BINGO!
Finally after a long, long time of testing, it was possible for me to made a really stable and extreme fast 486-system with an UMC-chipset.

ALL mainboards Shuttle HOT-433 and ECS UM8810-PAIO in the past time I tested are NOT stable at FSB 60 and 50 MHz – regardless of their revision, regardless of using EDO- or FPM-RAM and under any BIOS-conditions! I tested them with an AMD 486-DX5-133 ONLY, so I can not be sure for their (in)stability with other CPU-Models at this time. The problems I had are shown in this thread (Modern graphics on a 486), there is no need for description once again.

The instability-problems END for me with another mainboard (I never had before) shown and explained very well in this thread: 486 mobo + 586 chip

It is a „very young“ 486-board SYL8884PCI-EIO, with an UMC-chipset made in 1997! Maybe this is the reason for stability, I really don't know.

ONCE AGAIN: THIS BOARD IS ROCK STABLE AT FSB 50 MHz AND 60 MHz (at least with the AMD-486-CPU) – THEY ARE ABSOLUTELY NO PERFORMANCE-DROPS FOR THE CPU IN WINDOWS!

From this board they exist some different models from different manufacturers, so they vary regarding the jumper-settings a little bit.

The good: it works with EDO-RAM, has a CR2032-cell and a PCI-Generation of 2.1.

The bad: ugly onboard-IDE-controller, no PS/2-support, bad layout for the place of the both COM-ports (very difficult for using of PCI-cards because of collision, if one or both COM-cable(s) is/are used)

The onboard-IDE-controller does work fine for my board with 2 (two) IDE-devices only, and at PIO-mode 3 (not 4) for a FSB of 50 or 60 MHz (PIO-mode 4 causes crashes/freezes during booting of DOS/Windows at 60 MHz FSB; at 50 MHz they occur errors inside of Windows).

Apart from 3,3 V and 5 V, my mainboard does not offer CPU-voltages of 3,45 V and 4,0 V, so I had to find and use an AMD-486-DX5-133 which can work at 5 V. And I was lucky – I found one working at this voltage at 200 MHz at FSB 50 MHz!

The L2-cache is 512 KB, the modules are 10 ns (only the TAG-module is 12 ns – at this time I do not have one with 10 ns).

About the RAM: shown in the pictures is the very one EDO-RAM-module of 64 MB at 50 ns, which works fine with this board. I had no luck using one or two double-sided modules with 32 MB each (two different brands).

So in conclusion it is possible to run the system with all BIOS-settings set at their fastest/maximal values without errors and problems! 😉

The result is an extremely fast CPU- and RAM-performance, which is (apart of the pentium- instructions and the FPU-performance) competitive to a mid-ranged Pentium-I-system without MMX from 1996/1997!

Other problems:

1. Sadly, the board was not able to boot at FSB 66 MHz with the same 486-133-CPU (at 200 MHz = 3*66) and with another AMD-486-120WB (at 133 MHz = 2* 66). HAS SOMEBODY MORE EXPERIENCE if the AMD 486 is able or not able to boot at FSB 66 in generally?
2. After using 3D with OpenGL (Quake I and II), there is not possible to shutdown Windows to the end (Windows hangs at the message-screen „Windows is shutting down” )

Windows-Results 3D:

Quake I (800x600 dpi ,windowed mode): 30,3 FPS
Quake II (800y600 dpi, Fullscreen): 18,1 FPS (best: 18,2)

Other DOS-results with at 200 MHz (no pictures):

DOOM (timedemo 3): 76, 21 FPS (978 realtics)
Quake I: 21,7 FPS
PcP-Bench (VGAMODE): 30,6
3dbench 1.0c: 115,9
the only strange thing is, CACHECHECK means, there is Fake-L2-cache (not sure why – the system works perfectly)

By the way: the CPU works fine at 180 MHz (3x 60), too. But in this case, the cache-timings in BIOS must be decreased from 2-1-2 (for 50 MHZ FSB) to 3-1-3, and this results in a drop of the RAM-performance from 70 MB/s (at 50 MHz FSB at 2-1-1) to 60 MB/s, as well in a drop for Quake I and II of about – 2,0 FPS (because of the lower CPU-Frequency I think). So the higher FSB itself was not helpful in this case 😉

The rest of the hardware:
VGA: GeForce 2 MX-400, 32 MB (driver 6.31)
Sound: Creatice Sound Blaster 16 (CT2290)
LAN: 3Com EtherLink III (disabling it temporary in Windows for the Quake-tests results in + 1,0 to + 2,0 FPS)
USB: Opti FireLink 82C861 (I let it enabled during the Quake-tests; disabling it would add 1 or 2 more FPS to Quake)

Well, let's start with lots of pictures

more pics

even more pics

even more pics

even more pics

deleted images of SiSoft Sandra

deleted images of SiSoft Sandra

Oh that’s sweet!!!

Any chance of a 3dm 2000 run?

Can you compare with radeon 9250 or radeon 7500?

Very cool @gonzo.

Did you target the SYL8884PCI-EIO motherboard because of that other thread you linked, or it fell in your lap by coincidence ?
It looks like a late budget version when the world moved or was moving en masse to Pentium.
Most of these assemblies are not very fast on a clock to clock basis, but scale pretty well.

UMC's UM888# chipset is notorious with its week IDE controller, so your findings are inline with what we have seen before.

You should be able to get to at least 3-1-2 timings with carefully curated level 2 cache chips, assuming the board has the chops for it.

The 50ns EDO memory modules you used are proven quality. Not each and every one of them, but quite a few.

No 66MHz FSB you say ?
This is strange.
Do you use active cooling with Peltier, or riding on air ?
If on air - good chance that's the problem.
Otherwise - either the mobo or the RAM cannot take it.
This chipset and many motherboards based on it can take 66MHz FSB. Not all of them stably, but at least the systems light-up and you can do certain things with them.

About number 2 issue with Windows not able to fully shutdown. I have seen this problem on other UMC boards. So you are not alone here. : )

Are you sure the level 2 cache is actually operational ?
The speedsys and cachecheck suggest otherwise.
Judging by them i am certain L2 is not ticking.
But i think the fastest BIOS timings for DRAM are compensating to quite extent.
This is very interesting because your numbers are only slightly lower than the known top scores.
If you can get the L2 cache up and running you may be able to outdo some of the current best boards.

Did you run some more involving tests, or for not you are staying with some of the common videogames from the time period ?

Will link your post in the directory on the first page.

Sphere478 wrote on 2024-03-20, 04:33:

Oh that’s sweet!!!

Any chance of a 3dm 2000 run?

Can you compare with radeon 9250 or radeon 7500?

Unfortunately, on this system 3Dmark 99 MAX does not start even with the 486-mode under no conditions I tested on other UMC-boards with a GF 2 in the past , too (98SE or ME; some few newer driver-versions than the 6.31) - see this thread: Modern graphics on a 486

If there is a 486-mode for 3Dmark 2000, it would be very helpful and nice if you upload it here once again, thank you - so I can try it.

I will take a look for a R9250 or R7500 the next days.

pshipkov wrote on 2024-03-20, 07:10:

Did you target the SYL8884PCI-EIO motherboard because of that other thread you linked, or it fell in your lap by coincidence ?

To be honest, it was pure luck, so yes, "fell in my lap by coincidence" 😀

Thank you very much for the info about the L2-cache - today I will replace all the modules with another ones having 15 ns, and see what happens. I hope, they also can work at 2-1-2 at FSB 50 MHz

Can the issue with no booting at FSB 66 by caused by the VGA (too high frequency)? I tested this with 2 other VGAs (GF 2 MX 100 and Matrox Mystique) - sadly, no success, too.

By the way: in this thread Modern graphics on a 486 I had reach some very good results in GLQuake and Doom (even the UMC-boards are not stable), maybe you can add them with the remark "not stable system" to the charts, if this make any sence for the statistic. For example, there is a result for GLQuake @800x600 dpi of 38 FPS (AMD 486 @ 180 MHz / GF 2 MX 400 / HOT-433).

BTW, I see, in my posting from yesterday they are missing the screen-shots from the BIOS - here they are.

Well, the L2-cache on the SYL8884 is actually really NOT working (thanks to pshipkov once again!).
How I tested it – I simply removed all cache-chips incl. the TAG-module and made all DOS-tests once again. All scores are exactly identical to yesterday (with the L2-chips inserted).

In the next step I changed all chips with another ones (all 5 pieces) – sadly, the same results!
To be sure, I did this once again – still the same results!

I can not believe, that all chips are broken...

So I tested 256 KB L2 instead of 512 KB (different chips, again) – still the same results (= L2 inactive)!

Finally, I tested 128 KB L2 with other chips – and it works (at the second POST-screen there is shown „128 KB Cache enabled“! For 50 MHz FSB, I had to adjust only the cache-timings at 2-2-2.

The big disappointment: in Speedsys 4.78, the RAM-speed is greatly reduced, and the speed of the L2-cache itself is LOWER THAN OF THE RAM-modules without L2-cache activated (see yesterdays measurement)! Inside of Windows, using 16 MB of RAM + 128 KB L2, the performance is terrible (or something is still wrong with the L2-cache)! By the way – CACHECHECK still means, there is no L2!

This let me think, that maybe the STABILITY of this board at a high FSB is caused BY THE MISSING (NOT WORKING/DISABLED) L2-CACHE!

In the next step I tried to use 3 other boards WITHOUT L2-cache with the same RAM-module of 64 MB EDO and 50 ns and see what happens in DOS (at HOST : PCI = 1 : 2):

the ECS UM8810P-AIO Rev. 1.1 (latest BIOS) was not able to boot with the Am486 @ 200 Mhz (even 3x 50 MHz = 150 MHz was possible), and no boot at 3x 60 = 180 MHz, too
a HOT-433 Rev. 4 was able to start at 3x 66 MHz and do my DOS-tests, but entering the BIOS they are lots of vertical blue-white lines with the GF 2 MX visible, so I think the Northbridge is not stable enought
a HOT-433 Rev. 1 was able to start at 3x 66 MHz, too, sadly not using the GF 2 MX (because of its PCI 1.0 – version), but a Matrox Mystique (not interesting for me at this time)

In addition, the L1-measurements of both HOT-433 at 66 MHz compared to the SYL8884 at 50 MHz are worse/slower, the same fact is for the RAM-results (because of the increase of the DRAM-waitstates from 0 to 1).

So, in conclusion, the SYL8884 seems to be a really very fast mainboard and (regardless of its maybe cheap design and components) it is suitable for the building of a very fast 486-system!

In the next days I will try to find a solution for the L2-problem on this board. If the L2 can not be enabled for what reason ever, this should not be a big problem because of the excellent measurements without L2 activated.

By the way: the total RAM-amount in all Speedsys-tests with the SYL8884 is reduced by 2 MB (e.g. 14 MB instead of 16 MB; 62 MB instead of 64 MB). Very strange....

and the rest of todays pics

On one of your screenshots, there is "Write Back Caching Accelerated".
Usually, when the Bios writes some blabla about Write Back, it means it is a fake cache Bios, meaning it always shows 256kB Cache even if there is none available.
But maybe not the case here, but I guess you should re-consider of updating the Bios.
Where is the menu setting about disabling / enabling L1 and L2 cache?
I mean L1 cache is working fine, but obviously L2 cache is disabled.

But as you already have written by yourself, this board is quite fast without L2 cache, so you also can just remove all the cache chips and call it a day.

EDIT: Your last screenshot shows, that the L2 cache is actually working. ....
Maybe not too much of a gain. I don't know.

Ok, L2 cache seems to be back in action, but reduced to 128Kb.
With 15ns rated chips the 2-2-2 timings is actually very good, but the smaller buffer is helping there.

L2 cache is always the weakest point for 486 overclocking past the 40MHz FSB.
If you can - spend some time to rotate the available chips around - there is chance you find a working 256Kb configuration.
This part is always annoying and takes longer than anticipated.

If you run some of the simpler tests like Doom, Quake 1, and GL Quake - what are the results with 128Kb L2 cache ?

Also, you didn't clarify - are you using air cooled CPU or there is a Peltier in the mix ?
I am asking this because active CPU cooling helps with all sorts of seemingly unrelated issues - for example suddenly L2 cache works, or 66MHz FSB becomes possible and so on.

gonzo wrote on 2024-03-20, 18:02:

I can not believe, that all chips are broken...

During Bios L2 initialization when SRAM is not fast enough for your ridiculous FSB overclock it fails to be detected

gonzo wrote on 2024-03-20, 18:02:

Finally, I tested 128 KB L2 with other chips – and it works (at the second POST-screen there is shown „128 KB Cache enabled“! For 50 MHz FSB, I had to adjust only the cache-timings at 2-2-2.

you found one set of fast enough SRAMs

gonzo wrote on 2024-03-20, 18:02:

The big disappointment: in Speedsys 4.78, the RAM-speed is greatly reduced, and the speed of the L2-cache itself is LOWER THAN OF THE RAM-modules without L2-cache activated (see yesterdays measurement)! \

Read speed will probably always suffer with L2 enabled, you need time to look up if address is cached in TAG ram. Bad implementations will probably wait for that read before starting ram access so you suffer additional ram latency. You would need 2-1-1 to show advantage.

gonzo wrote on 2024-03-20, 18:02:

By the way – CACHECHECK still means, there is no L2!

Its confused because everything is too fast, you can clearly see 16 to 19ns step at 128KB

gonzo wrote on 2024-03-20, 18:02:

In the next days I will try to find a solution for the L2-problem on this board. If the L2 can not be enabled for what reason ever, this should not be a big problem because of the excellent measurements without L2 activated.

its already active, you would need to experiment with 10ns SRAM chips, cant get those in DIP, requires making SOJ-DIP adapters.

gonzo wrote on 2024-03-20, 18:02:

By the way: the total RAM-amount in all Speedsys-tests with the SYL8884 is reduced by 2 MB (e.g. 14 MB instead of 16 MB; 62 MB instead of 64 MB). Very strange....

1MB cacheable BIOS
1MB 15-16 hole

pshipkov wrote on 2024-03-20, 19:50:

Ok, L2 cache seems to be back in action, but reduced to 128Kb. If you can - spend some time to rotate the available chips around […]
Show full quote

Ok, L2 cache seems to be back in action, but reduced to 128Kb.
If you can - spend some time to rotate the available chips around - there is chance you find a working 256Kb configuration.
If you run some of the simpler tests like Doom, Quake 1, and GL Quake - what are the results with 128Kb L2 cache ?
Also, you didn't clarify - are you using air cooled CPU or there is a Peltier in the mix ?

I reduced the L2 to 128 KB with appropriate chips by myself (it was just a test), but the performance in Win98 with 16 MB RAM is terrible, as I wrote yesterday. Big problems to start every software. So I am not sure if the 128 KB-L2 is working correctly. Sadly, no 3D-tests with 128 KB at the moment.

Yes, I will test some other chips for 256/512 KB L2. Still not sure, but as "rasz_pl" means in his posting, maybe the L2-cache is working, but the high FSB causes a wrong interpretation in Speedsys and Cachechk. So I will try if the L2 can be "found" by the software at a lower FSB (33/40 MHz).

For the CPU-cooling I use a cooler from socket 7/370 with a fast (high rotating) fan mounted on it. Unfortunately, one of the both white plastic-"noses" at the 486-socket of the SYL-board was broken, so I used thermal glue (a good expensive one!) to mount the cooler on the CPU. I am pretty sure, they are no problems of overheating. As shown above, yesterday I tested the same CPU at 3*66 MHz FSB on two HOT-433 (Rev. 4 and Rev. 1) without any problems for the CPU.

CoffeeOne wrote on 2024-03-20, 18:11:

Where is the menu setting about disabling / enabling L1 and L2 cache?
I mean L1 cache is working fine, but obviously L2 cache is disabled.

Yes, there is a BIOS-option for enabling/disabling the L2-cache. The measurement-results are equal, regrdless of "enabled" or "disabled", and regardless of L2-chips inserted into their sockets, or not inserted (= L2-sockets are empty).

by the way, I am still missing some comments about my idea from yesterday:
maybe the STABILITY of this SYL-board at a high FSB is caused BY THE MISSING (NOT WORKING/DISABLED) L2-CACHE!

I will test this idea in Windows in few days, using a HOT-433 Rev. 4, witch works at 60/66 MHz FSB, - until now, this board was equipped always with L2-cache and it was always unstable at 60 MHz FSB (this problem is common for all other boards except of the SYL-board I tested in the past).

rasz_pl wrote on 2024-03-20, 22:37:

you found one set of fast enough SRAMs

Thank you for all the infos, rasz_pl, maybe this one sentence is the most important for the L2-problem 😀
I think, maybe my TAG-module of 12 ns is the problem here.
Today I will test if the L2-cache can be recognized at lower FPS (33/40 MHz).