Not best but not so bad speed (usually at late 486 chipset L2 speed is 41-42MB/sec with SPEEDSYS)
Sad there no MR.BIOS for ALI M1487 but there are AWARD/AMI/PHOENIX BIOSes for M1487
Maybe you can try them?

Its running an AMI bios at the minute, tried the 3 bios files that are listed along side the board I using a jumper reference and it won't post with those!

There are a few other M1487 boards out there so can certainly try a few more.

https://theretroweb.com/motherboards/s/tmc-my … ci48af-ver.-3.0
https://theretroweb.com/motherboards/s/newtec … t486-terminator
https://theretroweb.com/motherboards/s/newtec … smt-smt4321-22b
https://theretroweb.com/motherboards/s/newtec … smt-smt4321-22a
https://theretroweb.com/motherboards/s/edom-wintech-mp060
https://theretroweb.com/motherboards/s/esa-tf-486 (1999-2000!)
https://theretroweb.com/motherboards/s/gigabyte-ga-5486al
https://theretroweb.com/motherboards/s/abit-ab-pb4-rev-1.3

thanks, will work through these later

CRG wrote on 2026-08-13, 10:39:

A few weeks ago I purchases an old CNC computer

loved both videos! 😀

CRG wrote on 2026-08-13, 10:39:

The issue comes in that the L2 cache will not perform better than about 45MB/s when tested with cachechk or 34MB/s in speedsys. The fastest timings I can set in the bios are 3-2-2-2.

The issue is indeed L2 timing. Theoretical L2 3222 is close to EDO theoretical max 5222. Note that Acer/Ali datasheet is liberal with facts aka lies https://theretroweb.com/chip/documentation/m1 … cc190749081.pdf
they claim:
(fastest) FPM setting corresponds to 2-2-2
(fastest) EDO setting corresponds to 1-1-1
which is impossible due to the way DRAM works 😀 EDO still needs two cycles, one for CAS to go low - ram locks in address, outputs data, and another where CAS goes high - EDO still outputs data, ignores address.
So in reality you have to add 1 cycle to Acers imagination 😀
EDIT: I was in fact wrong, chipset does 1-1-1 and this is possible with 60ns EDO ram because one 33MHz cycle is 30ns and fits witching "Fast EDO Page Cycle Time: 25ns". I am an idiot 😀 I was in fact remembering 2-2-2 being the minimum on Pentium boards, and those run at 66MHz.

posts like this Some EDO results for ALI M1487/89 make me suspect Acer was liberal with truth in more places, maybe that chipset doesnt support EDO timings at all? 😀

https://www.youtube.com/watch?v=YHSdYeh1lbU @16:05 you got SRAM and DRAM confused around
Hidden refresh has nothing to do with CPU doing refresh, it hides ram refresh from ISA bus.

You change too many things at once, like EDO and CPU in one go.

Dont rely on "memory bandwidth" displayed by Speedsys. I realize its very convenient and neat single number, but its also a Lie 😀 We have to ask outselfs what does it represent exactly? "memory bandwidth" of what? because that number is Higher than any other singular speed rating Speedsys shows you (L1, L2, mem throughput). So what is it? it might be theoretical speed of reading 4 bytes in a loop from L1 or something stupid like that, aka its useless.
Speedsys is not the greatest for this. Use CACHECHK for everything.

What CACHECHK shows is not slow cache, but EXTREMELY fast ram for 33Hz FSB 486. Here another AMD 133Mhz for comparison Re: Compaq (197005-001) cache module - open source reproduction
in fact your Ram is so fast CACHECHK cant tell you have any L2 enabled 😀 Try EDO with no L2, try FPM with no L2, FPM with L2. Compare all in CACHECHK.

CRG wrote on 2026-08-13, 10:39:

The initial thought was dodgy refinished cache so I've tested another set of known good 15ns chips although just 128kb of cache and the results are identical.

try with FSB set to 20MHz fsb = open jp5 and jp6. I would experiment with other permutations of jp5 and jp6, maybe you can get stable 40Hz fsb, and last combination might be 16MHz fsb - also handy for checking L2 stability.
If you can make L2 3-1-1-1 work in lower FSB you will know problem is reprinted slow SRAM chips. If you make it work again test with CACHECHK. Make screenshots of everything.

try with EDOM Award BIOS if the manual fits 100%.

CRG wrote on 2026-08-13, 10:39:

L2 cache is set to write back in the bios.

WB cache is great if
1 your cache subsystem supports Dirty bit
2 L2 is significantly (lets say at least 30-50%) faster than RAM
3 depending on workloads
Manual for m1489 mentions 7+1 and 8+always dirty configurations. always dirty mode will make cached ram way slower.

CRG wrote on 2026-08-13, 10:39:

Some of you might have seen my videos on this including the gaff I made by mixing up SRAM and DRAM (idiot)

me meee, teacher mee! I seen it! 😀
You can tell Im replying line by line instead of reading whole post.

Thanks for the detailed info rasz_pl.

I spent no less than 4 hours last night trying every bios I can get my hands on with some interesting results.

It seems its only the Cardinal Ami bios that doesnt like any sram timing other than 3-2-2-2. But one odity with that bios compared to all others is that it needs that jumper changed as per the 486. All other ami bios roms and indeed award bios roms need the jumpers set as per the diagram for the 5x86 to even boot!

In terms of speed, award roms lets the cache go down to 2-1-1-1 but the results aren't really any faster.

This is running on award and yoy can now see the difference between L2 and the EDO.

So just use the award bios?

Not that simple, with any bios other than the Cardinal one the floppy drives dont work! With most Award bios roms you also get a conflict at port 278 (even if I disable the parallel port).

Running the award bios rom, the boards not stable at 40mhz fsb whereas the cardinal ami bios seems happy enough.

Trying to use the bios from the board I'm using for jumper settings doesnt post.

So I'm left thinking there is some oddity with this board, it is what it is and I just need to use it as is.

I found another pci skt 3 board in my collection (forgot I had it), its the same chipset but a pc chips with fake cache. Does look like it'll take real so I think we need to sort that next tgen decide which board is better to move forward with.

Please make and compare screenshots of SpeedSys' cache chart to compare your settings.
You can save the charts as .PCX file and convert them to .PNG file to post them here. (Please no .JPG's of them)

NV8T is the original, write-through only Am486DX4.
SV8B is an Enhanced Am486DX4, that uses compatible pinout with the IntelDX4WB aka "P24D, but 3.3V".
SV8B/5X86-P75 is the correct jumpering for your Am5x86. You can see there is only one different jumper. That must be the CLKMUL. If the jumper is open your Am5x86-P75 will act like an Enhanced Am486DX4 and run at 100 MHz. With the jumper closed it will run at 133.

On pages 3-4 https://www.ardent-tool.com/CPU/docs/ST/486_5x86/4193.pdf are the pinout differences between different 486 CPUs that have to be dealt with using jumpers. That will give you an idea where they go.

Having a mixture of L1 write-through and write-back jumpers will invariably give you a wonky system. But it probably doesn't explain why 3-2-2-2 works and 2-1-1-1 doesn't.

Out of simplicity, are you sure you're running at 33 MHz and not say 40 or 50? Single bank is unlikely to work at -1-1-1-1 at 40 or 50, so that would instantly explain it.

Note, that some consider the EDO support in the M1489 so fast that they consider disabling the external cache. So you might do some benchmarks before concluding you definitely want the cache.

edit: here's another note https://www.ardent-tool.com/CPU/docs/AMD/486_5x86/19178.pdf this time from AMD, suggesting how to support "standard" (NV8T) and "enhanced" (SV8B) CPUs on the same motherboard.

CRG wrote on 2026-08-14, 17:09:

The attachment 20260813_234423.jpg is no longer available

jakethompson1 wrote on 2026-08-14, 19:43:

Note, that some consider the EDO support in the M1489 so fast that they consider disabling the external cache. So you might do some benchmarks before concluding you definitely want the cache.

yep, thats why I suggested "Try EDO with no L2, try FPM with no L2, FPM with L2. Compare all in CACHECHK."

Yes, and please save and send pictures of the SpeedSys charts.
We can read so much about your board's configuration out of it.

Thanks gents, I'll get some screen shot up later.

Picking up on a few comments I'm 100% sure its running at 33mhz fsb. I can switch it to 25 or 40mhz and the clock adjusts accordingly. Side note, one of the bios roms I tried when set to a 40mhz fsb posts showing an AMD 5x86 P90. Was that ever a thing?

5x86 133 - P75 (4x33 MHz)
5x86 150 - P75+ (3x50 MHz)
5x86 160 - P90 (4x40 MHz)

The latter ones were ultra rare or unicorns.

However, most of the 5x86 133 should run at 4x40 MHz with standard cooling and voltage.
For attempts on 3x60 and 4x50 MHz you can read in other threads here.

Ok I've spend a few hours and did lots of testing with the original Ami BIOS vs Award BIOS and combinations of 512k L2, no L2 along with FPM or EDO.

I'll have to split this into several replies starting with AMIBOS, 32mb FPM (2x16 60ns)

No L2 cache - cachechk

No L2 cache - speedsys

512k L2 cache - cachechk

512k L2 cache - speedsys

As noted previously the system won't get past post on this bios unless cache is set to 3-2-2-2. The CPU jumpers here have also been set as described for the 5x86 P75 (from the post above).

In particular note the terrible L2 performance.

Next up is the same AMI BIOS but switching to EDO ram

Not much of any speed increase to speak off changing to EDO and equally poor L2 performance!

Next up we change to an Award bios but back to FPM memory.

This will let me push the cache out to 2-1-1-1 but the floppy drives don't work!!!

Note a similar system ram speed to before but once we add the L2 it's now performing a lot better.

Lastly we have Award BIOS wit EDO (2x 16mb 60ns)

The interesting thing here is that without the cache the system memory performs really well, it seems its as fast as the L2 when that's installed although with L2 the system ram seems to slow!

Thank you for posting those. Those cache timings are far worse that 3-2-2-2. Those are de-turbo like speeds.

One of the differences between the NV8T and 5x86 pinouts also is going to be System Management Mode (power management). Chkcpu's 5x86 upgrade page alludes to this: http://www.steunebrink.info/amd5x86.htm so an easy test would be to completely disable power management in the BIOS.

For comparison here is a tuned UMC 8881+8886B. It is right inline with yours that has the broken floppy support.

1CACHECHK v4 2/7/96  Copyright (c) 1995 by Ray Van Tassle. (-h for help)
2 CMOS reports: conv_mem= 640K, ext_mem= 64,512K, Total RAM= 65,152K
3 "AuthenticAMD"  486 Clocked at 133.6 MHz
4 Reading from memory.  
5 MegaByte#:      --------- Memory Access Block sizes (KB)-----
6       1    2    4    8   16   32   64  128  256  512 1024 2048 4096 <-- KB
7 0:    8    8    8    8    8   18   18   18   18   18   --   --   --    µs/KB
8 1:    8    8    8    8    8   18   18   18   18   18   28   28   28    µs/KB
9 2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18  <--- same as above.
1019 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35  <--- same as above.
1136 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52  <--- same as above.
1253 54 55 56 57 58 59 60 61 62 63  <--- same as above.
13
14 Extra tests----
15 Wrt  16   16   16   16   16   16   16   16   16   16   16   16   16<-Write mem
16 This machine seems to have both L1 and L2 cache. [read]
17    L1 cache is  16KB -- 137.4 MB/s   7.6 ns/byte  (346%) (221%)   3.9 clks
18    L2 cache is 512KB --  62.1 MB/s  16.9 ns/byte  (156%) (100%)   8.6 clks
19    Main memory speed --  39.7 MB/s  26.4 ns/byte  (100%) [read]  13.5 clks
20    Effective RAM access time (read ) is 105ns (a RAM bank is 4 bytes wide). 
21    Effective RAM access time (write) is  60ns (a RAM bank is 4 bytes wide). 
22    "AuthenticAMD"  486 Clocked at 133.6 MHz.  Cache ENABLED.
23 Options: -t0

And here is a tuned SiS 496

1CACHECHK v4 2/7/96  Copyright (c) 1995 by Ray Van Tassle. (-h for help)
2 CMOS reports: conv_mem= 640K, ext_mem= 31,744K, Total RAM= 32,384K
3 "AuthenticAMD"  486 Clocked at 133.6 MHz
4 Reading from memory.  
5 MegaByte#:      --------- Memory Access Block sizes (KB)-----
6       1    2    4    8   16   32   64  128  256  512 1024 2048 4096 <-- KB
7 0:    8    8    8    8    8   18   18   18   18   26   --   --   --    µs/KB
8 2:    8    8    8    8    8   18   18   18   18   26   26   26   26    µs/KB
9 3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19  <--- same as above.
1020 21 22 23 24 25 26 27 28 29 30 31  <--- same as above.
11
12 Extra tests----
13 Wrt  16   16   16   16   16   16   16   16   16   16   16   16   16<-Write mem
14 This machine seems to have both L1 and L2 cache. [read]
15    L1 cache is  16KB -- 137.7 MB/s   7.6 ns/byte  (326%) (221%)   3.9 clks
16    L2 cache is 256KB --  62.2 MB/s  16.9 ns/byte  (147%) (100%)   8.6 clks
17    Main memory speed --  42.2 MB/s  24.9 ns/byte  (100%) [read]  12.7 clks
18    Effective RAM access time (read ) is  99ns (a RAM bank is 4 bytes wide). 
19    Effective RAM access time (write) is  61ns (a RAM bank is 4 bytes wide). 
20    "AuthenticAMD"  486 Clocked at 133.6 MHz.  Cache ENABLED.
21 Options: -t0

That looks like de-turbo mode.
Have you already played with turbo switch or turbo jumper?

Then just for the fun of it I pushed the CPU to 160mhz with a 40mhz fsb and ram cachechk on the award bios with both L2 installed and not installed.

Again note that system ram seems to perform as fast as L2 when there's no L2 installed!

Picking up the point on the turbo button. The boards turbo jumper is open which I assume is turbo off? I can do more testing changing the turbo switch later.