I'm glad you appreciate the Inwin case. I've always had a thing for them. They remind me of the old Acer 486 desktop cases. Prior to containing retro computer parts, it was part of my main system until I moved to mATX (also a desktop). I go out of my way to get the desktop cases.

I also like full towers. Prior to this setup, the VL/EISA board was in a huge full tower AT case with all black drive panels (it just needed a few finishing touches). It looked really sharp, but it was huge and weighed a tonne.

I'm pretty sure on my VL/EISA board that the bus speed is determined by a DIP clock generator chip. I would be able to swap it if I could lay my hands on a pin compatible unit that did 60MHz. However, I have not been able to locate the datasheets for my clock generator, so I have no idea if there is anything that could work in it's place.

I believe the 14.xxMHz metal can crystal sets the speed of the OSC signal on the ISA bus. Screwing with it is a bad idea. I've tried this in the past, and and it just messes things up.

I am not exactly sure which of the ICs on my board is the North Bridge. There are 3 or 4 of them in the set. At 66MHz, I did none of them got warm enough to make me want to remove my hand.

I will check up on the 5x86 steppings for you later. It's really weird. I used to have an ADZ and an ADW. Now I appear to have two ADZ chips. My memory is lousy, but I think I must have acquired a 3rd 5x86 later on, and misplaced or traded my ADW.

On a Cyrix 5x86 chip, isn't LSSER always set to 0 by default unless your BIOS enables it?

Anonymous Coward wrote:

On a Cyrix 5x86 chip, isn't LSSER always set to 0 by default unless your BIOS enables it?

I'm not sure what the chip default is, but every motherboard I've tried this CPU in, the CPU/MB combined default is always such that LSSER = 1, which is not optimal. On an M919 motherboard, the BIOS actually has a user selectable option to "enable LSSER", which sets it equal to 0. This is the only 486 motherboard I've seen with selectable Cyrix enhancements in the BIOS. It also allows for you to enable Linear Burst modes.

Anonymous Coward wrote:

I'm pretty sure on my VL/EISA board that the bus speed is determined by a DIP clock generator chip.

I think that DIP is just the PLL multiplier chip, but I could be wrong. My understanding is that all motherboard frequencies get re-generated from the 14.31818 MHz can oscillator. So the PLL multiplies this 14.3 MHz by ~2.328 to yield a 33.33 MHz FSB. The PLL on my Biostar MB8433-UUD is a UM9515-01. I haven't been able to find the datasheet either.

So if that oscillator can hypothetically be increased to 19 MHz, then 19 x 2.328 x 3 = 133 MHz. You could run your Cyrix 5x86 at 133 MHz on a 3X setting. The FSB would be fast at 44 MHz, so you'd want to drop the PCI bus down with a 2/3 divider for 30 MHz operation.

Assuming the ISA bus uses a 5/9 multiplier from the crystal can for 8 MHz operation, then with a 19 MHz can, the ISA bus would be at 10.6 MHz. This should hopefully still work with many ISA cards.

There are a lot of assumptions here, but it seems doable.

What ATX->AT do you use? How does it attach to the power buttons? And where did you find this I/O Shield for the AT keyboard connector?

I just picked up one of those NiCE Super EISA boards. It has the generic 06/06/92 AMIBIOS. What are the specifics on cache support? Most references I can find show that at least L2 is write back on the SiS EISA chipset. Also any incompatibility with VLB cards? I found some old Usenet posts that indicated that some board revisions had compatibility problems. Is the generic SiS EISA CFG file sufficient for this board, or does it need a vendor specific CFG?

The correct EISA config files should be here:

http://www.mmnt.net/db/0/0/ftp.bluefeathertec … er%20EISA%20Cfg

The L2 cache is indeed write back, but it can be switched to write through in the BIOS if you need it. I am not 100% sure on the cacheable area of this board when in write back mode. I have two 64kx4 tag RAMs, one 64kx4 "dirty bit" for writeback operation, and 8 pieces of 128kx8 for a total of 1MB cache. Most boards only allow caching of up to 128MB in writeback mode with 1MB cache installed, but I suspect this one might do all 256MB...though I only have 128MB of 30pin SIMM RAM installed so I can't test it out (this stuff is expensive).

I think my board is Rev 1.2. That may have been the last revision (not sure though). I have only really experienced one VESA incompatibility so far, and I have tried pretty much all the major VLB cards except for really exotic ones. I noticed that when I have ET4000W32P cards installed, sometimes they don't get along with the soundcard or I/O card installed. It's some kind of weird address conflict. For example, I had to try 3 different models of YMF-71x based soundcards until I found one that would boot with the ET4000W32P. If there is a conflict, the system just doesn't turn on.

Overall I really like this board though.

Good to know. I grabbed the EISA config files. I'm guessing that L1 writeback is out of the question given that you had to force it off on the 5x86. Did any EISA 486 boards actually support it? I know the AIR 486EI was new enough to get 3.3v support, but it uses the same SiS chipset as this board. Heck, did any of these even support APM or LBA?

As much as I like my Micronics board, the cruddy Phoenix BIOS and the lack of hardware ROM shadowing with more than 16MB of RAM installed is just annoying. It'll be nice to have a purple screen AMI BIOS again! I think I can deal with have one VLB slot too (although a 2nd for a VL IDE controller would have been nice). I think the Micronics has it fair share of VL Bus wonkiness. It didn't like my Genova 8500VL Cirrus Logic CL-GD5428 card at all!

I hate the fact that the Super EISA board needs 30pin SIMMs, but being able to take 256MB across 16 slots is just insane for 1993! The Micronics used 72pin SIMMs, but only supported 64MB and required jumpers to set the memory size.

The sixteen 30-pin SIMM slots is indeed very cool...but it is a hell of a lot more practical to be able to use 72-pin SIMMs to get up to 256MB. One thing I don't like about the memory expansion on this board is that after the first 64MB you can only upgrade to 128MB, 192MB or 256MB. I'm not sure why I suspect that it might have to do with memory interleaving. My system is currently running with a 2-way (effectively 64-bit) interleave. I know there is a 386/486 chipset from Headland that supports 4-way (128-bit) interleave. I suspect the SiS chipset on this Nice board may too...but I can't find the datasheets for the chipset, and I don't have the ability to test it at the moment.
The Nice board does not do writeback L1 cache. The only EISA board that may is the AIR 486EI, but I have never had one to verify. That board only goes up to 512kb cache though. It would still be pretty neat if it supported writeback L1 though.
I don't know much about non-AMI BIOS prior to 1995. All of my 486 VLB boards have always been AMI. It works and I like it.

Why on earth do you want to run a VL IDE controller on a system with EISA slots? For heaven's sake, get yourself an EISA SCSI controller.

Anonymous Coward wrote:

The only EISA board that may is the AIR 486EI, but I have never had one to verify. That board only goes up to 512kb cache though. It would still be pretty neat if it supported writeback L1 though.

If that is a chipset level feature, I doubt it.

I don't know much about non-AMI BIOS prior to 1995. All of my 486 VLB boards have always been AMI. It works and I like it.

Yes, the AMI Hiflex BIOS reigned king in the early 486 era. Very few boards used Phoenix since it sucked (mostly OEMs). Award was pretty much non-existent too. When AMI switched to the WinBIOS in 1994, everyone switched to the Award 4.50 core.

Why on earth do you want to run a VL IDE controller on a system with EISA slots? For heaven's sake, get yourself an EISA SCSI controller.

I already have a AHA-2742W EISA SCSI card. IDE is mainly for CD-ROM and CF use. DTC did make an EISA IDE controller (DTC2290), but they seem to be extremely rare.

I am moving back into my own thread rather than stealing the one started by NJRoadfan. Here is a recap of some of my posts from here:
Another 486 build.... with EISA!

I now also have a Tyan S1437 motherboard in addition to my Nice SuperEISA. I still lean in favour of the SuperEISA because of the better build quality and flexibility when it comes to switching between an oscillator and clock gen (despite what I said in an earlier post, I found out later that this board actually can use metal can oscillators). I am attempting to run a Cyrix 5x86 at 133MHz on the SuperEISA using a non standard clock of 44.236MHz. Of the four 5x86 chips I have none of them are stable at 133MHz with my VRM which puts out just 3.45V. I will install a trim pot on my VRM to see if anything more can be done.
I am not confident I can reach 133MHz with a Cyrix though, so I ordered a late model amd 5x86-133 to see if I can get it to overclock to 180MHz using a 45MHz clock. The three am5x86 chips I have do not boot at 180MHz at 3.45V. However, they do work pretty well at 3x44.236=133MHz. If 180MHz is attainable, it should give much better memory performance at 45MHz than at 60MHz because of reduced wait states. Memory performance at 45MHz is actually pretty incredible, especially with the L2 cache set to write through. I'm not sure why, but on the SiS411 boards, write through L2 gives better performance than write back. This phenomenon was first observed by NJRoadfan and I confirm it is the same on my boards.
In order to test stability of the 45MHz bus, I have finally decided I should install Windows NT4. My current 2GB drive is a little small, so I ordered three 36GB SCSI drives to test out. Two Quantum/Maxtor Atlas IIIs and an Atlas IV. Hopefully the drives will not be all worn out.

Results will likely be posted here later next week.

If you are successful, it would certainly pave the way for more Cyrix 5x86-133 systems, although I don't think too many users have 486 boards with those 4-pin crystal oscillators. All PCI-based boards I've seen have used PLL chips. Perhaps all VLB systems too. Even some 386's used PLL chips for this.

A few years ago, I tried swapping out the 14 MHz crystals in boards which used a PLL for clock generation, but it resulted in floppy drive read problems. I seem to recall not being able to go more than approx. 1 MHz above the crystal freq.

There are a number of VLB motherboards that use oscillators. Normally, even if they have the clockgen, the pads for both the clockgen and crystal are present and it's obvious where to solder if you want to do a conversion.

I now have an AMD 5x86 with a 2000 date code. I attempted to run it at 180MHz (4x45) but I have not had any luck at 3.45V. It made it part way through the NT boot loader before crashing. Has anyone ever managed to get one of these chips stable at 180MHz at the default voltage, or do they all need 3.7 to 4.0 volts for stable operation?

I am a little disappointed so far, because this chip doesn't really seem a whole lot better than the three AMD chips I already had.

I am almost finished testing this chip. I tossed it into my R418 PCI board so that I could have a better selection of voltages to choose from. I tried 3.6V, 3.7V and 3.9V. It didn't seem to have much of an effect. I don't want to bother with anything over 4V.

It seems to be more or less equal to the '96 ADW I had in terms of overclockability. With the two ADZs I have the system won't even boot at 180MHz. The ADW and 16BGC will only work at 180MHz with the L1 cache turned off. I don't even know how any of you guys can even consider 200MHz.

It will boot at 180 MHz at 3.45 V with L1 disabled?

I think those running at 200 MHz were at 5 V. I am not sure about the few examples running at 180 MHz.

I tested again. I can boot up DOS at 180MHz with L1 disabled and run tests, but it fails to load NT4. I feel pushing an am5x86 to 180 is even more hopeless than getting a Cyrix 5x86 to 133.

I was not successful at getting my year 2000 model 5x86-133 stable at 180MHz. This may have been because I was not able to get my VRM to exceed 3.95V. I decided to just let it be rather than risk frying an interesting chip.
I was also not able to get any of my four Cyrix/IBM 5x86-100 chips stable at 133MHz with voltages as high as 3.85V. However, during the testing process I learned something useful...mainly that I was not subjecting my chips to enough stress to properly determine long term stability. For example, I found out that my Cx5x86-100 chips were not even perfectly stable at 120MHz the way I had them setup. If you run Windows NT4, these things quickly become apparent. What I discovered is that all of them need 3.6V to do the job, rather than the 3.45V that pretty much all VRMs provide. I also discovered a fan is required at 120MHz to prevent overheating. I found out that my memory timings were too aggressive. I had to to switch from "fastest" to "faster" in the BIOS (at 40MHz) to smooth things out. Strangely, this has a very noticeable impact on the CPU score in speedsys (just like the FPU problem when benching 386s)...in my case, from about 62 down to 54. Under Wintune 98 in Windows NT I do not believe there was much of a negative impact. Also, when I enable writeback L2 cache on my board, speedsys memory scores drop a lot but the CPU increases slightly. Under Wintune 98 memory and CPU both show significant increases in performance with writeback L2 enabled. Basically, I do not think that speedsys is a very reliable tool for benching older systems.

Other interesting stuff. I am currently running my Cyrix 5x86 in my SuperEISA board on a 60MHz FSB (clock doubled). It actually doesn't give much a speed boost (as far as I can tell) over 3x40, but in my cause it is very helpful since my system is not able to boot in 3X mode (BIOS incompatibility), and I am not able to change the clock multiplier from the Evergreen 5x86 utility in NT4. In order to get 3x40 working in DOS I have to boot into 2X mode, drop down to 1X and then back up to 3X. My motherboard does not have a jumper setting for 60MHz (but it does for 66MHz), but I was able to disable the clock generator and switch to a 60MHz metal can oscillator. To my surprise this board actually works quite well at 60MHz, better than the M-tech R418 PCI board which is quite newer. Perhaps the most surprising thing of all is that my ARK1000 VLB cards do not complain at all. This is insanely out of spec for VLB. I recommend ARK1000 cards to anyone who wants to run with a 50MHz bus. I have had three of these cards, and they all work at high speeds. One VLB wait state is required however

Booting into Windows NT4 has always been a minimum requirement for my systems to establish stability, particularly for overclocked 486 systems.

I am surprised you noted a significant increase in memory performance when L2-WB was enabled over WT. Perhaps you can compare WB/WT in Doom and Quake.

Will your system end up with the Am5x86-160/4x or the Cyrix 5x86-120/2x? Did your CMOS settings of the Am5x86 at 160 MHz pass the NT4 stability test?

The Am5x86 also failed NT4 with the aggressive memory timings I was using. I am going to stick with the Cyrix 5x86-120, because with enhancements on it seems faster (feels faster in NT4 at least). I have also gone back to my S1R3 chip and I have BTB enabled.

I will try doing some doom benchmarking. Quake is out of the question for now since I don't have a copy.

I have finally gotten around to installing Windows 2000 on this system. I have tried with both the am5x86 at 160 (4x40) as well as the Cyrix 5x86 at 120 (2x60). I have not yet decided which CPU I'm going to stick with. The Evergreen Cyrix control driver for NT4 does work in Win2k by the way.

The install procedure was interesting. First I could not get the system to boot using the floppy disks. I got an error with hal.dll. However, I was able to upgrade from Windows NT4. But, in order to get my AHA 2742W SCSI controller recognised I had to copy arrow.sys and a .inf/.oem file to a floppy disc to load the drivers as Win2k no longer supports this card.

The install took quite a long time. I wasn't timing it, but I estimate about 2 hours. My 3com 3C597TX ethernet card and OPL3SAx sound card were automatically detected and work out of the box. However, my ARK1000VL graphics adapter was not supported. I had to use the NT4 driver instead.

I have not yet upgraded to SP4, but I plan to soon as well as Windows 2000 Lite utility. I need to figure out how to strip this OS down to make it nice and lean like NT4. I'm probably not going to run any games on this install (can't install a 3D card anyway), but it seems to be fast enough to do the basics.

I have not gotten around to updating Windows 2000 yet, but I have the system dual booting with DOS and wfwg311. I have decided to stick with the am5x86 at 160, as it is a lot easier to get running...there are no software utilities to deal with. The Cyrix 5x86 at 120 isn't very fast in my system as my board is too old to support the advanced features that give it a real speed boost. In DOS games, the am5x86 is faster for anything that doesn't rely on an FPU.

Recently I started benchmarking the VLB cards I own in this system. I was also able to get my Hercules Dynamite Power working again. It had a bad solder joint on one of the resistor packs. I believe my Hercules Terminator 64 DRAM is also suffering the same fate to a lesser extent. If you own a late model Hercules VLB card that has failed, chances are they also have bad solder joints.

Here are the results of my testing so far. This is with an am5x86-160 in WT mode, SiS411/406 chipset with WB L2 cache, 64MB DRAM, 1024kb cache. Tests were done with a YMF-719 soundcard installed and enabled. Windows tests were done at 1024x768@64k in non interlaced mode.

Number 9 Motion FX 771 (S3 Vision968)

DOS
Doom = 1428 realtics
pcpbench = 11.2fps
pcpbench vga = 25.8fps
3dbench2 = 88.9
mvspeed = 314.9fps
Quake = 16.6fps
Landmark V2.00- 16948
vspeed = 18.43 million bytes/sec

WFWG311
Wintune 2.0 = 12.915kpixels/sec
Windows Speed = 999
Speedy = 55.79
Graphics Winmark = 13.0

ATi Graphics Ultra Pro Turbo VLB(Mach64 VRAM)

DOS
Doom = 1643 realties
pcpbench = 11.4fps
pcpbench vga = 26.0fps
3dbench2 = 80.7
mvspeed = 341.8fps
Quake = 16.6fps
Landmark V2.00= 11045
vspeed = 15.48 million bytes/sec

WFWG311
Wintune 2.0 = 11.077kpixels/sec
Windows Speed = 621
Speedy = 38.23
Graphics Winmark = 8.59

ATi Graphics Xpression VLB (Mach64 DRAM)

DOS
Doom = 1644 realties
pcpbench = 11.4fps
pcpbench vga = 26.0fps
3dbench2 = 80.9
mvspeed = 342.8fps
Quake = 16.5fps
Landmark v2.00 = 11045
vspeed = 15.48 million bytes/sec

WFWG311
Wintune 2.0 = 8303kpexls/sec
windows speed = 579
speedy = 28.55
Graphics Winmark = 7.45

ATi Graphics Ultra Pro Turbo ISA (Mach64 VRAM) 8MHz

DOS
DOOM = 4326 realtics
pcpbench = 6.5fps
pcpbench vga = 18.2fps
3dbench2 = 36.0
mvspeed = 52.8fps
QUAKE = 13.2fps
vspeed = 3.32 million bytes/sec
landmark v2.00 = 2540

WFWG311
Wintune 2.0 = 6285kpixels/sec
Windows Speed = 196
Speedy = 11.46
Graphics Winmark = 3.84

Hercules Terminator 64 DRAM VLB(S3 Trio64)

DOS
DOOM - 1320 realtics
pcpbench - 11.8fps
pcpbench vga - 26.7fps
3dbench2 - 90.3
mvspeed - 476.2fpS
QUAKE - 16.9fps
vspeed - 20.64 million bytes/sec
landmark v2.00 - 20062

WFWG311
Wintune 2.0 - 10049pixels/sec
Windows Speed - 996
Speedy - 30.64
Graphics Winmark - 10.9

OCTEK VL VGA-1000 (Ark logic ARK1000VL) could not run 1024x768x64k in non-interaced mode

DOS
DOOM - 1306 realtics
pcpbench - 11.9fps
pcpbench vga - 26.6fps
3dbench2 - 90.3
mvspeed - 476.2 fps
QUAKE - 16.6fps
vspeed - 19.17 million bytes/sec
landmark v2.00 - 19660

WFWG311
Wintune 2.0 - NA
Windows Speed - NA
Speedy - NA
Graphics Winmark - NA

Hercules Dynamite Power VLB (ET4000W32P)

DOS
DOOM - 1363 realtics
pcpbench - 11.9fps
pcpbench vga - 26.6fps
3dbench2 - 90.3
QUAKE - 16.6fps
mvspeed - 476.2fps
vspeed - 19.17 million bytes/sec
Landmark v2.00 - 19660

WFWG311
Wintune 2.0 - NA (Test could not finish)
Windows Speed - 957
Speedy - 62.31
Graphics Winmark - 7.99