Disable the L1 and L2 cache in the BIOS and it will perform like a fast 386DX-40 or slow 486DX25.

There are many games that won't run properly unless you disable the caches.

You could probably fiddle with jumper settings and get it down to a P120 or P100 as well

I started my DOS nostalgia trip on a 233MMX and it's a good option for later ('94+) games, but too fast for some early stuff. Even Syndicate and Dune2 have given me some trouble on a Pentium.

You can mess around with the cache, etc but that's far too easy! Having the period correct hardware is heaps more fun.

It seems to handle most of the games I have played on it - Doom, Doom II, Radix, Duke Nuke 3D, and Terminator Future Shock. I am not sure of the dates on those.

The last one (Terminator) was the funny one and the reason I built the box in the first place. I found a spot I cold not get past running it on a pentium 3/windows 98 box. I think it is related to processor speed, not not wndows versus DOS. It is still a little touchy at 233 but you can get through it.

I did think about reducing the mutiplier - but aren't those locked on Intel Chips?

I think the best CPU for DOS gaming is an AMD P75.

Not on these early chips! You can also reduce the FSB / Bus speed if your mainboard has that option...

You should be able to set it to 133, 166 and 200 MHz.

many dos games need 8088 4MHz

The main problems with faster processors are:

- Games that do not adjust their speed or have fixed settings. Those games were designed to run in a 8086 at 4.77 o 8 MHz, or allows you to select from some fixed speeds of that time (like 80286 at 12 or 16 MHz). A 386 or better processor will be too fast for them, so try using moslo or another slowing program.

- Games that crashes when using faster processors (Jazz Jackrabbit, for example). Mainly, they crash with an "error 200". The problem is a library from Borland compilers that causes a zero-division error in fasters processors. There are some programs that can patch those programs to run (or, if your processor is slow enough, disabling the cache may run those programs without patching). There is another version of that error found in Microsoft libraries, but I don't remember what error was (but finding a solution may be easy).

The rest of the programs will auto adjust the speed with "fixed" events (like vertical retracing in VGA), so they won't run faster in your computer. I used DOS (MS-DOS 6.22) until my AMD Athlon at 1 GHz and had almost no problems; in later systems I used DOSBox.

It all really depends on what DOS game you would like to play on it, actually. Especially games that depend on the cpu cycles and are related to it (CPU) proportionally and indefinitely.

Later DOS games will run comfortably in faster systems.

If you're ask me, I'm perfectly alright with a P233MMX. But again there comes the question - "perfectly alright for what?". 😁

For earlier games, I prefer to run under Dosbox - you can fine tune the speed optimally.

For resource hungry dos games, (mostly the last batch of the final few years of DOS games before Windows9x killed them), it's better to have a reasonably "fast" DOS-based system.

I remember when I had a Pentium 133, my friend compared the in-game mission and world loading times of Interplay's Conquest of the New World - and he was amazed that it was "significantly shorter (the loading time) in my P133 vs. in his friend's 486DX2 based system.

Other games that might benefit from "fast" systems are Flight Sims and Racing Games. Turn based strategy games will also benefit from faster systems, since when playing against the computer, the time taken for the computer to "think" the next move, will be substantially shortened and will result in smoother and more convenient gameplay.

Besides my actual 486DX2-66, I also have a PII400 - which I use for general gaimg and for later DOS games (and Windows 95 games too).

DOOM and Descent play much more smoothly in my PII-400 when compared to running in my 486.

Zup wrote:

- Games that crashes when using faster processors (Jazz Jackrabbit, for example). Mainly, they crash with an "error 200". The problem is a library from Borland compilers that causes a zero-division error in fasters processors. There are some programs that can patch those programs to run (or, if your processor is slow enough, disabling the cache may run those programs without patching). There is another version of that error found in Microsoft libraries, but I don't remember what error was (but finding a solution may be easy).

It was either this or a similar issue that I was warned about before. I forget the full details, but iirc there's an issue with several (otherwise non speed-sensitive) games where the inner loops (in cache) run too fast on processors above ~400 MHz.
Disabling cache can solve this issue, but may make the system too slow for some games. Disabling just L1 on a PII or Socket 7 over 400 MHz may also work at a more acceptable speed though. (Katmai PIII should be similar, and on a K6-III/+ you'd need to disable L1+L2 and leave only the board-level cache)

Albeit, with socket 7 parts, you could also just use a lower bus or multiplier setting.

There's also a separate issue a few games have with sound issues on faster CPUs. The floppy disk version of X-Wing plays garbage notes (or nothing) in adlib/SB with CPUs much over 133 MHz. Apparently there's a few other games like that but I don't know any off-hand.

"High-end" DOS rigs often have other limits as well. From a Pentium III upwards, you start losing control over the ISA bus timings. Some PIII have these BIOS options, some don't.
You might, of course, find software solutions, BIOS patches etc. to tune 8-bit and 16-bit I/O recovery (delays). This is sometimes needed for very old ISA cards.
The upper memory area (640k to 1 MB) is usually smaller and more fragmented as well, you might find that you cannot load all your TSRs in high and upper memory.
If you use a PCI sound card, the emulation software will have its own speed limitation. For example, the SB Live!'s emulation works correctly between cca 200Mhz and 800 Mhz (I cannot remember the exact numbers) on a PIII. This utility depends on EMM386.EXE, which again has speed issues...

I'm starting to think my 233 is about right. Sure it's a little fast for DOS, and the hard drive is a little big (6.4 gig) but I don't mind things being a little souped-up, That drive (ST34622A) is just plain better - faster, quieter, and of course bigger than the 1275 Mb Conner I had in my original 486....

Hi, this is my first post here, apologies in advane if it is in the wrong place. Does anybody know if I could use an intel atom machine to play dos games? Is it possible to disable the cache?
thanks

ncmark wrote:

So - how fast is too fast for DOS?

Anything over a pentium 2 would probably be too fast. I think the pentium 2 was actually designed to accelerate 16-bit code which DOS is mostly made of. I think anything more than a pentium 2 would be too much. The pentium 3 for instance, has sse instructions, and I don't think any DOS games use sse instructions.

"too fast" depends entirely on what you want to do. For games like Duke3D, there's no such thing as "too fast".

Atom would probably work fine, as long as you have an ISA slot for a sound card.

Holering wrote:

ncmark wrote:

So - how fast is too fast for DOS?

Anything over a pentium 2 would probably be too fast. I think the pentium 2 was actually designed to accelerate 16-bit code which DOS is mostly made of. I think anything more than a pentium 2 would be too much. The pentium 3 for instance, has sse instructions, and I don't think any DOS games use sse instructions.

This is an old thread so I think the OP's question is no longer relevant.
Still, you raise some interesting points.
The Pentium Pro was faster than the Pentium in 32-bit code, but had weaknesses which resulted in slower performance in 16-bit code (same clock speed) compared to Pentium. The Pentium 2 included some enhancements so its performance didn't suffer as much with 16-bit code (still a bit slower than Pentium (MMX?) at the same clock speed).
I'm pretty sure that there are no DOS games that use SSE instructions, although technically that would be possible. But instructions that are not used are not a problem - most DOS games didn't use MMX either (did any?), and both the Pentium MMX and the Pentium 2 have those. Even x87 instructions weren't commonly used before Quake.

Topic maybe too old, but I still try. In more modern CPUs like PI to PIII turning off both caches (CPU Cache and Motherboard Cache) will turn your CPU to 8086, right? So I have a question. Which settings turn my CPU to a 386? Exactly to 386 not 8086? Motherboard Or External Cache Off and CPU Cache On? What about CPU Cache Off and Motherboard or External Cache On? Maybe it turns my CPU to a 286?:). Cause the slowest CPU I have is PI-166MMX, I have 486DX2, but I haven't motherboard to it.

mr3dx wrote:

Topic maybe too old, but I still try. In more modern CPUs like PI to PIII turning off both caches (CPU Cache and Motherboard Cache) will turn your CPU to 8086, right? So I have a question. Which settings turn my CPU to a 386? Exactly to 386 not 8086? Motherboard Or External Cache Off and CPU Cache On? What about CPU Cache Off and Motherboard or External Cache On? Maybe it turns my CPU to a 286?:). Cause the slowest CPU I have is PI-166MMX, I have 486DX2, but I haven't motherboard to it.

There is no one set of settings that will cause all cpus to perform at the same level. Generally, the slower a cpu's native speed, the more it will slow down when you disable caches. That applies up to about P55C. Pentium II and III are more dependent on caches, so they actually tend to slow down a lot more when disabled. In many cases, they slow down too much. Check the thread in my signature about disabling caches. People here have contributed results for many different cpus precisely so that questions like your can be answered. 😀

As clueless1 said. Best you can yo is to try to find out 386-ish speed by running some old benchmarks/diagnostics tools.
For example, Central Point's System Information can show how your system compares to others.
But keep in mind this doesn't resemble real-world performance.
Some of these programme do measure speed by measuring how many
cycles some instructions need to perform a given task.