Another thing you can do to speed Doom up is reduce the screen size further in the options menu.

I think I used to play with one level of border around the screen on a 386SX33.

Well here's a short recording of DOOM on my 386DX40.

http://www.youtube.com/watch?v=mleD8AdqSdo

Its very choppy at high detail/full screen but at low detail with a couple of notches reduced screen size its fairly smooth.

that's actually not bad for a 386. i've played it much worse. 16mhz doom X_X

From what I can remember of a 486 sx-33 it was a decent machine to play doom on.

I have a beast of a 386DX-40, 64meg ram, 2meg cirrus logic 5434, enchanced IDE controller, I ran some benchmarks and these are a few observations.

DOOM, ET4000(1meg) vs CL5434(2meg) - exactly same framerate (7.8fps timedemo).
4meg ram vs 64meg - no difference

Quake,3dbench, exactly the same. pcpbench would'nt run on the ET4000 but did on the CL5434.

Doom will never run perfect on a 386DX40, if you run in low detail its quite playable but not 100% smooth.

Just tried mortal kombat, runs slow but smooth animation.

Its a myth that a 386DX40 is comparable to a 486SX, their is nothing wrong with your machine but if you want to improve it for dos, 8meg of ram but I doubt games that require 8meg will run good on the 386, and maybe a faster disk controller otherwise the ET4000 is as good as it gets, the CL534 is vastly superior if you want to run windows.

pcpbench would'nt run on the ET4000

you need to load UNIVBE first and then it will run.

Its a myth that a 386DX40 is comparable to a 486SX

In my own testing I found that my 386DX40 with ET4000 video is pretty comparable to a 486SX25 with crappy Trident video.
In real world usage at least (i.e. gaming). Most benchmarks will favour the 486SX, although funnily enough 3dbench is actually a couple of FPS higher on the 386.

If I put the ET4000 in the 486SX then it pulls ahead in most games.

Well the olnly problem is, that 386 does not have full speed L1 CACHE, and 486 does. L2 cache is just a bonus. And games and bencmarks that use many program code brances, are pretty even. But if you get some 486 optimized code that assumes 8kb L1 cache , then things change. And all those programs and games that lagged in 386, were released the time, where 486 was mainstream, and Pentium Classic was latest hype. Or even later.

Cyberdyne wrote:

But if you get some 486 optimized code that assumes 8kb L1 cache , then things change.

That makes a lot of sense...

We had a technique for evening up the odds in deathmatches. The best player had to play on the worst machine.

I think there is a bug with my version of Mortal Kombat in my 386. Maybe it's your problem too?
For example, things crawl soo much, but so much I can't even get in the menu without waiting for two minutes for the logo to fade. But it's so slow it's obvious there is some serious stall here, some kind of bug I can't identify right now (maybe I have to boot without EMS or without XMS or load factory settings in BIOS, will try later again)
MK1 runs like this, MK2 even crashes. But another reason could be that I have a Gravis. I am wondering if I could maybe install both a sound blaster and gravis in this machine and select. I think I don't have space too 😀

Just a little thing I remembered. Doom required a good videocard with good bandwidth between system memory and video memory. In the past, while I was trying some weird experiments to plugin a very old ISA gfx card in a very fast CPU (Pentium 3 class) and running various software, I realized that Doom wasn't very optimized in that aspect. The ISA gfx card could do maximum 40fps when doing 32bit copies to 320*200*8bpp but strangely Doom couldn't make over 10fps. I suspected it's copying data one byte at a time (8bit). I learned the reason later, that it's not using classic 13h mode but something similar to Mode-X because it needed to use page flipping. While there would be some advantages, I saw more disadvantages though and have been discussing the subject at some places.

Anyway, fact is a fast Pentium with very slow card would not give the best performance with Doom, while a slow 486 with fast VLB card could possible run fine. CPU is a bottleneck too but check your card. This thing with unoptimized copying to videoram (because Mode-X planar modes have weird pixel formation and makes it easier to just copy 8bit at a time that do a fast 32bit blit) seemed weird when I realized it and most people don't believe me 😀

Another solution though, I recently learn from the discussion on the link, the Mode-X engine wasn't always in Doom, it came possibly after v1.2. So if you can downgrade to an older Doom version there might be a good advantage if your gfx card is not good enough. I must try this soon even in my 386. I have also learned that the engine running in pure 13h mode before they changed it to use Mode-X alike engine was also used in Heretic and Hexen. This is the only I could try yesterday in an EEE asus PC booting with DOS USB. Ok, modern CPU for Doom and Heretic, only bottleneck the Videoram. Some 32bit copying to vram tests shows around 200fps maximum in 320*200*8bpp. So, this is the possible maximum. Running timedemos both in Doom later version and Heretic, Doom did around 60fps, Heretic around 200fps. This is a good evidence it's true. Gotta find a way to downgrade Doom version now and try this too. Also, in a 386 would I still gain something from that? (too much loss from slow CPU already)

p.s. I was getting crazy about this strange matter with Doom in the past? 😀

I remember back in 1993-94 when I had a 386dx/40 (actually the same board thar dirkmirk has right now), 4MB RAM and one very simple HMC 256kb svga card. I used to play Doom2 and Morkal Kombat 2. I thought they played fine on my 386, but it was just me being used to that. Later I saw my friend's 486/66 running Doom2, and it was much smoother than mine.
Anyway, a couple years later I upgraded to a Pentium 100 with 8MB RAM and a trident 8900 1MB card(which was already very old, but I got it very cheap from a friend to use it just as a temporary solution, so as to try some svga games). Then Doom played much better, and also the first Need for speed run at high resolution. It wasn't perfect, but playable, and certainly much better compared to a 486.
Finally, I remember at least a couple of articles in Greek computer magazines, reviewing 386/40 systems, and saying that such a fast 386 would be comparable to a 486, and in fact when used for programs that would rely much on the clock speed, it is faster than a (slower clocked) 486. But I don't know if that's true or not.