Found this info today on the SP3G:

Q04) What is the better cache sheme: write-through or write-back ?
How can I upgrade my SP3g with a dirty Tag RAM ? Does it increase
the performance ?

The ASUS SP3G can drive write-through or write-back cache. You can use both
modes with and without dirty tag RAM present. In write-through mode, dirty
tag RAM has no function. The BIOS default is "write-back", resulting in
some 5% performance loss, because if there is no dirty tag RAM, then there
are lots of unnecessary memory writes, since each cache line refill will
need to be preceeded by a write back of four DWORDs, even if none of them
had been modified.
In write-back mode, dirty tag RAM increases real-world performance by
approx. 1% over write-through. The small positive effect of write-back on
PC compatible hardware is caused by the lack of write-allocate. In case
of the Saturn chipset, the number of clocks for some transactions is
higher for write-back and the Saturn has already very efficient write
posting buffers which compensate much of the inefficiency of not having a
dirty tag RAM.
Starting with rev. 1.8, the SP3G got a socket for a dirty tag RAM. With
this RAM installed. Suitable chips for 256kB cache size are 64K x 1 RAMs
with 15 ns access time, e.g. Cypress Cy7C187, AT&T ATT7C187-15,
Matra MHS HME 65787-15, Motorola MCM6287C-15 and Toshiba TC5562-15 or any
other equivalent.

-----------

Q05) What type of simms can the PCI/I-SP3g use ?

The SP3g can only use pairs of simm modules, because the SATURM II chipset
alsways enables memory interleave, and to do so you need at least to
modules. So if you plan for example 16MB for your system count the price
of 2 * 8MB.
The board can use simms with or without parity as it doesn't use the parity
bit on the simms.
With 4MB and 16MB modules you can either use pairs of single sides or
double sided ( = density, I like that one Stefan). Even though people
reported that the successfully mixed a single sided and double sided 16MB
module on one memorybank.
With 8MB and 32MB modules you can only use double sided (= density) simms.

It is a quirky board/chipset, and I guess that is why we know so little about it. I like that!

vetz wrote:

It is a quirky board/chipset, and I guess that is why we know so little about it. I like that!

For sure.. I like a challenge!

Looks like my freshly-burned chips should be sitting at the post office waiting for me! Stay tuned....

YES! 😎

After waiting patiently the chips have arrived...
Freshly burned chip works flawlessly thanks to a very helpful VOGONS member. Write-Back mode!!!

Time for some benchmarks, and then I'll put this 2nd build together in a nice New Old Stock AT case that I have sitting here...

Running :
Asus VL/I-486SV2GX4 with new EPROM updated 0402.001 Beta Bios (from 1999)
256KB Cache
Intel 486DX4-100 (&EW)
16MB RAM
4GB CF Card on Promise Controller

Did you give up on the PCI/I-SP3G?

Nope! I just wanted to make sure these EPROM chips worked in both Asus VL/I-486SV2GX4 boards. Now, for the second board (the rev 2.1) I need to make a decision on the processor to use. I'm thinking I may throw my Cyrix 5x86 120 in there instead of the A80486DX4-120SV8B, and because it can't run the 120Mhz AMD processor in L1 Write-back mode anyway. Thoughts?

So when all systems are considered, I'd have:

Build 1: 1 x BIOSTAR MB-8433UUD-A Rev. 3.1 with AMD-X5-133ADZ (Am5x86-P75) in Write-Back Mode [Build Completed]
Build 2: 1 x Asus VL-i486SV2GX4 Rev. 2.0 with Intel i486DX4-100 &EW in Write-Back Mode [Configured today]
Build 3: 1 x Asus VL-i486SV2GX4 Rev. 2.1 with Cyrix 5x86 120 in Write-Back Mode
Build 4: 1 x Asus PCI I-486SP3G Rev. 1.8 with AMD A80486DX4-100SV8B in Write-Back Mode

Now I wish I had an X5-150, 160 or Cyrix 5x86 133 to test with. This is starting to get fun!

😎

You want to compare the Biostar MB-8433UUD vs. the Asus PCI/I-SP3G using identical hardware and indeally with a 33 MHz FSB.

If you are requesting recommendations on a final build for the PCI/I-SP3G given your existing builds, you might try running an X5-133ADZ at 160 MHz. Are there any accounts of a 133ADZ not working reliably at 160 MHz? I'd suspect that most issues associated with 160 MHz would be with the FSB on the motherboard side and too tight of timings.

feipoa wrote:

You want to compare the Biostar MB-8433UUD vs. the Asus PCI/I-SP3G using identical hardware and indeally with a 33 MHz FSB.

For benchmarking purposes, what do you recommend I use for processors in each ?

If you can get the PCI/I-SP3G working with L1 write-back mode, then either an AMD-X5-133ADZ or Intel DX4-WB-100

feipoa wrote:

If you can get the PCI/I-SP3G working with L1 write-back mode, then either an AMD-X5-133ADZ or Intel DX4-WB-100

Sounds good. I'll pop in the AMD Am5x86-P75 ADZ and see if it will work with WB mode. According to that chart in the link you posted, it should work.

I know there is a 2GB DOS partition limit size in DOS 6.22, and of course there is the old 504MB drive limit fixed with firmware upgrades... But can you think of a reason why my 4GB CF cards are being autodetected at 1836 MB in the BIOS on both of these GX4 boards? I'm using a Promise VLB controller to host the CF card.

Artex wrote:

I know there is a 2GB DOS partition limit size in DOS 6.22, and of course there is the old 504MB drive limit fixed with firmware upgrades... But can you think of a reason why my 4GB CF cards are being autodetected at 1836 MB in the BIOS on both of these GX4 boards? I'm using a Promise VLB controller to host the CF card.

There is a 2GB limit in the BIOS AFAIK (you should be able to manually input that for your CF card). If you want bigger disc you need to go SCSI or Drive overlay software.

vetz wrote:

Artex wrote:

I know there is a 2GB DOS partition limit size in DOS 6.22, and of course there is the old 504MB drive limit fixed with firmware upgrades... But can you think of a reason why my 4GB CF cards are being autodetected at 1836 MB in the BIOS on both of these GX4 boards? I'm using a Promise VLB controller to host the CF card.

There is a 2GB limit in the BIOS AFAIK (you should be able to manually input that for your CF card). If you want bigger disc you need to go SCSI or Drive overlay software.

Ahh... I thought that perhaps that limit might be removed with the BIOS update. Any overlay software you can recommend for DOS? I guess it's not a huge deal as I could squeeze several DOS games on a 2GB partition, but I would like to get Windows 95 installed on one of these boxes.

Windows 95 is okay with 2 GB. I have windows 95 on a 512MB cf card and its tight but alright
98+ you would need more

I'm still in the mindset that the BIOS should be able to see the full size the of the card with this latest (and latest available) BIOS update. Pretty sure one of the earlier BIOS revisions enabled LBA support, so I'm going to fiddle around with this tonight.

Alright folks.. another update. Looking back on my previous entry, here were the results for the MB-8433UUD-A motherboard running an AMD X5-133ADZ:

System 1 Specs:
BIOSTAR MB-8433UUD-A
AMD AM5x86-P75 / AMD X-5-133ADZ
32MB EDO RAM
256KB L2 Cache
STB S3 Trio64V+ (2MB)
4GB CF Card (CF->IDE Adapter)

System 1 Benchmark Results:
SpeedSys Overall Score: 49.97
DOOM: 2134 gametics in 1776 realtics
Superscape: 704 fps
PC Player Bench: 17.7
Quake: 969 Frames 84.4 seconds 11.5 FPS

Today I threw the same processor into my Asus PCI I-486SP3G (Intel S82423TX) Rev. 1.8, Saturn-based motherboard that uses memory interleaving (similar to dual channel memory by today's standards). I'm sure you'll agree that these are some interesting findings. 😎

System 3 Specs:
Asus PCI I-486SP3G
AMD AM5x86-P75 / AMD X-5-133ADZ
32MB RAM
256KB L2 Cache
ATI 3D Rage Pro II (2MB)
4GB CF Card (CF->IDE Adapter)

System 3 Benchmark Results:
SpeedSys Overall Score: 49.88
DOOM: 2134 gametics in 1612 realtics
Superscape: 768 fps
PC Player Bench: 17.8
Quake: 969 Frames 82.3 seconds 11.8 FPS

Still cannot get the MB to 'see' the full size of the 4GB CF card. It's showing up as 1806MB even with the latest BIOS. Also interesting is that the processor (at least at POST), shows up as a 80486DX2

When running the CPU identification utility, everything's reported properly. I was able to find an older Google Groups post that provided jumper settings to get this processor working, and in WB mode - nothing was in the original manual.

Coin-modded RTC!

Just some hardware porn here...

The benchies...

Good job documenting it all Artex!

Could you run Cachechk v7 on both systems and share the results?

Read speed:
cachechk -d -t6

Write speed:
cachechk -d -w -t6

Could you also upload the screenshot of the full Speedsys report?

What cache and memory settings are set in the BIOS for both motherboards?

Will do!! I have some more goodies to share today as well... Dropped in my Cyrix 5x86 120 in my rev 2.1 Asus VL/I-486SV2GX4 board. 😎