For a 486, L2 cache,

0kb of cache vs 128kb of cache .... definitely a decent different.
128kb of cache vs 256kb of cache ... not a big different. Its there, but it won't punch you in the face.

What video card you are using? This also makes a difference.

Regarding being the perfect processor for 3D gaming, that is kinda a bit hyperbolic. And it really really depends on the game.

Doom likes smooth loading to cut down on lags: smartdrive, decent hard drive (not slow ones, mine was 4500rpm, 5400rpm is better) and 8MB to 16MB ram.

I used to have a 486 strictly DOS 6.22 with default settings, but I overclocked to DX 50 from 486DX 40 (AMD), cache was 256K as well all 9 sockets filled. Controller and Video card on VLB slots, I had ATI ISA, bought a Tnesg ET4000w32i/p made by Hercules in 93~94 remember everything were highly hyped, I should had not listened to the excitable idiot who sprouted tseng all the time. S3 VLB cards is better and little cheaper, even Cirrus logic if your board is VLB.

If your board is a PCI
Tnesg is not a requirement anymore, you can use a S3 DX 4MB or GX 4MB even Trio64V+ (2MB) instead.

Cheers, Pentiumspeed

3D gaming is done with voodoo2 or later, or TNT2, geforce2 or geforce4. AGP, PII or PIII. Pentium straddles this transition from 486 era to PII/PIII and 3D gaming was rapidly evolving that quickly outstripped the 486, and fast Pentium 200-233 was the minimum. I was playing Myst series on PII 350 and unreal gold with weak AGP (was geforce2 MX 64bit as I had little money then).

Cheers, Pentiumspeed

For Doom, if everything else is right in the system, yes, a DX4 is an improvement over DX2. A proper DX4-100 is where it starts to reach its 35fps frame rate limit. Not in every situation, but in most. DX2-66 was the de facto top CPU when it came out as no one had a Pentium-66 in late '93.

And it needs a fast VLB/PCI VGA too.

About cache, I'm not sure. 256k vs 128k is surely much less of a diffrence than zero vs 128k.

canthearu wrote:

For a 486, L2 cache, […]
Show full quote

For a 486, L2 cache,

0kb of cache vs 128kb of cache .... definitely a decent different.
128kb of cache vs 256kb of cache ... not a big different. Its there, but it won't punch you in the face.

What video card you are using? This also makes a difference.

Regarding being the perfect processor for 3D gaming, that is kinda a bit hyperbolic. And it really really depends on the game.

That makes sense, with the cache. As for the video card...

pentiumspeed wrote:

S3 VLB cards is better and little cheaper, even Cirrus logic if your board is VLB.

I already have Cirrus Logic VLB video- It came integrated in my motherboard. It's a 5430 VLB 1MB model, so I'm not sure how much faster I could get with my ISA/VLB system.

I also find the name "pentiumspeed," ironic here, in a dumb way. Get it? Because DOOM isn't running unrealistically fast on my 486, and a Pentium would solve... that. 😵

I have a 486DX2-66 in one of those cheesy fake cache PCChips boards. Doom runs fine on it, however it does have a PCI video card, which probably makes all the difference.

The cirrus logic 5430 is pretty good for the class of system it is in (midrange 486)

An upgrade to 100mhz would certainly make the system run 3D games faster, but it will only be along the lines of 30-40% faster.

Video bus speed matters for getting that framebuffer filled but DOOM is cpu bound more than anything.

JudgeMonroe wrote:

Video bus speed matters for getting that framebuffer filled but DOOM is cpu bound more than anything.

It depends. For a 486 with VLB/PCI VGA, yes it is. For a late Pentium it's all about the video card. Doom uses mode X (Y?) and 8-bit video memory access and scales crappily on faster hardware.

And even on a fast 486 it heavily depends on the VGA. Back in the day I had an old Tseng ET3000 paired with an Am5x86 and the result was horrible. Replacing the VGA with a PCI S3 Trio64 made Doom fly. On the other hand, Heretic and Hexen was much better with the old Tseng as they render in much faster system memory then copy the finished frame to the VGA memory at once. Okay, 5x86 + Tseng ET3000 is a rather extreme example.

I remember in the mid-1990s we had a 486DX4-100 with 16 MB but I still usually played in low-res F5 mode for the speed bump.

Doom is a bit meager on a DX2. It was the perfect processor in its time because everything else was so slow and it was able to tackle a LOT of different applications.

A DX4 and ultimately an overclocked X5 are more of an enjoyable experience especially if you run torture tests like Episode4 and some "notorious" maps of Doom2.

A CF card may eliminate some of the lag in Doom(which comes from caching the assets during gameplay) - or maybe use a SCSI card with on-board mem.

The highest scores I got on 486s regarding Doom were with Matrox PCI cards. Doom does not need a lot of transfer speed to the video card(over 2MiB(excluding headroom) per second) but it does depend, as already pointed, out on raw CPU speed and on the speed the vga itself works due to depending on the internal "workings".
Cards like the TNT, Riva128, ArkLogic, RagePro and S3 Virge were slower but not much.
If you're bored, you could test your onboard VESA card in speedsys and x-vesa and see how it does compared to an external one.

I wish I still had my 90's PC it was an unbranded clone so I don't recall which motherboard it used or if it had cache.

But I'm guessing it did because I used to play Doom Deathmatch on a 486sx33, 8mb ram, 1mb Cirrus Logic VLB at the maximum screen size that still shows HUD (-1 from max?) and it played decent. I guess a good part of it was my expectations, realistically I was probably playing at a consistant 20fps. Later I upgraded to a DX4-100 and that was a noticeable improvement but I had never complained about the SX33 performance.

Me and my brother played Doom via a null modem cable between two Compaq machines that had cacheless 50MHz SX2 in them with VLB ET4000/W32 and the game was smooth with high detail and "full screen with HUD visible" size. Duke3D was completely unplayable even with smallest screen size, that game absolutely needs cache to be playable on a 486...

Duke3D is hurt by the lack of math co-processor (sx) too. I remember playing Duke on a DX2-66 and it was acceptable but not great.

Duke3D is hurt by the lack of math co-processor (sx) too.

Build does not utilize FPU af all.

The Serpent Rider wrote:

Duke3D is hurt by the lack of math co-processor (sx) too.

Build does not utilize FPU af all.

I only know the following because of NexGen CPU discussions on this forum.

To get a definitive answer on why slopes are so slow on 486SX processors, I turned to the most reliable source I can think of: the original Duke 3D source code released in 2003. After looking around for slope-related code, I finally found what I was looking for: the Build engine does in fact use floating point assembly instructions in its slope rendering routines (setupslopevlin_ and slopevlin_) inside the A.ASM source file. It's just a few short simple instructions (fild, fadd, fst, fstp), hardly worth mentioning, but it's enough to bring a non-FPU processor to its knees when invoked.

-https://fabiensanglard.net/duke3d/build_engine_internals.php

This is good to know !

Doom 1 timedemo on my 486DX2/66MHz (WT), 16MB RAM, 256kB L2 cache, Cirrus Logic CL-GD5428 1MB VGA.

https://www.youtube.com/watch?v=ZdsPgyZdCoc

Perfectly playable, but definitely not perfect. As alvaro84 said, to hit that 35 fps limit constantly, you need at least DX4/100MHz. Btw. L2 cache itself means +5 fps (*) in comparison with system with no L2 cache installed.

(* 25 vs 30 fps, see table at https://www.high-voltage.cz/2018/globalyst-51 … kolo-vylepseni/ - measured in timedemo)

Interesting - I remember a LONG time ago testing a DX2+VesaLocal board and it got around 27fps.

I was curious about the plane switching and it seems that on my vga-card 64000 plane-switches take around 0.05 seconds(7.8125e-7 seconds/switch). Not a realistic number by a long shot as there are most likely way fewer per frame drawn but neat to know.