First success!

1) AMD 5x86-133 initialized with default BIOS 4.27 setting

2) AMD 5x86-133 initialized as 486DX2-66 subroutine in BIOS 4.27

BIOS 4.27GN2 patched for the max performance of AMD 5x86-133 and nice CPU POST string.

Hi, all.

Please check the latest bios modifications for vip-io.

Tnx!

Hi, Babasha.

Please check the chipset utility for vip-io/io2.

Hi, Babasha.

Results of using the tuning utility are below.

Pci perforformance before and after

Yftul wrote on 2025-05-08, 17:43:

Pci perforformance before and after

So that indicares that performance got better?

Correct

Yftul wrote on 2025-05-10, 16:45:

Correct

I’ve got 486 VIP without the IO that I bought new a long long time ago. https://theretroweb.com/motherboards/s/fic-486-vip

I’ve spent years looking for a BIOS that will have good PCI performance and properly recognize an AMD133. Can’t wait to try out the bios. But I’ll have to wait until wednesday at least because I got a lot of stuff going on right now.

Sure. Please try it. Possibly you will need to use your own values for RX22 and RX51 registers, corresponds to the possibilities of your hardware.

PS. I used this datasheets for my experiments
https://dosdays.co.uk/media/via/VIA_82C496G.pdf
https://theretroweb.com/chip/documentation/vt … f6889725501.pdf
actually vip-io contains chipset 82486+82505, but seems it at least partially compatible with 496.

Yftul wrote on 2025-05-12, 11:21:

Sure. Please try it. Possibly you will need to use your own values for RX22 and RX51 registers, corresponds to the possibilities of your hardware.

The BIOS identifies my CPU as "Am486DX4B-S CPU at 150MHz". Mostly. Every so often I get "Am486DX4B-S CPU at 120MHz". It's an AMD X5-133 ADZ

Speedsys sees it as 133Mhz with a CPU score of 44. VESA video write speed in 16MB/s

Edit - Doom Frame rates are similar when using comparable PCI and VLB cards. About 74 FPS. So the PCI is performing well. The problem is that when I use a PCI video card, the ATA storage performance becomes very poor and occasionally fails completely.

You have a multiplier coefficient of 3 set, so the 5x86 processor support modification code is not activated.
As for the PCI performance speed—what kind of card is it, and how was the speed measured? DrHard, for relatively modern graphics cards (such as the Tseng et4000/w32p, S3 Trio64V+, S3 Virge), reports speeds of around 30 MB/s and higher. However, since your bus runs at 50 MHz, the PCI frequency is divided by 2. This could potentially be a factor.
Regarding ATA, these drives are highly sensitive to overclocking. Try replacing the drive with a CF card and see if you experience any speed drops with it.

I just tested your latest BIOS version and Am5x86 CPUS are now identified and configured correctly. So far this was only possible with the alternative BIOS version offered by FIC that sacrifices PCI support and L2 caches more than 256K.
But there´s still an issue that would be worth fixing, if possible: The "TAG - splitting" option has no effect, there´s no Dirty-TAG available under the 7+1 setting. This is also the reason for the very poor memory performance in L2 WB mode.
The second annoyance is that BIOS reduces memory timings massively as soon as you increase frequency or / and multiplier. Other BIOSes for this chipset do the same, so maybe it´s due to serious hardware limitations of the VT82C486A...

At the moment the 33 MHz Speedys results are: 24.61 / 39.99 / 109.41 and a memory bandwidth of 54.80 MB/s (far from Yftul´s impressive 204.54 MB/s)

Assuming the registers for enabling "Dirty TAG" 7+1 operation are the same for the 486A and it´s successor 496, I compared the default settings of BIOS 4.27GN2A and the alternative 3.276GN1 and I found there is no difference.
Still the latter version enables 7+1 for AmX586 and 4.27 does enable it for an Intel 486-DX4, but not for the Am5X86.

It seems this is not an issue of the chipset registers default settings (like in the case of the SiS 471), but some detail in the BIOS code when an Am5x86 is being detected and default registers are overwritten. Setting 7+1 in BIOS setup won´t overrule this.

Extremely high memory bandwidth results in Speedsys usually show that WB and Dirty L2 TAG are active, so I wonder if Yftul´s port_util script does the trick here?

Just some updates here:
I tinkered a bit with the "port.exe" tool to optimize memory and cache performance. (I didn´t care about PCI-performance so far.)
To keep it simple I run
1. "port.exe -c0x50=0x88"
2. "port.exe -c0x5e=0x80"
to enable WB L1 and WB L2 with one Tag-bit for dirty TAG.
and
3. "port.exe -c0x22=0x40" for the tightest RAS/CAS settings possible.
When I add
4. "port.exe -c0x51=0xc" (for 256K L2 cache, double bank, 2-1-1-1/2-1-1-1)
for fastest L2 cache timings there´s no improvement, so I think BIOS sets "2-1-1-1 / 2-1-1-1" by itself for 33MHz when "Turbo" is set in BIOS.

CTCM, Speedsys and Cachecheck results:

This seems to be the limit that can be reached with this chipset. The results for this VIA "VT82C486A" mainboard are identical with those on mainboards with the later "VT82C496G".
I get the same results with my Jetway "J-433-A"
Faster FSBs and CPU speeds improve L1 performance but heavily decrease L2 and memory throughput.

Oh, and a big Thank You to Yftul for providing the tool for setting the chipset registers!

majestyk wrote on 2026-02-24, 16:31:

so I think BIOS sets "2-1-1-1 / 2-1-1-1" by itself for 33MHz when "Turbo" is set in BIOS.

You can try checking the values using command "port -dump". It will print the entire port map of the chipset.

Here´s the one of my Jetway J-433-A:

RX51: Line 5, column 01 -> ox0c (so 2-1-1-1 / 2-1-1-1 is set for L2)
RX50: Line 5, column 00 ->0x8a (8a also implies that "burst write" is enabled by bit 1. On the FIC 486VIP-IO with it´s 486A chipset this isn´t supported, so we have to set 88 here)
RX5E: Line 5, column 0E ->0x88 (or 0x80)
RX22: Line 2, column 02 -> 0x40 (RAS / CAS)