You cannot run later Dos games at a satifactory fps on a dx2-66. Best case scenario is to do L1 and/or L2 cache disabling on the 486
for games like Wing Commander and other older Dos games that run too quick on a 486. Of course the 486 will run the majority of Dos
games ok.
You really still need a Pentium for games like Duke Nukem 3d, Terminal velocity, Quake Dos, Screamer, etc.

EDIT. Should have read Your post better. The 286 will cover the older software base fine.

So generally I can use 286 for old games, 486dx2 for most DOS games and Pentium+ for the most demanding DOS and Windows 98 games?

You can use a K6-2 or PMMX to cover a performance range from 386DX to Win98 era using cache disabling and multiplier switching. A 286-16 with turbo switch will cover everything down to an XT.

gdjacobs wrote:

You can use a K6-2 or PMMX to cover a performance range from 386DX to Win98 era using cache disabling and multiplier switching. A 286-16 with turbo switch will cover everything down to an XT.

I just can't be bothered with cache disabling and stuff like that. That's why I'm looking for a solution that will let me use 2-3 systems and be done with it.

Well, you can do it in software with the K6-2+ and K6-III+ CPUs (using SETMUL). I wouldn't be too concerned about it being complicated.

Multiple systems will do the trick just fine, though. If you have the space, more retro is a good thing. 😀

Hmmm

Disabling caches through BIOS / command line vs. building 3 PCs 🤣

If you go with a MMX 233 you can have:

386DX, 486DX2 and MMX 233 in a single machine 😀

PhilsComputerLab wrote:

Hmmm […]
Show full quote

Hmmm

Disabling caches through BIOS / command line vs. building 3 PCs 🤣

If you go with a MMX 233 you can have:

386DX, 486DX2 and MMX 233 in a single machine 😀

These days I find the P233 MMX too slow for games such as Quake. It will do 45 FPS in Quake at best and also trigger the Pascal compiler speed bug in many games.

I would build these 3 systems. 286-16 - 386DX40/486SX25 - K6-2/2+/3/3+ @~400 MHz (60+ FPS in Quake) to avoid the Pascal compiler speed bug issue. For SVGA gaming it could also be worth having fast P3 DOS system.

edit

I noticed I forgot the Windows 98 system, P3 1000 or faster with Voodoo 2 SLI + Geforce 4 Ti 4x00.

I struggled with exactly the same problem many years ago.
Since you're not interested in cache disabling, I'm not going to provide any suggestions around that, but perhaps just check out Phil's video on what you can do with a K6-2 in terms of achieving various levels of performance (if space becomes a problem, then this is a good alternative).

I started out with a Pentium 233MMX and tried to use that as my main DOS machine - as expected, it was way too fast and didn't even have a turbo switch.
I then experimented with various 386 & 486 CPU's and finally decided on an AMD 486DX4 100 MHz (the older NV8T derivative with write through cache).
I found that this CPU allowed me to play the majority of the DOS games that I wanted to play and, with the turbo switch, slows down to more or less the equivalent speed of a 486DX 33 MHz.

I have a Celeron 900 MHz that I use as my main "high performance" DOS & Windows 98SE PC (I've set up a "dual boot system", essentially just a start up menu, for this purpose).
Generally, it's fine for most games between 1996 & 2001 that runs on DOS or Windows 98SE, but I think this PC might a bit too fast for "late" DOS games or "early" Windows 9x games (that were released in late 1995 to early 1997 when your Pentium 1 was still the fastest CPU around). So, I'm considering building up either a fast Pentium 1 or a Pentium II based system for this purpose (as luck would have it, I do actually have a K6-2 450 MHz all ready to go).

Although I rarely play XT based games or games that demand a slow 386 CPU, I have been tinkering with the idea to built up a slow PC as well (just for the fun of it, since I enjoy tinkering with the hardware as well)
Unfortunately, the slowest PC I currently have is an AMD 386SX 25 MHz - does anyone have any experience with this CPU in terms of its "de-turbo" mode and how much it slows down? Can it get to XT speeds?

Half-Saint wrote:

More of a general question but would it make sense to use a 486DX2-66 for all DOS games and a Pentium (or better) for Windows 98 games?

Late Win9x games feel better on high P3.
Some DOS games will have problems on 486 66 MHz and you'll need set is slower. The good thing is DOSBox runs ok most DOS games, even on P3. While Win9x games emulation is still a problem.

There shouldn't be much difference between a 386DX-40 and a 486DX2-66, especially since the DX2 also comes with the turbo feature and can be slowed down.

386 may be slowed down more, but this possibility is useful mostly for CGA games.

gdjacobs wrote:

You can use a K6-2 or PMMX to cover a performance range from 386DX to Win98 era

K6-2 and K6-3 are slow for late Win9x games, beginning from ~1999.

I have got 3 machines that run pure DOS, a 386SX 25 MHz, that does all the old stuff without problem. A 486 DX/2 66 MHz, nice middle ground. Finally a AMD K5-166, that does everything else, it is probably the most versatile or the 3 and is certaily the most used.

Well, I already have a P5A-B based Super Socket 7 machine w/ AMD K6-2+/550ACZ and GeForce 2 MX 400.

This is one of the reasons I am working on compiling numbers for as many CPUs as possible with L1 cache disabled:
Is there a benchmark list showing results with L1 disabled?
Wouldn't it be nice to look at one chart and see that, for example, a Pentium 120 perform like a 20Mhz 386? SETMUL is really simple. Just make a batch file for a game if you want to use it. Example, name your batch file go.bat and the contents is:
setmul l1d
wc1.exe
setmul l1e

I've done charts like this many years ago 😁

The motherboard, memory timings and graphics card also matters a little. CPU architecture matters the most though.

Using an ISA card is a trick to make it even slower. What a nice graph will show are the "gaps", areas that you cannot reach with the cache tricks.

PhilsComputerLab wrote:

I've done charts like this many years ago 😁

The motherboard, memory timings and graphics card also matters a little. CPU architecture matters the most though.

Using an ISA card is a trick to make it even slower. What a nice graph will show are the "gaps", areas that you cannot reach with the cache tricks.

Maybe I'm the only one who is interested in this. 🤣. I just think it would be nice to have something documented and searchable, rather than depending on word of mouth. Well, I will try to do this. Perhaps my CPUs will be the only ones on the chart, but I'll give it a go.

I would have thought Phil's VGA Benchmark covers all this ground TBH https://docs.google.com/spreadsheets/d/1lvF9n … wpNU/edit#gid=0 , it has over 650 entries now so you can easily find a configuration for your needs. From experience, a Realtek ISA VGA card will make any socket 7 machine behave like a slowish 486, regardless of CPU fitted, see entry 63. So prehaps one of those and a good PCI VGA card in the same machine and problem solved without messing with cache or multipliers etc. I've not actually tried this, in fact just thought of it, but might work? Can you have a ISA & PCI card in same system? never tried.

I'd much rather have a simple command line entry than mess around with hardware that will probably affect the system in other ways not to mention the crappy video quality of old ISA cards.

BSA Starfire wrote:

I would have thought Phil's VGA Benchmark covers all this ground TBH https://docs.google.com/spreadsheets/d/1lvF9n … wpNU/edit#gid=0 , it has over 650 entries now so you can easily find a configuration for your needs. From experience, a Realtek ISA VGA card will make any socket 7 machine behave like a slowish 486, regardless of CPU fitted, see entry 63. So prehaps one of those and a good PCI VGA card in the same machine and problem solved without messing with cache or multipliers etc. I've not actually tried this, in fact just thought of it, but might work? Can you have a ISA & PCI card in same system? never tried.

Phil's VGA Benchmark is outstanding! I have many entries in it myself. But all the entries are with L1 cache enabled. I am looking at making a chart specifically for 486 and higher processors that shows performance with L1 (and in some cases L2) disabled. So if someone is looking at targeting a specific slow speed by disabling cache, they can consult this chart and say (for example), "I want a Pentium 120 because with cache enabled it's perfect for the late era DOS games I want to play, and with cache disabled, it's perfect for 386/20 games that I want to play."
It would be neat to see where all the processors fall in performance when you disable cache, especially some of the Cyrix and AMD CPUs, all in one list.

I envision having a number of "reference" benchmarks scattered throughout. So maybe I could take the average 386/40 from Phil's chart and place that into this chart so someone knows what a typical 386/40 scores in specific benchmarks. And the average 486/25, etc.

Hopefully that makes more sense.

clueless1 wrote:

Phil's VGA Benchmark is outstanding! I have many entries in it myself. But all the entries are with L1 cache enabled. I am lo […]
Show full quote

BSA Starfire wrote:

I would have thought Phil's VGA Benchmark covers all this ground TBH https://docs.google.com/spreadsheets/d/1lvF9n … wpNU/edit#gid=0 , it has over 650 entries now so you can easily find a configuration for your needs. From experience, a Realtek ISA VGA card will make any socket 7 machine behave like a slowish 486, regardless of CPU fitted, see entry 63. So prehaps one of those and a good PCI VGA card in the same machine and problem solved without messing with cache or multipliers etc. I've not actually tried this, in fact just thought of it, but might work? Can you have a ISA & PCI card in same system? never tried.

Phil's VGA Benchmark is outstanding! I have many entries in it myself. But all the entries are with L1 cache enabled. I am looking at making a chart specifically for 486 and higher processors that shows performance with L1 (and in some cases L2) disabled. So if someone is looking at targeting a specific slow speed by disabling cache, they can consult this chart and say (for example), "I want a Pentium 120 because with cache enabled it's perfect for the late era DOS games I want to play, and with cache disabled, it's perfect for 386/20 games that I want to play."
It would be neat to see where all the processors fall in performance when you disable cache, especially some of the Cyrix and AMD CPUs, all in one list.

I envision having a number of "reference" benchmarks scattered throughout. So maybe I could take the average 386/40 from Phil's chart and place that into this chart so someone knows what a typical 386/40 scores in specific benchmarks. And the average 486/25, etc.

Hopefully that makes more sense.

Oh! I get it now! can be a bit slow me! 🤣 I can probably help out with scores with Cyrix and AMD socket 7 chips & winchip(those are really my thing). Let me know if you want a hand.
Best,
Chris

Thanks, Chris! When I nail down a format, I will start a new thread with my scores pre-populated, then solicit help from you and anyone else interested so we can grow the list. 😀