadoff must be kept disabled since Memtest does not even start with that enabled. But other than that, just excellent!

So just one more test with L2 enabled to make sure everything works. And errors right away. For some reason cache is not happy with those settings. Is it 75 MHz speed that is the problem? Is is one of those settings? Is it the extra TAG SRAM I installed? Now I must test those one thing at a time. Why can't things be easy for once?

So there was just one setting that was too much for 75 MHz, RAS to CAS delay. It can be 2T for 66 MHz bus but needs to be 3T for 75 MHz bus. All other settings can be equal. Also it might need second memtest86 pass before the problems start to show up. So 2T for 75 MHz is just and just unstable. My guess is that the problem is with those 15ns TAG SRAMs. I'm sure 2T would be fully stable with 12ns SRAMs. I tried to remove the extra TAG SRAM I added but it did not change anything.

What do those BIOS settings mean. For excellent information, search PC engineers reference book on archive.org.

So it is best to keep DRAM Turbo Read Leadoff disabled, so no need to worry memtest86 did not start with that enabled. I also disabled Turn-Around Insertion that I previously had enabled. Disabled is faster since enabling it adds wait states. And since it is disabled by default, there should be no need to enable it.

Just one setting, Passive Release, is something I don't know yet should it be kept disabled or enabled. It might be that ISA sound cards work more nicely with that one disabled. But we'll see.

So we now have 128 MB RAM with fastest possible settings. Great! Time to choose the CPU next. Lots of benchmarking to do next week... But before starting those, I should select some video card for the tests. I'm going to use the one that most likely stays on this computer.

Surprised? It is just a typical Virge 325? Virge 325, yes, but not typical. This one is a hidden gem. You'll see when I put it against several other S3 cards and couple of other cards as well. But need to choose the CPU first.

I tried to start the CPU tests yesterday and immediately ran into a problem. Nothing really new and I should have expected it. I was going to use a 6GB Hitachi Microdrive since I have one that already has all the benchmarks etc in it. And Doom immediately showed lots of pauses while running the test. This is because the Microdrive has power management enabled by default and system BIOS does nothing to disable it (although I did not even look if there is a setting for that). XTIDE Universal BIOS, that I have on my other systems, does disable PM by default so there are no issues with Microdrives. I was planning to use one of those CF-HDD drives I made previously for my 486 systems. They have CF-to-IDE-adapter and Toshiba 60 GB 1.8" HDD. The Toshiba requires Advanced PM to be disabled as well and XTIDE Universal BIOS does that also.

There are software solutions that could be made but I'll leave those for later. Next I was going to use a 120 GB Hard Drive. I have a quiet 5400rpm Samsung drive that would be very nice. Of course it won't be a period correct drive. A period correct drive would be 1.2 GB, 1.7 GB, 2.0 GB or maybe even a 2.5 GB drive. I'm not sure if there were larger IDE drives available in 1996. But those are all noisy. I like the head clicking sounds but not the noise that comes from spinning the drive. I think 8-10 GB drives started to be a lot quieter and I could use those (but they are not period correct). But not all manufacturers made quiet drives even by then. I don't know who started those but I know there are quiet 8 and 10 GB Seagates and a noisy 20 GB Maxtor, for example.

But for now, I think I should do the testing with a period correct drive. Then test later if that 120 GB drive shows much difference.

There is another possible issue with later drives and DOS. Those drives are not made for DOS. They have large, 2 or 8 MB buffers, and they rely that OS flushes the caches before powering off the system. It is very much possible that those later drives simply have unwritten data in their caches when you power off your computer from DOS. This is the reason why XTIDE Universal BIOS disables write cache by default.

But BIOS from this system almost certainly won't disable write caching. Some software could be made to disable write caching when starting DOS or some parking program to flush caches could be used before powering off the system. Or at least reboot the system once with CTRL-ALT-DEL before powering off since drive reset should flush any unwritten data from the drive cache. There should not be any problems with period correct drives. They typically have 128k cache and they likely flush it more frequently. Those drives had to be DOS compatible.

Well, you could go to 32 MB in a 1996 system with Windows 95 B.
I've installed 32 MB in a standalone system for friends in 1996 with a Cyrix 6x86 P166.

I've preferred 32 MB / Cyrix P166 (133 MHz) over 16 MB / Pentium 150.

Due to the large RAM amount this system was more long lifed than expected because this was their first system that has been connected to the internet later.

Graphics card was a Matrox Mystique, but I've never seen any 3D game on this machine until its end.

Oh, it seems I'm a bit misplaced.

Your hardware seems to be more fitting to 1995 than to September 1996.
But I understand you have had to look on your budget due to your first summer job.