I had some fun this week doing a very basic analysis/benchmarking of my Intel 386DX/33 and 486DX/33 and 486SX/25 setup. I was primarily interested to investigate what does Turbo off do and how it compares to manually changing configuration of caches, CPU speed, etc. Another goal was to find a configuration that can run Donkey Kong at reasonable speed, hoping it can be accomplished with my 386DX/33.

Few findings:
- See the attached image for some detailed measurements
- Boot to MS DOS 6.22 with empty autoexec and configsys
- The only reliable way to run Donkey Kong is to set my 486DX to 2MHz and Turbo Off. Or Set it to 4Mhz and turn off L1. This is however not very practical as the setup of CPU MHz speed is done through 4 board jumpers. The computer also takes a good minute or two to boot to command prompt (with empty autoexec and configsys).
- I tried using SloMo, but it causes my system to hang, so need to figure how to overcome that, or use different utility
-486: Turbo Off with L1 On reduces performance to 1/3
- 486: Turning L1 Off reduces performance slightly more than just Turning Off Turbo
- 486: Turbo Off doesn't do anything when L1 is Off
- 486: L2 Off doesn't make any meaningful difference; my guess is that both LandMark benchmark and Donkey Kong fit into L1?
- 386 Turbo Off reduced performance to 1/2
- 386 Turning Off cache reduced performance to 1/2
- 386 Turbo Off and Cache Off reduced performance to about 1/4
- 386 Changing DRAM Wait state from 1 to 2 reduces performance by 1/3
- 386 Changing BUS Clock and Fast page Mode for DRAM doesn't impact performance in these tests

In the end I was little bit disappointed that I didn't manage to slow down the 386DX/33 enough to run Donkey Kong. Will have to try some additional slow down SW. I did try to also do a quick test with TestDrive 3, and it was nicely playable on 386DX/33 with turbo off, but way too fast on 486DX/33 with turbo off.
It was also interesting to see how important is L1 for 486 and cache for 386 - for these benchmarks.