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

So my 1996 Pentium back then was like this: […]
Show full quote

So my 1996 Pentium back then was like this:

• Pentium 120 MHz (very soon overclocked to 133)
• DataExpert EXP8661 motherboard with Intel 430VX chipset
• 16 MB RAM
• ExpertColor S3 Trio64 graphics card
• 1.2 GB Seagate HDD (believe me, this was HUGE compared to my previous computer)
• Keytronic keyboard and MS Mouse
• Forefront DH-1570 15” monitor

And from my previous computer:

• Logitech Soundman Games (very rare these days, I really wish I can found one again)
• Funai 2x speed CD-ROM drive + controller card with Panasonic interface (it wasn’t connected to sound card, not sure why. Maybe I still had Sound Blaster 2.0 when I got the CD-ROM-drive).

Wow, that's very similar to my system bought in Septemper 1995 (still in working order after 27.5 years):

• Original -> Current (if applicable)
• Pentium 120 MHz -> overclocked to 133 MHz
• 16 MB RAM -> added to 32 MB RAM
• Number Nine #9FX Motion 771 (S3 Vision 968 86C968) with 2 MB VRAM -> S3 Trio64V2/DX (86C775) with 2 MB VRAM
• No sound -> Sound Blaster AWE32
• No network -> Realtek RTL8139C NIC
• 1.08 GB Quantum Fireball -> 1 GB MMC card via CF adapter
• NEC MultiSpin CDR-273 4X CD-ROM -> AOpen 52X CD-ROM
• NEC MultiSync 15" monitor -> HP 15" LCD monitor (1024 x 768)

IMHO the motherboard (and CPU / RAM it supports) is the heart of "period correctness." OTOH disk drives, expension cards, monitors, etc. can be swapped for better experience and/or easier acquisition.

Yes, I agree that motherboard is the heart of period correctness. But it is not always so simple. How can it be period correct system if you use components that were not available when the motherboard was bought? Easy answer would be it can't but it is not so easy since computers can be upgraded. Then should I consider any PCI video card to be period correct, even those last Ati Radeons that were available on PCI? PCI Radeon would be very unlikely used to upgrade system like this. Perhaps Riva TNT, Voodoo 2 or Voodoo Banshee should be considered period correct since those are 1998 parts and might very well be used as upgrade parts back then. Maybe even PCI Voodoo 3 but I think that one is too new.

My original 1996 system was upgraded by changing the S3 Trio64 first to Virge and then to Hercules Stringray 128/3D (a Voodoo Rush card. I preordered it, what a mistake, but maybe more on that later). That wasn't the wisest upgrade path, I know... So in that sense I could very well use 1997 parts in this system I'm now building. And I would like to use more modern hard drive even though 120GB HDD can no way be considered period correct. But I'm not going to use 1998 parts here (but I'll test TNT for comparison) even though those are completely valid upgrade parts.

Anyway, I looked what 1996 HDDs I had.

Those slimline Seagates where introduced in 1995 but both of those are manufactured in 1996. Small (1.0 and 1.2 GB) but very good drives. That Quantum is 2.5 GB drive. It was surprisingly quiet considering its age. More quiet that PSU fan or CPU cooler (I'm going to replace those) or those Seagates. I also tested 3 GB Maxtor from 1998 and it was louder than any of the above drives.

Quantum had a problem (I got it few months ago along with some other stuff and I've not tested it before this). It was properly detected as 2.5GB drive but after POST, BIOS showed it as 1 GB drive. And it had partitions made accordingly. Looks like someone has limited the drive capacity for some reason. I need to find Quantum utility to restore it. But it was quiet enough so I could use this one if I'd really want a 1996 drive. But for the CPU tests I'm going to use the 1.2 GB Seagate.

dormcat wrote on 2023-03-29, 16:14:

[…]
Show full quote

• Original -> Current (if applicable)
• NEC MultiSpin CDR-273 4X CD-ROM -> AOpen 52X CD-ROM
• NEC MultiSync 15" monitor -> HP 15" LCD monitor (1024 x 768)

I think those two are downgrades and not upgrades.

aitotat wrote on 2023-03-30, 06:44:

dormcat wrote on 2023-03-29, 16:14:

[…]
Show full quote

• Original -> Current (if applicable)
• NEC MultiSpin CDR-273 4X CD-ROM -> AOpen 52X CD-ROM
• NEC MultiSync 15" monitor -> HP 15" LCD monitor (1024 x 768)

I think those two are downgrades and not upgrades.

While I like those two NEC products, CDR-273 was not the high-end SCSI MultiSpin series that gave NEC an excellent reputation at the early days of consumer CD-ROM. Rather, it was just an above-average generic ATAPI CD-ROM. It was a nice drive but once it stopped working after so many years it didn't stop me replacing it with slightly more "modern" drive with DOS support i.e. that I can find the official .SYS driver so it could be initiated in CONFIG.SYS.

The monitor was a different story: I bought the XP15 monitor in December 1995 for bloody NT$25,000 (one of the best monitor of its time) but sadly it didn't live long enough. It was replaced 5 years later with a Philips Brilliance 107P, then 107T2. After they all died in mid-2010's the computer had no dedicated monitor and had to cope with a retired Asus LCD (with a weird 1440 x 900 resolution and awful colors; my Mom bought it without consulting me). It was not until I found a nice old 4:3 HP monitor in an used electronics store and gave the retro rig a dedicated monitor.

Slow progress this week. I did test MMX 233 MHz yesterday. I also tested it with L1 disabled (MMX offers much more slowdown possibilities but I'll just test the MMX CPUs with L1 disabled for comparison with Pentium Classic). With 233 MHz and L1 disabled is seems the system is slow enough for Cycles and OPL3 to work. Indy Last Crusade did not work but then I noticed I didn't have the 486 patch like I did when I did testing with DX4 CPUs. Looks like from now on I should do these kinds of tests with and without the 486 patch.

This is becoming a boring CPU comparison. Since MMX 233 with L1 disabled is already slow enough, it simply means that the best 1996 CPU for this system would be Pentium Classic 200 MHz. It will be slow enough when needed and fastest of the 1996 CPUs. I already ordered one (even before the tests I did yesterday) since I couldn't resist getting one... (I almost hate when that happens and that have happened too many times this year!) I hope it arrives next week. But at least it might be interesting to test how 75 MHz bus changes things and of course how the Cyrix will perform.

And finally CPU benchmarks are complete. Here are the results.

There were surprises after all!

When L1 caches are disabled, MMX Pentium is faster than Pentium Classic with same clock speed. This was surprising since MMX Pentium was supposed to be Pentium with double L1 cache and MMX instructions. Clearly there are more differences than just that. Well, of course Pentium MMX has the debug registers for much finer slowdown than just L1 cache. All I tested was L1 cache for comparison with Pentium classic. But L1 alone gives very good results, actually even a bit better than with Pentium classic. MMX with L1 cache disabled is faster than classic but not too much.

Now the Indy Last Crusade 486 patch. I found that there is an update even after that. I call it a Pentium patch here. Looks like it works up to about 200 MHz just fine. Maybe it suffers from Turbo Pascal 200 MHz bug or similar? I tested them all, the original 1990 executable, 1991 486 patch and the "Pentium patch" that I think came with LucasArts Classic Adventures collection. Only CPU slow enough for the original release was Pentium 75 with L1 disabled (I'm sure MMX Pentiums can be slowed down enough but I did not test that).

Last surprise was Pentium classic 150 MHz. But I'll write about the CPUs more later.

Speedsys screenshot for MMX 233

MMX Pentiums first. Actually I only tested with a single MMX CPU (233 MHz). I under- and overclocked it for other speeds.

This was a really great CPU. It even overclocked to 262 MHz (3,5 x 75 MHz) without any issues and with default voltage! Just excellent. But It is too fast even with 233 MHz. The best CPU seems to be MMX 200 MHz. It does not suffer from the 200 MHz bug and disabling L1 alone gives excellent slowdown. Or alternatively 187 MHz (2,5 x 75 MHz) could be used for a bit better performance.

Of course 233 or even the 262 MHz is fully usable since MMX Pentiums have other slowdown options in addition to disabling L1. Certainly it could be slowed down just enough for the 200 MHz bug and some other setting in addition to disabling L1 could be used to find sweet spot for OPL speed sensitive games. But that is a little bit more work. 200 MHz is very easy. If you need a faster processor, then maybe K6-2+ or K6-3+ system would be a better alternative.

I would have chosen Pentium MMX 200 MHz if it would have been released in 1996. But no, it is (a very early) 1997 CPU.

Speedsys screenshots for 6x86

What about Cyrix 6x86? I have only one such CPU, IBM 6x86L pr200+. L means it is a low power (2,8V) version released in 1997. But otherwise it is the same as 6x86 pr200+ that was available in 1996.

This CPU is not for this system. It is simply the best CPU in 3DBench. Is that a 16-bit real mode benchmark? The others are 32-bit protected mode benchmarks. Maybe the excellent Cyrix 16-bit performance is what makes it so fast in 3DBench. Other than that it is slower than Pentium Classic 200 MHz in every other benchmark. And Quake performance if awful, barely faster than Pentium 90. Another interesting comparison is to compare pr200+ with P150 @ 75x2 (same bus, same CPU clock). Clock-for-clock, the Pentium is surprisingly close in all other benchmarks other than 3DBench and Quake.

But even bigger problem for this system is that with L1 cache disabled pr200+ is too fast! My guess is that only 16k primary cache gets disabled but 256-byte instruction line cache remains enabled. Pr166+ with L1 disabled is slow enough so it could be used in this system. But why? Pentium is simply a better alternative here.

I ran CHKCPU before running Speedsys. CHKCPU enabled CPUID instruction that is disabled by default. This is a compatibility thing to make 6x86 look like a 486 CPU since it is not fully Pentium compatible. Cyrix 6x86 is like a super fast 486 that works on a socket 7 motherboards.

6x86 is supposed to be faster on non-Intel chipsets and some BIOSes might tune the 6x86 better. So 6x86 might be faster on some other motherboards and there are many programs to tune 6x86. I did not test them here.

Pentium 75, 90, 100 and 120 MHz Speedsys screenshots

How good are the slow Pentiums? I wasn't going to test anything below P120 but then I started to wonder if any of them was slow enough for the original 256 color Indy3 without any patches. And one of them was. The only CPU that was slow enough was P75 with L1 cache disabled. It is the only Pentium that uses 50 MHz bus and that makes all the difference. I suppose I could have tested how 2x50 or 3x50 MHz CPUs do and maybe I will later.

Bus speed matters a lot. Pentium 120 MHz (2x60) is only a little faster than Pentium 100 (1,5x66). Pentium 120 is a good CPU otherwise since you can just set it to 133 MHz and it will almost certainly work at 133 MHz. If you really want a slow Pentium, then 75 MHz is the one to go with (because it is so slow with L1 disabled).

I didn't have 75 MHz Pentium so I used the 90 MHz one and underclocked it.

Speedsys screenshots for P133, P150 (2,5x60 MHz)

Looks like I forgot to save results from P166 and P150@2x75 MHz.

P133 is a good CPU. 66MHz bus speed helps here and makes it easily faster than 120 MHz Pentium. Pentium 150 is a disappointing CPU, at least with the default 60 MHz bus speed. On some benchmarks it is actually slower than P133! Pentium 150 should have been left unreleased. Unless you just change it to 2x75 MHz (same as Cyrix 6x86 pr200+). That makes it a completely different CPU! 60 MHz bus hurts a lot since with 75 MHz bus and 150 MHz CPU it is faster than P166. And this one works with 3,3V unlike the P166 I have. It requires 3,5V but I believe there are P166 CPUs that work with 3,3V.

P150 @ 2x75 is really a positive surprise. And since that CPU came with this motherboard, I would have kept it in this system if I had not ordered the P200.

P166 is a little disappointiment. It is easily faster than P133 (and P150 with 60 MHz bus) but it is no match for P150 using 75 MHz BUS. This CPU is already factory overclocked since 3,5V is required and this CPU cannot be overclocked at all. 187 MHz is simply unstable. I didn't test but this CPU should work @ 2x75 like the P150. Likely I could lower the voltage to 3,3V as well. It would give better speed but I wouln't feel comfortable underclocking it. That is why the P150 is a better CPU.

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.

@aitotat was curious what the special Virge card was? I’ve been messing around with Virge stuff lately, def peaked my interest. Thanks!

aitotat wrote on 2023-04-05, 17:19:

The best CPU seems to be MMX 200 MHz. It does not suffer from the 200 MHz bug and disabling L1 alone gives excellent slowdown.

Interesting findings. Out of curiosity, did you test multiple games to confirm that the MMX 200 doesn't suffer from the "Runtime Error 200" bug?

Good candidates are Jazz Jackrabbit (any version) and Tyrian (original release, not the Tyrian 2000 version).

Joseph_Joestar wrote on 2023-04-05, 18:32:

Good candidates are Jazz Jackrabbit (any version) and Tyrian (original release, not the Tyrian 2000 version).

I didn't test those now but I can do so. I did test Jazz Jackrabbit previously when I tried to build MMX 233 system. It worked although I expected it to fail. But I'll test those with MMX 200 and 233.

Chadti99 wrote on 2023-04-05, 18:15:

@aitotat was curious what the special Virge card was? I’ve been messing around with Virge stuff lately, def peaked my interest. Thanks!

ExpertColor Virge 325. I'll compare cards later but

Best of 1996 CPUs, Pentium 200 MHz

A good CPU and fastest of Pentium classics. Cyrix 6x86 is faster on some things but overall Pentium 200 was faster here. Quite uninteresting CPU otherwise. With L1 disabled it performs very close to all other Pentium classics with 66 MHz bus. With L1 disabled, it is the bus speed that matters. But that is a good thing since that means P200 is not too fast and can be slowed down nicely.

When compared to MMX 166, P200 is only a little bit faster. But runs hotter. It is another factory overclocked CPU that requires 3,5V and cannot be overclocked. I believe all P200s use 3,5V. It does not have 3,5x multiplier but it doesn't matter since even 225 MHz (3x75) is too much for this CPU. Quake tests did not finish with L1 enabled.

But some more performance can be gained. 187 MHz (2,5x75) is a little faster than 200 MHz. But not too fast so disabling L1 gives nice speed. I need to try if 3,3V could be used with 187 MHz. But maybe 200 MHz is the thing to have, after all, 200 MHz was the best in 1996.

Real world was 3 months behind "release" anyway. If however, you ordered vapoware, and if it stayed on schedule, then maybe you could order from Compaq or Dell or whoever was the favored anointed at that point, a month or so prior to official release, paying through the nose plus a kidney*, and get it within a week or two of release, but retail channels didn't get anything come through for ages.

(*because obviously you also needed the 64vs 32mb RAM, the 8x vs 4x CDROM drive etc or the 33mhz faster CPU wouldn't work... i.e. it was only available on the top end model.)

Tried to test the "200 MHz bug" with Jazz and Tyrian. I used Jazz Jackrabbit CD v1.2 and Tyrian v2.1. But they all work. I tested them with Pentium classic 187, 200 and 225 MHz. MMX Pentium also worked perfectly from 187 up to 262 MHz. Those versions are pulled from some collection and perhaps patched? (although file dates looks to be correct, 1994 for Jazz and 1996 for Tyrian).

Also tested another version of Tyrian but it was the same v2.1. I happened to find Jazz Holiday Hare 1995 shareware. It reported version 1.3. Those two I tested only with MMX 262 and they worked.

Can someone give me a link to shareware versions that are known to have the bug?

aitotat wrote on 2023-04-06, 07:53:

Can someone give me a link to shareware versions that are known to have the bug?

Jazz Jackrabbit shareware definitively has the bug. As for Tyrian, that game was made freeware by the original developers back in 2004. Versions 1.0 (shareware) and 1.1 (retail) both have the bug. Also, to test whether the bug is present in the versions that you have, you can run them on a faster system e.g. a Pentium III or something like that. It should immediately kick in there.

EDIT - here's a screenshot of both Jazz 1.1 (shareware) and Tyrian 1.0 (shareware) producing the error on my Athlon64 system:

You can find both shareware versions with a bit of googling.

I had little time today to do just one more quick test. I found and downloaded Jazz shareware 1.1 and tested it with MMX 262. And it works! I need to find both Jazz and Tyrian v1.0 next and test those. I'm going to test those on a faster system as well if they happen to work here.

aitotat wrote on 2023-04-06, 09:38:

I had little time today to do just one more quick test. I found and downloaded Jazz shareware 1.1 and tested it with MMX 262. And it works! I need to find both Jazz and Tyrian v1.0 next and test those. I'm going to test those on a faster system as well if they happen to work here.

Very intriguing! It seems that there might be a bit more to the "Runtime Error 200" issue than simply "CPU is over 200 MHz". I read somewhere that AMD K6 CPUs possibly need slightly higher clock frequencies to trigger this, but Intel CPUs were always described as susceptible to it.

Out of curiosity, I tested this on my Celeron 466, and it happens there as well. And to be doubly safe, I didn't select any sound card options in setup (left it at None) and the error still occurred.

EDIT - just checked the relevant Vogons wiki page and it implies that Jazz Jackrabbit running on a Pentium MMX CPU starts having issues above 266 MHz. Other CPUs like the Pentium Pro, Pentium II and AMD K6 appear to have different thresholds.