RetroPCCupboard wrote on 2025-07-24, 21:17:

I am sure many of you have built a similar machine. What kind or scores were you shooting for if playing WC1? Did you use a PCI or ISA video card? What model graphics card did you use?

I was able to reach 386 speeds with a PCI Matrox Millennium II. You can see some of my tests here, and my slowest result here.

From my experience, if your motherboard can downclock a Pentium MMX CPU to either 133 MHz (2x66) or just 100 MHz (2x50) with both L1 and L2 caches disabled, you should be able to reach slow 386 speeds without any trouble. If necessary, you can stack the SetMul BPD and VPD parameters on top of that for some extra slowdown. Additionally, loosening the memory timings in the BIOS might make the system even slower.

Joseph_Joestar wrote on 2025-07-24, 21:34:

RetroPCCupboard wrote on 2025-07-24, 21:17:

I am sure many of you have built a similar machine. What kind or scores were you shooting for if playing WC1? Did you use a PCI or ISA video card? What model graphics card did you use?

I was able to reach 386 speeds with a PCI Matrox Millennium II. You can see some of my tests here, and my slowest result here.

From my experience, if your motherboard can downclock a Pentium MMX CPU to either 133 MHz (2x66) or just 100 MHz (2x50) with both L1 and L2 caches disabled, you should be able to reach slow 386 speeds without any trouble. If necessary, you can stack the SetMul BPD and VPD parameters on top of that for some extra slowdown. Additionally, loosening the memory timings in the BIOS might make the system even slower.

Thanks for the reply. Your findings are interesting. I haven't tried any games yet. I do have a Matrox Mystique 220, which I guess will be similar to your millennium II. I haven't used it in any builds due to reported DOS incompatabilities. So interesting to see that those issues only happen at certain speeds.

Basically what I am trying to do is cover thr 386-486 ranges with two machines. I have the 100Mhz system I already described, and also a 300Mhz Pentium MMX system. My thinking is that I want to avoid the need to reboot to change speed. Which means I don't want to manipulate the L2 cache (which can only be done through BIOS).

So I have benchmarked my existing 300Mhz system with L2 cache enabled using the PhilsComputerlab 136in1 project sheet. And I am comparing that to various configurations of Pseudo 386 that I am experimenting with. L2 cache is disabled on this machine. For reference, these are scores expected for some reference systems:

386-25 = 10.8
386-33 = 13.9
386-40 = 16.0
486-25 = 20.9
486-33 = 28.0
486-66 = 46.2
486-100 = 70.1
486-133 = 79.9

This is my results:

And as a line chart:

Zoomed in to the 486-66 and below:

So, based on these results, I can see that EDO vs SDRAM makes a slight difference, but not much. Negligable once down to 386 speeds.

Video card makes a larger difference. The Cirrus is about 25% faster than the Trident at full system speed (equivilent to about 486-33mhz) but when you get down to 386 speeds, its less.

With these bemchmark results I am thinking the psuedo 386 will use Trident CPU and 8mb EDO. This will lead to the following performance characteristics for the two machines:

Neither system will cover the 486-100 and 486-133 range, and not sure it that is a problem. The pseudo 386 looks like it will be able to drop to the speed of a fast 286 with this configuration. It can get close to 25mhz and 33mhz 386, but not 40mhz.

If you need this capability then it's probably best to just use 86Box. I found my PIII 900 to be too slow with caches disabled. Slowing down to 486-66 is probably best as 386-33 is too slow.

There's no getting around toggling the L2 cache on/off in the BIOS. If you need slow 386 boards just don't support changing that at runtime.

Depending on your motherboard and case. If you have a turbo button. It might be possible to have that change a single CPU multiplier jumper. But again you'll need to reboot. Slow 386 ran DOS, it's not trying to boot into a full Windows environment which will just take forever.

486-100 and 133 aren't a thing that games ever targeted, 486 33 and 66 are what you need for that era, then you jump to slow Pentiums. If you want to do benchmarking of games with different configurations and seeing how things actual run, then you need real hardware. Having a single system that hits the different performance levels is to allow you to run games that have a variety of speed requirements on a single system.

SScorpio wrote on 2025-07-25, 15:26:

There's no getting around toggling the L2 cache on/off in the BIOS. If you need slow 386 boards just don't support changing that at runtime.

Yes, that is why I am building two systems. One with cache on, and one with cache off.

The one with L2 cache enabled will be used to target games from slow 486 up to fast Pentium MMX. The one with L2 cache disabled will target slow 386 to slow 486.

SScorpio wrote on 2025-07-25, 15:26:

Depending on your motherboard and case. If you have a turbo button. It might be possible to have that change a single CPU multiplier jumper. But again you'll need to reboot. Slow 386 ran DOS, it's not trying to boot into a full Windows environment which will just take forever.

486-100 and 133 aren't a thing that games ever targeted, 486 33 and 66 are what you need for that era, then you jump to slow Pentiums. If you want to do benchmarking of games with different configurations and seeing how things actual run, then you need real hardware. Having a single system that hits the different performance levels is to allow you to run games that have a variety of speed requirements on a single system.

Yes, which is why I want two machines. One primarily a fast Pentium MMX running Win95, but can be slowed to 486 speeds whilst in DOS mode without needing reboot. The other will have DOS 6.22 and probably Windows 3.11, and is just for DOS or early pre-9x Windows games.

AlexZ wrote on 2025-07-25, 15:07:

If you need this capability then it's probably best to just use 86Box. I found my PIII 900 to be too slow with caches disabled. Slowing down to 486-66 is probably best as 386-33 is too slow.

With Pentium II and III, that's correct. Pentium MMX is more flexible in terms of granularity of slowdown. 386-33 is too fast for Wing Commander 1

Slowing down a system depends a lot of the memory type, FPM or EDO, and 60 or 70 nanosecs, video card type: ISA, PCI, AGP
I tried slowing down few retropc for playing Test Drive 3: The Passion.
For a Pentium 100 with ISA Trident TVGA9000, I reached 6.6 score in 3d Bench:
https://youtu.be/JRvI4yQMh-U?feature=shared&t=713
And for a Pentium II 266 I reached 4.9 score. Very interesting about this one, an IBM PC 300GL, with AGP integrated video card. AGP is linked directed to the CPU, so it will slow down more then with an ISA or PCI card:
https://youtu.be/F4_zILdYxAA?feature=shared&t=3117

b0by007 wrote on 2025-07-25, 20:05:

Slowing down a system depends a lot of the memory type, FPM or EDO, and 60 or 70 nanosecs, video card type: ISA, PCI, AGP I trie […]
Show full quote

Slowing down a system depends a lot of the memory type, FPM or EDO, and 60 or 70 nanosecs, video card type: ISA, PCI, AGP
I tried slowing down few retropc for playing Test Drive 3: The Passion.
For a Pentium 100 with ISA Trident TVGA9000, I reached 6.6 score in 3d Bench:
https://youtu.be/JRvI4yQMh-U?feature=shared&t=713
And for a Pentium II 266 I reached 4.9 score. Very interesting about this one, an IBM PC 300GL, with AGP integrated video card. AGP is linked directed to the CPU, so it will slow down more then with an ISA or PCI card:
https://youtu.be/F4_zILdYxAA?feature=shared&t=3117

Yes, true, memory timings can make a difference. I am running mine at 70ns. ISA vs PCI seemed a huge difference at higher speeds but less so when you bog the CPU down by disabling caches etc. My Pentium MMX 300Mhz system is running an AGP card.

I find it interesting how different CPUs respond to disabling their cache. Results can be wildly different. I find it very surprising that a PII or PIII is slower than a Pentium MMX when caches are disabled. But the results don't lie.

Architectures always trade efficiencies.

The p2 has a longer pipeline, so a cache miss costs more clock cycles.
Since you're disabling the cache, every read is a cache miss.

Memory also often trades latency for bandwidth.
The p2 uses 64bit memory, and Iirc it sends more than requested - to fill the cache pre-emptively, which is disabled. So it's sending a lot of extra junk that has to be sorted though and that probably has a cost too. Imagine sending a full shipping container with only one parcel you want.
So you're requesting 8 bits, getting... I think 4x64, and dumping most of it.