Gahhhrrrlic wrote on 2026-04-24, 01:25:

Hey everyone. I recently broke out my old 386 to play with it and started reminiscing about the past. One thing led to another […]
Show full quote

Hey everyone. I recently broke out my old 386 to play with it and started reminiscing about the past. One thing led to another and I found myself trying to reprogram doom to use the FPU as a co-processor. Crazy right? I just wanted Doom to play well on my 386... but I don't have to tell you guys what motivates us to do seemingly impossible things 😉

Well that downward spiral lead to nearly a month of being absorbed in doom's source code, re-learning assembly for the first time since university and plugging away at hot loops and micro-optimizations. Long story short, I believe I have made an improvement. I started with Viti95's 0.9.7 build of FastDoom as it's the fastest and most faithful build I am aware of and managed to optimize some of the code and offload work to the FPU.

The main problem I'm having is my hardware isn't fit for doing benchmarking. It's not a pig... but it's not great either. I'm severely bottlenecked by the only graphics card I own. I don't have particularly fast ram or cache (or much of either) and I only have a couple of FPUs to try. I've been mostly standardizing my work by means of Dosbox but I feel my efforts have fully matured and now is the time to move from code to hardware, to really see if all the work made any difference.

I'd like to request the assistance of a colleague who is better equipped than I, to run some of the EXEs I've compiled, to see if lifting the bottlenecks I have would make a difference. Specifically I'm hoping for:

AMD-DX 40
At least 128kb of fast L2 cache (I only have 64)
8mb of fast ram preferably with 0 wait states
A screaming video card! Something that is faster than the ISA bus
A selection of FPUs including Cyrix or Green (I have tried ULSI and IIT)

I think that ought to do it. The 386 is primarily limited by the video bandwidth (basically the isa bus) so I'm very curious to see what would happen if that were maxed out. My gains are all at the low resolution end and disappear quickly at max res.

If you are willing to help benchmark a few EXEs on your rig to validate the changes I've made I'd be most appreciative. You can respond here or PM me. Thank you very much!

Until 486 DX33 or similar really the limit is CPU not ISA bus, I have a 486DLC33mhz, tested with VLB and no difference, tested a intel 486DX40 too and VLB vs ISA was still not a big difference. When i started to see a big CPU waste is for example a DX2-66, i tested VLB vs ISA and was a timedemo demo3 in low, 62fps vs 39fps, this is like 40% CPU wasted, and my video card is a fast one, if not fastest, Trident 8900D 0WS (im talking isa stock speed no overclocked)

About my computer is 486DLC33, 256kb cache, 8mb 60ns 0WS, Cyrix 487, 8900D 0WS

Any optimizing for doom is welcome, since even fastdoom is slow for today standard in this computers

I can help benchmark. I have multiple VESA local bus 386 motherboards and a decent selection of fast ISA and local bus videocards.

I can help as well if needed. Specs:

AMD 386DX (40MHz)
IIT 4C87DLC-40 (FPU)
PCChips M321 (128KB cache)
Tseng ET4000AX 2theMax 4000S (ISA)
8MB ram

Don't know if the ET4000AX is a screaming video card though. 😀 Just PM me if needed.

I have a 386 host that I can use for testing. The configuration is as follows:

CPU: AMD 386DX-40 & Cyrix 486DLC-40 (switchable)
FPU: ULSI US83C87-40
Motherboard: FIC 386-VC-H (256KB L2 cache)
Memory: 32MB
Graphics Card:
CL-GD5429, 2MB Video RAM, ISA
Trident 8900D, 1MB Video RAM, ISA
Trident 8900D-R, 1MB Video RAM, ISA

Wow thanks very much for the quick replies! It means a lot that you're willing to help. Already from reading your specs I feel like a kid in a candy store 🤣.

Just a follow-up question for TheElf and Zwrr:

TheElf: Do you happen to have any conventional 386 core cpus? The DLC is an enticing bench option because it represents the end of the road for 386 rigs but I wouldn't want to exclude the 386 itself as that was the target I was going after. I could never get mine up to 40mhz either without risk as my crystal is soldered on.

ZWRR: Do you have any other FPUs such as the Cyrix Fasmath? I started my tests with the ULSI, seeing gains but was quite surprised (and dismayed) that the IIT chip was actually slower than no FPU at all. I realized that each vendor's architecture really did have a personality of its own when it comes to what kind of code it prefers so I'd like to know if the FasMath does as well with Doom's calculations as it does in synthetic tests. If so, that'd be a reason to own one 😉

I'd be happy to reimburse for any additional hardware if that's required (within reason of course heh). I'm a big fan of the guy from CPUGalaxy on Youtube as he has all the FPUs and can probably find a niche for each of them.

I have a 386 system with the cyrix fasmath fpu

The CPU is a cyrix 486dlc at 50 mhz, I think the system was originally a 386 w 25mhz bus speed. The game performance is generally akin to a 386 at 40mhz, and is much slower than a true 486 at a similar clock speed. It also has 128k l2 cache.

If the unusual cpu doesn't muddy the waters too much I could do some testing.

Gahhhrrrlic wrote on 2026-04-24, 23:51:

Wow thanks very much for the quick replies! It means a lot that you're willing to help. Already from reading your specs I fee […]
Show full quote

Wow thanks very much for the quick replies! It means a lot that you're willing to help. Already from reading your specs I feel like a kid in a candy store 🤣.

Just a follow-up question for TheElf and Zwrr:

TheElf: Do you happen to have any conventional 386 core cpus? The DLC is an enticing bench option because it represents the end of the road for 386 rigs but I wouldn't want to exclude the 386 itself as that was the target I was going after. I could never get mine up to 40mhz either without risk as my crystal is soldered on.

ZWRR: Do you have any other FPUs such as the Cyrix Fasmath? I started my tests with the ULSI, seeing gains but was quite surprised (and dismayed) that the IIT chip was actually slower than no FPU at all. I realized that each vendor's architecture really did have a personality of its own when it comes to what kind of code it prefers so I'd like to know if the FasMath does as well with Doom's calculations as it does in synthetic tests. If so, that'd be a reason to own one 😉

I'd be happy to reimburse for any additional hardware if that's required (within reason of course heh). I'm a big fan of the guy from CPUGalaxy on Youtube as he has all the FPUs and can probably find a niche for each of them.

I have 386SX40, 386DX33 and 386DX40 motherboard in collections,they are just in shelf collecting dust, i dont have any interest in 386, back on time, i jump from 286 to 486DLC directly

Forget, I have a 486SLC40 too

Oh i add something in all my test, latest one is fastest, no 0.9.7

theelf wrote on Yesterday, 02:36:

I have 386SX40, 386DX33 and 386DX40 motherboard in collections,they are just in shelf collecting dust, i dont have any interest in 386, back on time, i jump from 286 to 486DLC directly

Forget, I have a 486SLC40 too

Ah, well if you don't mind blowing off the dust, if you can put together the 386DX40 + Cyrix 387 that would be perfect.

Anyone else with such a configuration is welcome too. I just think the classic 40 + Cyrix is the best balance of powerful + common enough

Gahhhrrrlic wrote on Yesterday, 05:29:

theelf wrote on Yesterday, 02:36:

I have 386SX40, 386DX33 and 386DX40 motherboard in collections,they are just in shelf collecting dust, i dont have any interest in 386, back on time, i jump from 286 to 486DLC directly

Forget, I have a 486SLC40 too

Ah, well if you don't mind blowing off the dust, if you can put together the 386DX40 + Cyrix 387 that would be perfect.

Anyone else with such a configuration is welcome too. I just think the classic 40 + Cyrix is the best balance of powerful + common enough

No problem, just post something interesting and i will build the 386

Anyways, for ones that have 386 or DLC, using a reduced wad help from little to a lot, for example, jaguar one, https://www.wad-archive.com/wad/a105129389b27 … bd9ed402aea7491

In my case, some levels that in PC i get not very nice fps with jaguar wad is much more playable

Sadly the wad miss E3M2, then hang in this eppisode

?? What's the difference? Did the Jaguar wads have more optimized BSP searching or something?

TheElf: I sent you a PM.