Chkcpu wrote on 2022-01-17, 20:35:

Here is the DX-6900/M912 VER 2.21r Award BIOS with patch J.1. Apart from correcting the fake 256K Cache Memory display, this pat […]
Show full quote

Here is the DX-6900/M912 VER 2.21r Award BIOS with patch J.1.
Apart from correcting the fake 256K Cache Memory display, this patch J.1 BIOS it identical to the original 09/08/95-UMC-498GP-2C4X6B13-00 VER 2.21r BIOS.
I’m now working on a patch J.2 version to add full Am5x86-P75 support.

Jan

Yes your J.1 Patch works perfectly with my M912, better then AMI Bios, my config on this Borad is Cache with 512 kbytes of 4x Chips with 128k x8 -15ns only in Bank 0. In the Post Screen it is correctly shown as 512 kbytes Cache !

Are you ready with your J.2 Patch ? my current CPU is AM486DX120 with WB, it works good with your J.1 Patch, but i want to upgrade to x5-133 CPU.

Does WB work ok (on both L1 and L2) with the Award bios?

Henk386 wrote on 2022-10-01, 19:23:

Are you ready with your J.2 Patch ? my current CPU is AM486DX120 with WB, it works good with your J.1 Patch, but i want to upgrade to x5-133 CPU.

Hallo Henk386,

Thanks for your positive report about the M912 patch J.1 Award BIOS!

feipoa, the OP of this thread, tested the patch J.2 version of this BIOS but found several issues.
Although the Am5x86-133 and Cx5x86-133 now run fine on this J.2 version, the incorrect speed indication on the Am5x86 still needs to be fixed.
Also the lower IDE transfers speed of this Award BIOS troubles me.
I still need to find solutions to these issue before releasing a definitive patched M912 Award BIOS.

Unfortunately, we don’t have the UMC498F chipset datasheet. So I may have to find a way to reverse engineer the complicated AMI WinBIOS to see what it does differently from Award.
This is turning into a bigger project than I thought. 🙁

I have started on a (definitive) patch J.3 BIOS but it is still early days.
In the meantime, I’m releasing the stable (but not yet perfect) M912 patch J.2 BIOS. 😉
When you are going to test it, please let us know how your Am486DX4-120 is reported by the BIOS.
feipoa’s report on this J.2 BIOS can be read on page 6 of this thread.

So, why going through all the trouble of patching an Award BIOS while the AMI 1995X works fine!??
I got a lot of requests for a better M912 Award BIOS, both on Vogons and via e-mail. I don’t know if most of them were v1.4 or v1.7 users, but several M912 owners said they preferred the Award BIOS and a few told me they disliked the AMI WinBIOS intensely. 😉
Anyway, this 09/1995 Award BIOS appeared to be a great showcase for my DIY BIOS patching story here on Vogons!

Jan

Am386DX-40 wrote on 2022-10-03, 14:41:

Does WB work ok (on both L1 and L2) with the Award bios?

yes both works perfectly L1 and L2 cache are both working very well, see it at the speed sys 4.78 capture.

Chkcpu wrote on 2022-10-03, 15:41:

When you are going to test it, please let us know how your Am486DX4-120 is reported by the BIOS.

So, why going through all the trouble of patching an Award BIOS while the AMI 1995X works fine!??

So the AMI BIOS not works so good with my system with the AMI i had w3.11 crashes with AWARD not. So i also can get some better Memory performane with the AWARD.

I will test your j.2 the next days intensively.

regards Hendrik

So some short test with dx120 and dx133 with J.2 BIOS, there is no advantage between J.1 and j.2 using a DX120 i think J.2 works fine with j.2.

when switching to x5-133ADW CPU with is normaly 4x 33 Mhz i could not get any jumper settings for multi =4 on this board, i tried some setting on page 3 and 4 in this posts and some different from the pcchips manuals, but sometime system boot and sometime not...

it shows every time "Am486DX4-S CPU at 100 Mhz"

this means every time 33Mhz x3 and no other CPU-id with have normaly shown with the x5 CPU, some at this time i could not get Patch J.2 work with an DX5 CPU.

It is strange is already think that the x5 Amd with 486 core translate the multi from 2 to 3 and from 3 to 4 ?

So your J.2 Patch runs the x5-133ADW now i switched the BusClock from 33 to 40 Mhz and it is shown as an normal DX120 WB ? So running System Speed Info shows different CPU-id but the 16kbyte L1 Cache of that x5-133 ??
So the Cache works, yes it is 16kbyte is that size from the X5-cpu but CPU-ID and System Speed is exactly the same from the DX120-x4 CPU.

I think the multi is not translated from 3 to 4 ? and thats the reason why it is the same speed, is the translation implemented in the BIOS ?

regards Hendrik

So i found out some really useful information about the J.2 Patch and the x5-133 cpu !

So sorry about that first jumper for multi must to be set to "2". If it is set to "2" on the board the cpu translate it to "4" internally.

But it is something wrong with the J.2 Patch and Post Screen for CPU id.

Multi x4 works now, but the x5 cpu was shown in the post screen with..

a.)
BusClock=33 > Post Screen 100 Mhz > real = 133 in system speed 4.78 and also real with points compared to dx120
b.)
BusClock=40 > Post Screen 132 Mhz > real = 160 in system speed 4.78 and also real with points compared to dx120

Here the screenshots taken from settings BusClock=40 multi=4

If seems that there is an error offset of 33 Mhz when x5 cpu is installed with J.2 Bios ?

By the way...i have to install a voltage socket converter, because the mainboard offers only 3,3 volt, that is to low for an overclocked 133 to 160 x5 cpu, but 3,45volt with the voltage soket converter works fine with oc at 160 Mhz.
regards Hendrik

@Henk386,
Thanks fort he detailed test reports on the M912 patch J.2 BIOS!
Especially the 132MHz reading on the POST screen when running the Am5x86 at 4x40 is very interesting. It looks like the BIOS is detecting the FSB wrong when using the x4 multiplier!
When running at 4x33, the BIOS thinks it is running at 4x25, and with the 4x40 setting the BIOS indicates 132MHz which is 4x33!

Do I understand you correctly that the BIOS reports the correct 120MHz when running the Am486DX4-120 or Am5x86-133 at 3x40? And that the speeds are only wrong when using the x4 multiplier?

To help me troubleshoot this bug, can you make a memory dump of a part of the BIOS Data area?
Just run Debug from the DOS-prompt and enter this dump command from the Debug prompt -
d 0:480
Then quit Debug with the q command.

Please make 2 dumps, one with the Am5x86 at 3x40 and the other at 4x40.
I’m very interested in the values at memory addresses 04D8h and 04D9h.
I hope this will confirm my FSB bug hypothesis. 😉

Thanks, Jan

Chkcpu wrote on 2022-10-04, 21:12:

Please make 2 dumps, one with the Am5x86 at 3x40 and the other at 4x40.
I’m very interested in the values at memory addresses 04D8h and 04D9h.
I hope this will confirm my FSB bug hypothesis. 😉

So i hope this is helpful for you, i have taken to screenshot dumps
a.) x5-133 CPU, with BusClock=40 and Multi=3
b.) X5-133 CPU, with BusClock=40 and Multi=4 (but beware of Multi=4 it is jumpered on the mainboard with 2, because of the fact that no jumper ist possible for 4 physically)

perfect if you will fix it sometime, i have taken about 19 different configurations on that board for 160cpu but i think i have find one with agressive settings, but i must wait for ultra high speed cache that i have ordered, now i have only 15ns cache and will upgrade to 10ns cache, and hope that would be the key to operate 160mhz with agressive working in all aplications.

Chkcpu wrote on 2022-10-04, 21:12:

@Henk386,
Do I understand you correctly that the BIOS reports the correct 120MHz when running the Am486DX4-120 or Am5x86-133 at 3x40? And that the speeds are only wrong when using the x4 multiplier?

BIOS reports correct 120 Mhz when running Am486DX4-120 at 3x40Mhz

regards Hendrik

Hi Henkdrik, thanks for the Debug dumps. It makes the issue crystal clear.

The values from the BIOS Data area show me that the CPU type and multiplier setting are correctly detected in each case, but that the FSB detection is incorrect when the multiplier is x4.
In that case the BIOS thinks the FSB is 25MHz when it actually is 33MHz, and 33MHz when it actually is 40MHz. Consequently the reported CPU speed is incorrect as well.

Now what causes this is still a mystery, but I know now where to look for. 😉

I will report back when I have more.

Cheers, Jan

Hello Jan,

Thanks for your answer, i am very glad about you, that you have the advantage of programming an finding errors for such old BIOS in this case it makes life easier to handle with vintage Hardware.
So i am testing very intensively this board, i have perfect setting for DX4-120 @ 120 and DX5-133 @ 133, but 133 is much slower in every 3D-application Benchmark, only some better but very small results about L1 Cache. So it make no sense to upgrade from 120 to 133 in result all Games performs slower. But i am working on perfect agressive settings for DX5-133 @ 160, they need very high speed compoments, so i am waiting for ultra speed cache, they are in shipping process, perhaps in 4 weeks i will do again testing process for dX5-133 @ 160, the performance is great, but have some cache problem, i hope with the new cache DIP IC´s not....
i will report here if settings are stable.

regards Hendrik

Henk386 wrote on 2022-10-06, 11:12:

Hello Jan, […]
Show full quote

Hello Jan,

Thanks for your answer, i am very glad about you, that you have the advantage of programming an finding errors for such old BIOS in this case it makes life easier to handle with vintage Hardware.
So i am testing very intensively this board, i have perfect setting for DX4-120 @ 120 and DX5-133 @ 133, but 133 is much slower in every 3D-application Benchmark, only some better but very small results about L1 Cache. So it make no sense to upgrade from 120 to 133 in result all Games performs slower. But i am working on perfect agressive settings for DX5-133 @ 160, they need very high speed compoments, so i am waiting for ultra speed cache, they are in shipping process, perhaps in 4 weeks i will do again testing process for dX5-133 @ 160, the performance is great, but have some cache problem, i hope with the new cache DIP IC´s not....
i will report here if settings are stable.

regards Hendrik

Could you provide a list of tests you perform? (I suppose Phil's + others?)
I'd like to give it a go this weekend. I don't have much time, but if I don't have to look for myself, following a list, I'd try "bot-like" (brain dead 😀).

Thanks a lot to Jan for his help.

Edit:
What are the settings I should enter in BIOS to make it as fast as possible?

Nexxen wrote on 2022-10-06, 12:42:

Could you provide a list of tests you perform? (I suppose Phil's + others?) I'd like to give it a go this weekend. I don't have […]
Show full quote

Could you provide a list of tests you perform? (I suppose Phil's + others?)
I'd like to give it a go this weekend. I don't have much time, but if I don't have to look for myself, following a list, I'd try "bot-like" (brain dead 😀).

Thanks a lot to Jan for his help.

Edit:
What are the settings I should enter in BIOS to make it as fast as possible?

Hi there...,

Yes some of the collection of Phil´s dosbench 1.4, but also some real working test.
my BIOS setting are tested by following test`s
- CPU identification Utility v1.25
- cache chk 4
- System Speed Test v4.78
- Superscape 3D Bench 1.0
- Chrisbench 320 & 640
- Topbench 0.38a
- Doom Demo at Full Details
- PC Player
.....also testet
- Windows 3.11 connecting via LAN to NAS Drive, copy files from NAS and from drives to drives
- playing audio on Windows 3.11
- playing DOOM2 at MS-DOS v1.66 at a minimium of 30 Minutes
- several cold and warmboots

so now i offer the Setting for an AMD DX4-120 with 8 kByte L1-cache and WriteBack for the BIOS Patch J.2 for AWARD-BIOS m912

please notice ! the settings are depending on the Hardware with is not common for some people! it depends of which cache chips and which RAM chips you use and will not work for all other common RAM Modules and Cache Chips, notice this could be the fact if the setting might not work on your M912 !

Hardware for the following BIOS Settings are:

- only one bank of RAM installed, in PS/2 Socket, the Module is about 64 MByte with parity and 60ns. This modules are hard to find !
- cache with a total amount of 1024 kbyte L2 Cache with 15ns, don´t use 20 or 25 chips. chips model number is "IS61C1024-15N"
- also use tag chip with 15ns !
- no other VLB Cards except only one Graphic Card

now first the motherboard setting that are also important:
-JP16 = ON (Vesa Clock over 33)
-JP17 = OFF (Vesa Wait State = 0)

BIOS Setting now:
BIOS FEATURES SETUP......
- CPU Internal Cache = Enabled
- External Cache = Enabled
- Memory Parity Check = Enabled
CHIPSET FEATURES SETUP......
- Auto Configuration = Disabled
- DRAM Wait State select = 1 WS
- DRAM Page Mode = Fast
- L2 Cache Read Wait State = 3-2-2-2
- L2 Cache Write Wait State = 0 WS
- L1 Cache Update Scheme = Wr-Back
- Video BIOS Cacheable = OFF (might work ON on some other cards)
- ISA Bus Clock Option = CLKI/4
- Local Ready Delay Setting = Delay 1T
- Alt Bit in Tag SRAM = 7+1 Bits

get very fine result in System Speed, all working stable under stress testet over some days.
If you get no L1 Cache or no L2 Cache or no Write Back Mode, it could depending on your selected Hardware RAM and cache, that this settings are to agressive. In Future i will do some more test with 133@160, but this require more better Hardware on Cache and/or RAM Modules.

So i hope i could help you.

regards Hendrik

Henk386 wrote on 2022-10-06, 20:31:

Hi there..., […]
Show full quote

Nexxen wrote on 2022-10-06, 12:42:

Could you provide a list of tests you perform? (I suppose Phil's + others?) I'd like to give it a go this weekend. I don't have […]
Show full quote

Could you provide a list of tests you perform? (I suppose Phil's + others?)
I'd like to give it a go this weekend. I don't have much time, but if I don't have to look for myself, following a list, I'd try "bot-like" (brain dead 😀).

Thanks a lot to Jan for his help.

Edit:
What are the settings I should enter in BIOS to make it as fast as possible?

Hi there...,

Yes some of the collection of Phil´s dosbench 1.4, but also some real working test.
my BIOS setting are tested by following test`s
- CPU identification Utility v1.25
- cache chk 4
- System Speed Test v4.78
- Superscape 3D Bench 1.0
- Chrisbench 320 & 640
- Topbench 0.38a
- Doom Demo at Full Details
- PC Player
.....also testet
- Windows 3.11 connecting via LAN to NAS Drive, copy files from NAS and from drives to drives
- playing audio on Windows 3.11
- playing DOOM2 at MS-DOS v1.66 at a minimium of 30 Minutes
- several cold and warmboots

so now i offer the Setting for an AMD DX4-120 with 8 kByte L1-cache and WriteBack for the BIOS Patch J.2 for AWARD-BIOS m912

please notice ! the settings are depending on the Hardware with is not common for some people! it depends of which cache chips and which RAM chips you use and will not work for all other common RAM Modules and Cache Chips, notice this could be the fact if the setting might not work on your M912 !

Hardware for the following BIOS Settings are:

- only one bank of RAM installed, in PS/2 Socket, the Module is about 64 MByte with parity and 60ns. This modules are hard to find !
- cache with a total amount of 1024 kbyte L2 Cache with 15ns, don´t use 20 or 25 chips. chips model number is "IS61C1024-15N"
- also use tag chip with 15ns !
- no other VLB Cards except only one Graphic Card

now first the motherboard setting that are also important:
-JP16 = ON (Vesa Clock over 33)
-JP17 = OFF (Vesa Wait State = 0)

BIOS Setting now:
BIOS FEATURES SETUP......
- CPU Internal Cache = Enabled
- External Cache = Enabled
- Memory Parity Check = Enabled
CHIPSET FEATURES SETUP......
- Auto Configuration = Disabled
- DRAM Wait State select = 1 WS
- DRAM Page Mode = Fast
- L2 Cache Read Wait State = 3-2-2-2
- L2 Cache Write Wait State = 0 WS
- L1 Cache Update Scheme = Wr-Back
- Video BIOS Cacheable = OFF (might work ON on some other cards)
- ISA Bus Clock Option = CLKI/4
- Local Ready Delay Setting = Delay 1T
- Alt Bit in Tag SRAM = 7+1 Bits

get very fine result in System Speed, all working stable under stress testet over some days.
If you get no L1 Cache or no L2 Cache or no Write Back Mode, it could depending on your selected Hardware RAM and cache, that this settings are to agressive. In Future i will do some more test with 133@160, but this require more better Hardware on Cache and/or RAM Modules.

So i hope i could help you.

regards Hendrik

Yes, you succeeded. Thanks

I have:

-15ns but 256KB
- 1 32MB 60ns
- Amd DX-4 100, 586 P75 133 + @160 and Cyrix 586.

I'll try to post my results in some benchmarks that I ran. At a point only L1/L2/Ram speed in Speedsys.
Ok, I hope it rains heavily this weekend, to let it hurt less 😀

I tested Bios X and D versions.
I'll try J2 + X + D again.

Chkcpu wrote on 2022-10-03, 15:41:

Also the lower IDE transfers speed of this Award BIOS troubles me.

I just found out about this conversation. Very interesting.
Also quite a coincidence, just as I am hitting that roadblock with the exact same Award BIOS and its DOS IDE speed. (Though I haven't yet put back the AMI WinBIOS to cross-check things. Nor did I benchmark my other VLB Multi-I/O card...)
In Windows 95, with the VLB-card's driver, the Atto benchmark result seems pretty good.

Quoting myself from here 486 VLB UMC-Chipset, what is it?:

I have an ISA Multi-I/O card that, according two to benchmarks, performs a bit better in DOS. 1566 kB/s instead of 1323 kB/s. But then I would lose the Windows 95 performance of the VLB card. In practice; loading up Doom 2 to menu screen using the VLB-one takes 15 seconds. With smartdrv 14 seconds. I suppose the SD card's access time is helping out here.
The VLB Multi-I/O card FAQ says, that one is not supposed to use its DOS sys driver when LBA scheme is enabled. LBA is essential for interfacing the 8GB Drive.
Another thought is that the AMI BIOS "IDE 32-Bit Mode" toggle is surely not available, nor working with the Award BIOS that I am using. And I just found that exact concern mirrored in this recent topic pages 6 and 7.

I tried to find out more about the IDE speed issues on my 486 VLB system.
On second look, it is not an Award BIOS issue, nor is it Cx5x86 specific. Since I got the same low 1323 kB/s IDE speed with the original AMI WinBIOS from 1993 using an Am486DX4. That is with LBA, Block Mode and 32-bit enabled in the BIOS.

I did find a trick to get the VLB IDE card up to 1725 kB/s using part of its DOS driver's functionality. Which reads as 1888 kB/s in CoreTest. The trick is explained in my "486 VLB UMC-Chipset, what is it?" link above.

Tomorrow I'm putting this m/b in a better desktop case and I'll test bandwidth with CF and HD.

I have quite some items in my to-do-list. 😀
I guess I'll do more stuff having it assembled and ready to use.

Covid leave is over and I'll have less time again 🙁

Yesterday I tried version J2.

I can't see any improvement over X version. They are close.
I used 64MB on two sticks, 2-1-1-1, 0WS, T2 (lowest possible).
I ran Doom fast, they are equal in results. Unfortunately my VLB video card wouldn't work and I had to use an ISA Cirrus. Results were far less impressive.
Isa 3535, VLB 1700 (from old notes). Huge gap.

Speedsys is almost equal, no huge speed improvement 😀

Speedsys pics coming later.

Jan, thanks for your work. I think I'll need more time to find what really stands out with your BIOS.
I don't think I'll find a new VLB video card soon, they cost half a kidney (black friday thanks!) but still too much.

EDit: I used an IBM 5x86-100. Still need to go with dx4-100/120 + AM5-P75.

Did all tests of the Phil pack.
With the 5x86-120 (3x @40mhz) the J2 BIOS is less stable with results. I prefer the X version over this one.

I'm going to keep the latter, but honestly it's about personal taste, in games it'll do as good as X.

Statistically the difference isn't high enough, it's within that +/5% that varies among runs, in Doom 1705 vs 1690.
Quake @360 14 vs 11.1 is a lot, @640 6.4 vs 4.4 FPS. The only big difference.

All other test (VGA) are so close it's like 0.5/1%.

I'm not going to do AM5-P75, gonna be the same %. Unless you need some special testing I'm reboxing the mobo.

Personal note: I have no idea of what you do, how you do BIOS mods. I report back to give you an idea of progress.
Your work is greatly appreciated and I hope in a significant breakthrough in this M912 performance run 😀