Hi Feipoa,

On the AMD5x86-160 I believe I get around 26fps at the mentioned settings but I’ll go back and double check. The easiest way to find one is to search Diamond S220 on eBay, the main chip is clearly marked “Rendition” to be certain.

No worries about the typos, your family scenario sounds much like mine 🤣.

feipoa wrote on 2021-11-08, 03:03:

I'm sure you have other mistakes in this thread that haven't been caught.

Yep. I keep fixing stuff when spot an inaccuracy. Yesterday noticed little something in old post about vlb eide controllers. Most things should be well curated by now, but I bet there are many mistakes still left untouched.

A summary of the mid-to-end of line 486 motherboards "research" early in the thread.

The effort started with building a long list of motherboards.
Most of them got discarded based on various criteria.
15-ish made to the hands-on examination phase.
Most of the results are shared in the thread, but got too big and hard to follow, so putting together this compact summary split in 6 sections, as follows:

--- 486 class CPU at 160MHz

The sweet spot of all things 486.
Easy to achieve with Am5x86-133 processors + air cooling.
Plenty of motherboards handle it well, but there are 3 that stand out:

Asus PVI-486SP3
One of the most flexible 486 assemblies. ISA, VLB, PCI buses. Rock stable and hustle free package.
Best at offline compute - (de)compression, encoding/decoding, 2D/3D rendering, image editing, spreadsheets, etc.
Its PCI interface is slightly behind the curve at DOS interactive graphics. Comes back on top of any ISA/PCI board when paired with fast VLB video cards like Ark1000VL or S3 Trio64.
Its PCI graphics are top notch at accelerated Windows GUI and in line with the best ISA/PCI mobos (which excel at that with more modern PCI cards like Matrox Millennium).
Best on-board IDE controller - period 😀

Asus VLI-486SV2GX4
The best 8-bit pixel pusher. Fastest board for DOS interactive graphics. Especially if paired with fast VLB video cards like Ark1000VL or S3 Trio64.
With S3 Trio64 it is on par with the best ISA/PCI boards at accelerated Windows GUI. But only with 16-bpp and up to 1024x768.
Lacks in offline compute.
When paired with strong VLB EIDE controller (like Promise EIDE2300 Plus) it takes a PCI UltraDMA/33 tech to outdo it at disk i/o.

Biostar MB-8433UUD-A
The best implementation of UMC's UM8886BF late 486 chipset.
Probably the fastest ISA/PCI board out there. Also - rock solid.
Its on-board IDE controller and drivers are lacking, but the PCI interface provides options for Ultra DMA class upgrades.
Limited out-of-the-box L2 cache size. Also, extremely picky about cache chips that can hit the tightest BIOS timings.
If 1024Kb L2 cache mod is applied the board pulls even further ahead - best offline compute, best Quake 1 score, best at Windows GUI (on par with LuckyStar C2), second only to VLI at DOS interactive graphics.

--- 486 class CPU at 180MHz

Several motherboards can do it on a test bench.
Few can do it well.
Two are worth putting inside a PC case.
Building such a computer is not trivial, so selecting the right components makes big difference.

LuckyStar LS-486E rev:D
Clearly the best option in this cathegory, but only after replacing the on-board 15ns L2 cache chips with 10 or 8 ns ones.
It was in a group of late and fast ISA/PCI boards ticking great at 160MHz, but that was it - one of many.
Apparently there is plenty of built-in padding that allowed it to scale much better than anyone else past the 160MHz.
Very fast, very stable and just works.
Handles well Voodoo3 which at this CPU speed allows decently performing 3D accelerated graphics.
For reference - Voodoo3 seems to be the best option for 3D accelerated graphics on a 486-class hardware.
For some reason does not like Gotek floppy emulators. Looks like inherent issue unrelated to overcloking, etc.

Biostar MB-8433UUD-A
Similar to LSD - fast and solid. Scales slightly worse than LuckyStar D.
UMC's IDE drivers lead to instabilities and cannot be used = less-than-ideal IDE performance.
Not a big deal - Promise UltraDMA-100 PCI to the rescue.
Voodoo3 incompatibility is a drawback. Voodoo2 is still a great option but requires extra PCI slot. SLI makes the situation even worse.

PC-Chips MM919 v3.4B/F
With 1024Kb L2 cache module this is the fastest system at 180MHz.
Still, leaving it in third position because of the next factors:
- Voodoo3 incompatibility.
- 1Mb L2 cache modules are very rare.
- If base frequency is above 40MHz automatic PCI bus divider kicks-in. To work around that Turbo has to be OFF until POST completes.
- UMC IDE drivers issues like with UUD.

Honorable mention - Chicony CH-471B revision 2 - truly unique assembly.
1024Kb L2 cache, ISA/VLB, on-board Appian ADI/2 EIDE controller, clockgen with 60/66 MHz frequency.
So far the only ISA/VLB board in the 3x60 department.

--- 486 class CPU at 200MHz

Very few boards can do this well at 3x66MHz, more can handle 4x50MHz.
3x66 is better in general than 4x50 despite the requirement for PCI bus divider.
LuckyStar LS-486E rev:D and Biostar MB-8433UUD-A do best at 3x66.
Finding the right set of L2 cache chips for the Biostar UUD board can be particularly unpleasant.
Chicony CH-471B rev.2.0 is another impressive assembly - the fastest classic ISA/VLB motherboard out there. It actually outdoes some of the later 486 assemblies in some disciplines by relying on Ark1000VL ticking at cool 66MHz.
Few SiS 496/497 PCI/ISA boards with 1Mb L2 cache can handle reliably 4x50 and follow very close behind the two above.
Then everybody else (at 4x50).

Honorable mention - Chicony CH-471B revision 2 - truly unique assembly.
1024Kb L2 cache, ISA/VLB, on-board Appian ADI/2 EIDE controller, clockgen with 60/66 MHz frequency.
Handles reliably the 200MHz (3x66) which makes it the most scalable and because of that the fastest ISA/VLB motherboard.
Even better - without bus dividers. Yes, thats right - Ark1000VL VLB ticking at 66MHz.
Impressive.

--- Pentium Overdrive at 100MHz

This one is very easy - Asus VLI-486SV2GX4.
While it is awesome, nothing is truly perfect.
There is one drawback - when L1 cache is set to WB mode any configuration with more than one memory module (in the first slot) can be unstable in some situations. The motherboard cannot recognize modules bigger than 32Mb, so this is the limit.

Close second are LuckyStar LS-486E rev:D and Biostar MB-8433UUD-A.
They do better at accelerated Windows GUI and some offline compute tasks.

--- Cyrix 5x86 at 133MHz

I don't have direct experience with these CPUs.
Adding the section for completeness.

There are 2-3 threads talking about that.
LuckyStar LS486E rev.C2 and Cyrix 5x86@133
The World's Fastest 486
The Ultimate 486 Benchmark Comparison
Users Ph4ntom, Rad and Feipoa contributed most of the information there.

Performance seems to be slightly lower than Am5x86@180MHz in most tasks.
Looks like the best boards are again LuckyStar LS-486E and Biostar MB-8433UUD-A.

At least Feipoa and Rad have assembled PCs around this configuration, so it will be up to them to clarify.

--- Cyrix 5x86 at 150MHz

There is limited information about the subject.
Cyrix 5x86 120(FSB40,60) 150MHz(FSB50) Speedsys result
Hypothetical 5x86
The Ultimate 486 Benchmark Comparison
Also, several other mentions here and there.
Mostly Feipoa's contribution based on his Biostar MB-8433UUD-A setup.
Does not feel like a long-term stable solution, but still worth mentioning it for completeness.
Performance seems to be approaching Am5x86@200MHz and POD100 levels in the different tasks.

---

This is what i got so far.
Closing the 486 chapter with this post but reserve the right to revisit the notes if new facts emerge.
Also, starting to investigate some options for long-term stable 200MHz system. I think it is possible, so will see ...

Nice summary.

I would add that at 3x60 MHz, the UUD board would only work with 256K when the fast timings were applied. With 1024 or 512K, timings needed to be reduced and reduced timings w/1024K are slower than with fast timings at 256K. I think you should also note the BIOS L2/RAM timings employed for the LSD and UUD board at 3x60.

It would be good to know if the POD100 limitation on the 486SV2GX4 is also afflicted by 512K-WB.

re: Cyrix/IBM 5x86 at 133 MHz - would be good to point out that only the LSD and UUD boards seem to work reliably with a 66 MHz FSB, and only the UUD board has a functioning floppy controller at this freq. I do plan to see if I can work around the LSD's floppy limitation by using an LS-120 drive with one of those SuperDisk EEPROM's loaded on a network or lotech card. The purpose of this is to get bootablity back to a 3.5" floppy. User LuckyBob was kind enough to send me an image of a SuperDisk's EEPROM.

Good notes. Thanks.

VLI POD100 (2.5x40) issue is related to L1 cache in WB mode.
I am going to swap around bunch of memory modules today and see if there is a magic combination that results in fully maxed-out system.

Made an edit above UUD@1024K@160MHz.

@pshipkov Can you share jumper settings for POD in the LS486 Rev. D ? Im not running stable at 100MHz for some reason.

On the other hand all the same hardware is stable all day long with an AM5x86 on the same board at 180MHz.

Quite odd. I used this guide from the manual.

whats the benefit of scsi over ide has there been a benchmark?

Chadti99 wrote on 2021-11-17, 01:49:

@pshipkov Can you share jumper settings for POD in the LS486 Rev. D ? Im not running stable at 100MHz for some reason.

Have you confirmed that your POD can handle 100 MHz? If it runs fine at 83 MHz and not 100 MHz, then maybe it is the CPU. Are you running the POD with the diode mod so that it runs at 4.1 V?

feipoa wrote on 2021-11-17, 02:23:

Chadti99 wrote on 2021-11-17, 01:49:

@pshipkov Can you share jumper settings for POD in the LS486 Rev. D ? Im not running stable at 100MHz for some reason.

Have you confirmed that your POD can handle 100 MHz? If it runs fine at 83 MHz and not 100 MHz, then maybe it is the CPU. Are you running the POD with the diode mod so that it runs at 4.1 V?

Yes this is a diode modded POD confirmed stable at 100MHz on at least three other boards. I’ve tried running at both the 4v and 5v setting. Quake crashes and Windows boots but begins to throw exceptions not long after. I’m going to try swapping the single 64MB module out for something smaller as my next troubleshooting measure.

@chadti99
That's odd.
I don't think i did anything special for the POD100 CPU.
Will check and confirm soon.

I was actually thinking to run a proper POD100 study and incorporate some of the findings from the 486@180/200MHz investigation.
I have one post in the thread, but the info there is limited.

@zapbuzz
Will try to summarize my answer in few short sentences.
If we look at IDE/EIDE/UIDE/SATA controllers/drives and their SCSI counterparts from any given time period we will notice that the best members of the first group deliver better peak performance than the best members of the second group, but at the expense of consumed more system resources. While this is the general case - don't take it as the 11th commandment - there can be exceptions to the rule.

For a gaming computer or single user workstation the devices in group 1 are more practical - cheaper, more compatible, better peak perf.
If you are running a system that has to handle lots of disk i/o - it makes sense to go SCSI - it will reduce the load on the system.
There can be specific cases where video game, app or the operating system complete faster a group of parallel tasks that include disk i/o if there is a SCSI in the mix. For example - OS boot-up time may be shorter.
But again - from my experience, a well oiled IDE/EIDE/UIDE/SATA subsystem always comes on top of any SCSI combo from the same period.
But again - there is a lot of charm in retro SCSI which is great motivation factor to use it. 😀

The only decently comprehensive benchmark i have seen around is early in this thread.
It is limited to VLB EIDE adapters compared to VLB SCSI Adaptec card.
You will notice the above trend reflected in the shared perf numbers.
Same applies to ISA and PCI adapters before and after that time period.
The whole subject begs for a proper study.
Maybe one day ...

If you search online there is good info available for y2k and later controllers.

@Chadti99
Everything seems to be in order with LuckyStar rev.D and POD100.
Video reference here.
GlQuake, SpeedSys, Quake 1, Doom.
Tested with 3d offline rendering and few other time consuming things that are too long and boring, so not in the frames. Everything checks out. Always did with this board.
Quake 1 score is actually 24.3 fps, but after quitting Win95, the SB's CTCM thing kicks-in which affects Q1's perf, so in the video it shows as 23.7.
All BIOS settings on max. L1 and L2 cache in WB mode. Used 32Mb RAM only, otherwise perf drops in Windows with the limited 256Kb L2 cache in WB mode.

Checked with a second LSD board - same - things are as expected.

pshipkov wrote on 2021-11-17, 08:01:

@Chadti99 Everything seems to be in order with LuckyStar rev.D and POD100. Video reference here. GlQuake, SpeedSys, Quake 1, Doo […]
Show full quote

@Chadti99
Everything seems to be in order with LuckyStar rev.D and POD100.
Video reference here.
GlQuake, SpeedSys, Quake 1, Doom.
Tested with 3d offline rendering and few other time consuming things that are too long and boring, so not in the frames. Everything checks out. Always did with this board.
Quake 1 score is actually 24.3 fps, but after quitting Win95, the SB's CTCM thing kicks-in which affects Q1's perf, so in the video it shows as 23.7.
All BIOS settings on max. L1 and L2 cache in WB mode. Used 32Mb RAM only, otherwise perf drops in Windows with the limited 256Kb L2 cache in WB mode.

Checked with a second LSD board - same - things are as expected.

Thanks for checking, I’m sure I’ll determine the root cause, those jumper settings I posted were correct?

I’m running a 64MB stick in L2 WT so I’ll try 32MB with L2 in WB to be sure that’s not the problem.

Your jumper settings are correct.
64mb + l2 in wb works too, just a bit slower in Windows.
64mb + l2 in wt is fine as well but perf is lower across the board.

I bet on RAM incompatibility.

The POD is an odd beast.

I’m completing Quake benchmarks now with a single 32MB EDO stick. I’m also seeing slightly better Quake timedemo scores with L2 in WT. I’ve noticed this on other boards as well, with the POD specifically. The GX4 being the exception.

POD100
32MB
LS486D
256k

GLQuake
Voodoo 3 - 39.7 fps
Voodoo 4 - 39.7 fps

Turok
Voodoo3 - 23.6
Voodoo4 - 23.4 fps

Lost video on the 2nd pass of the Turok Benchmark with the Voodoo4, assuming power related.

Check this out guys, framerate counter in MotoRacer! Lower left is FPS, unsure what the other info is. Testing POD100 vs AM586180.

Just running the counter at the start of the first race and not moving:

320x240
POD100 13-21 FPS
AMD180 13-20 FPS

feipoa wrote on 2021-11-08, 00:37:

I don't have a Rendition Verite based card. Anything special about it

I once owned a Diamond Stealth S220 new when the card was still relatively current. I was really not impressed. It had the worst 2D output quality of any graphics adapter I'd owned up to that point. I remember the colours being really dim and washed out. Kind of like crappy S3 Virge cards, but worse. I also remember having issues with screen corruption using the drivers it shipped with. It was definitely shitty drivers, not bad RAM. Also, I was not impressed with the 3D performance. Though, to be fair my first experience with 3D graphics was a voodoo3, so that was what I compared it to. I remember to get unreal going smoothly, I had to drop down to 512x384 or some other weird intermediate resolution. I only had this card for a few months before I binned it.
Unless there was some kind of fix for the crappy 2D output, I wouldn't even consider one of these. Are any of the Verite cards better than the S220?

Good to see things ticking.
You are consistently getting slightly lower numbers despite both of us using the same hardware.
OS here is Win95 OCR2, but remember seeing same perf with DOS 6.22.
Wonder why the difference.

I also noticed that Asus VLI and PVI boards like better L2 in WB mode in general.

Not sure what to make out of the Motoracer frame rate. Never tried it myself. So what is your stance on the numbers ?

@Anonymous Coward
Wonder if you got a bad card back then.
The few screenshots i saw online from Rendition cards look fine to me.
But yes, looks like drivers and 3D accel was not stellar, especially if your prior experience was V3.

@pshipkov Do you have the jumper settings to run an AM5x86 at 160 or 180 on the Gigabyte GM486AS? Wanted to test the Diamond S200 aka Rendition V2100 for feipoa on this board.

pshipkov wrote on 2021-11-17, 18:51:

Not sure what to make out of the Motoracer frame rate. Never tried it myself. So what is your stance on the numbers ?

I’m not sure just yet either but it’s a Direct3D game that works fairly well on a fast socket 3 so I’m glad I have a way to gauge performance other than eyeballing it. I would of thought the POD100 was pulling better numbers over the AM5x86.