Comparison of AMD X5 133MHz vs Cyrix 5x86 100 (3x33) and 120 (3x40) performance on the Tiny 486 SBC system:

The Cyrix at 100MHz is getting more than 90% of the performance of the AMD, yet at only 75% of the frequency. Increasing the bus up to 40MHz makes a substantial difference, especially on this board which is just a mediocre performer, memory-wise. Everything gets a nice bump from that faster bus.

The cyrix register enhancements also make a noticeable difference. However with my 5x86-100GP overclocked to 120MHz, Quake is unstable with any of the enhancements turned on. Without enhancements turned on it will happily play the timedemo, over and over again. At stock 100MHz speeds the enhancements are solid; not one crash.

Giving it a bit more voltage at the overclocked speed may help, but the options just aren't available on this industrial board.

I suspect I'll leave the Cyrix chip in, and running at 120MHz, but with the tweaks disabled. Shame Cyrix never properly resolved the bugs with the enhanced features in the design; it's clearly far closer to the 6x86/Pentium than a 486.

Did you try the AMD at 40mhz?

No I didn't. The Cyrix was an opportunity to increase speeds a little, but the primary goal was to have something that was more configurable than the AMD. I suspect that at 160MHz it would be ahead of the 120mhz Cyrix again a little in certain tests, if the Ultimate 486 benchmark thread is anything to go by.

The Cyrix chip can have the multiplier changed, as well as do all sorts of funky stuff with the cache - with the AMD I'm stuck with turning the cache on/off. I'd like to do away with the use of moslo for older games since it doesn't always work that well.

For work performed per clock though, it's clear the later design of the Cyrix has the AMD beat quite conclusively.

The next round of testing will be to see if I can get the Cyrix chip to perform close to some of my (now in storage) older machines, such as a 25mhz 286 and a 386dx 40.

Reran selected benchmarks with the Cyrix 586-120 set to various slowdown states:

With purely software settings via SETMUL, I can get approximately the following speeds:

- Native
- 486DX2-66 (multiplier set to 1x)
- 386DX-40 / 486DLC-40 (multiplier set to 1x, L1 cache disabled)

Disabling L2 cache in the BIOS gets additional speed options down to 386DX-25 / 386SX-33 / 286-25 levels.

With multiplier to 1x and L1 disabled, Wing Commander 1 (the most speed sensitive of the games I want to play) is at the faster end of playable. The most optimal setting is with both L1 and L2 cache disabled, but multiplier set to 3x. At 1x it drops to speeds approximately the same as my 286-25, which is definitely playable, but not quite as smooth as you would want.

Interestingly the benchmarks show that L2 cache has a negligible impact on Doom scores; they are far more impacted by L1 cache than anything else.

For an all-in-one, exclusively Dos system, I think this makes the Cyrix chip a more flexible option than the AMD.

So I think I have finally gotten to the bottom of the digital audio crackle / popping issues.

Whilst playing around with BIOS options to tweak the recently installed Cyrix cpu, I went hunting to see if there was anything related to DMA settings, latency, timers etc. There wasn't anything, but there was a setting labelled "IOCHREADY", which based on my (limited) understanding allows an ISA device to asset control of the bus for longer than normal.

The default setting has this disabled, but I tried turning it on and the cracking and popping during digital audio playback is gone completely. I've been back through all of the tested games and every one which I noted as "working with YMF, but noisy" is now sounding perfect. The worst offender for this was Stargunner, with its constant streaming background tracker music; it was so awful to be unplayable. Now it is perfect.

I have another 60 or so games to test, on top of the 230 or so which I've already completed, but based on the experience so far, I'm quietly confident that they should all work properly, too. It appears as if I now only have issues with Jill of the Jungle and all of the Legend Entertainment games (via SoftMPU). Jill I can deal with - it's only my kids who play that one, but I'd really like to get the Legend titles working with MT-32, but it just seems so buggy.

megatron-uk wrote on 2024-05-11, 14:18:

So I think I have finally gotten to the bottom of the digital audio crackle / popping issues. […]
Show full quote

So I think I have finally gotten to the bottom of the digital audio crackle / popping issues.

Whilst playing around with BIOS options to tweak the recently installed Cyrix cpu, I went hunting to see if there was anything related to DMA settings, latency, timers etc. There wasn't anything, but there was a setting labelled "IOCHREADY", which based on my (limited) understanding allows an ISA device to asset control of the bus for longer than normal.

The default setting has this disabled, but I tried turning it on and the cracking and popping during digital audio playback is gone completely. I've been back through all of the tested games and every one which I noted as "working with YMF, but noisy" is now sounding perfect. The worst offender for this was Stargunner, with its constant streaming background tracker music; it was so awful to be unplayable. Now it is perfect.

I have another 60 or so games to test, on top of the 230 or so which I've already completed, but based on the experience so far, I'm quietly confident that they should all work properly, too. It appears as if I now only have issues with Jill of the Jungle and all of the Legend Entertainment games (via SoftMPU). Jill I can deal with - it's only my kids who play that one, but I'd really like to get the Legend titles working with MT-32, but it just seems so buggy.

Weird that the BIOS allowed or even defaulted such an important ISA signal to be turned off, haven't seen that before... 🤔

Eivind wrote on 2024-05-11, 15:13:

Weird that the BIOS allowed or even defaulted such an important ISA signal to be turned off, haven't seen that before... 🤔

The only thing I can think of is that it is something specific to being an industrial SBC board, and not a general purposes design. Perhaps it's there to guarantee the bus doesn't get hogged by a greedy ISA device? Anyway, for our gaming purposes the default 'off' mode clearly isn't what we want!

megatron-uk wrote on 2024-05-10, 13:31:

I'm wondering if the issue is actually something related to the VIA chipset DMA implementation.

Yes sounds like something is messing with it - are there any interesting options in the BIOS to fiddle with around the ISA bus?

I can also recommend the Crystal based ISA cards - some of them are super cheap and small. Great SB Pro emulation and if you get a CS4236 based card then the FM sounds pretty nice too. Avoid the CS4235, bad FM.

badmojo wrote on 2024-05-13, 05:59:

megatron-uk wrote on 2024-05-10, 13:31:

I'm wondering if the issue is actually something related to the VIA chipset DMA implementation.

Yes sounds like something is messing with it - are there any interesting options in the BIOS to fiddle with around the ISA bus?

I can also recommend the Crystal based ISA cards - some of them are super cheap and small. Great SB Pro emulation and if you get a CS4236 based card then the FM sounds pretty nice too. Avoid the CS4235, bad FM.

It was the IOCHREADY option in the bios. The default state is "disabled", but when toggled on all of the audio clicking is immediately gone.

As I mentioned above, I can only assume that since this is a dedicated industrial board, that having some kind of deterministic setting for cards attached to the ISA bus is desirable. I have never seen that option on any other bios.

Even in Stargunner with the constant streaming audio the digital sound is now flawless.

Back to game testing... I'm up to around 280 titles now, some noticeable problems with the most recent batch:

- every single Legend entertainment game locks up with SoftMPU: 100% failure rate!

- fallout 1 drops back to a dos prompt with no errors

- Shadow of the comet crashes

- Plan 9 locks up at the first gaming screen

There are a few more, but mostly not audio related.

Great result! I've never heard of that option either.

megatron-uk wrote on 2024-05-10, 16:34:

Comparison of AMD X5 133MHz vs Cyrix 5x86 100 (3x33) and 120 (3x40) performance on the Tiny 486 SBC system: […]
Show full quote

Comparison of AMD X5 133MHz vs Cyrix 5x86 100 (3x33) and 120 (3x40) performance on the Tiny 486 SBC system:

amd-vs-cyrix.png

The Cyrix at 100MHz is getting more than 90% of the performance of the AMD, yet at only 75% of the frequency. Increasing the bus up to 40MHz makes a substantial difference, especially on this board which is just a mediocre performer, memory-wise. Everything gets a nice bump from that faster bus.

The cyrix register enhancements also make a noticeable difference. However with my 5x86-100GP overclocked to 120MHz, Quake is unstable with any of the enhancements turned on. Without enhancements turned on it will happily play the timedemo, over and over again. At stock 100MHz speeds the enhancements are solid; not one crash.

Giving it a bit more voltage at the overclocked speed may help, but the options just aren't available on this industrial board.

I suspect I'll leave the Cyrix chip in, and running at 120MHz, but with the tweaks disabled. Shame Cyrix never properly resolved the bugs with the enhanced features in the design; it's clearly far closer to the 6x86/Pentium than a 486.

Love this benchmark, i have a 386SX in a ISA card, but never tested a 486/586, thanks

the cyrix benchmarks are more or less 30% slower than my cx586 PC, not bad !