Sounds pretty sweet! Got any pictures?

CapnCrunch53 wrote:

Sounds pretty sweet! Got any pictures?

I knew I was forgetting something!

Here is the Initial benchmark that I ran... I realized after that I forgot to bench the L cache... that's ok, I didn't get around to turning it up past 3-2-3 and 3, 3 as well as the PCI divider is at 1:1/2 (I'm not gonna push the divider). The CPU seems very stable so far at 3.6v.

Uploaded with ImageShack.us

This is the rear view of the case... I left the brackets out for now so the air can flow slightly better through this nearly unvented case.

Uploaded with ImageShack.us

This is the front vew of the beautiful TEAC CD Rom and TEAC combo floppy drive (yes both drives work). I'm missing a few HDD bay covers in the left of the picture.

Uploaded with ImageShack.us

This is a view of the family that runs the show. Counterclockwise from the left: GUS (he's a big boy, might even make it pro football),
AWE 64 Gold (she's just joined the school choir),
Voodoo1 3dfx with his wife S3 trio Diamond Multimedia (yeah, she's the one with the diamond),
2GB CF HDD (I guess that's the photo album),
2x 32MB RAM (odd family pet, but I guess rams make you learn responsibility at a young age),
256KB cache (Bank accounts are pretty small nowadays),
IBM 100HF CPU OC'd to 133 (this would be the landlord and he's probably peeved about the pet),
and finally the property that they live on... Shuttle HOT-433 Rev. 4.

Uploaded with ImageShack.us

EDIT: I forgot to mention that the reason you can't see the CPU is because I'm too lazy to remove the 12v heatsink/fan that I added after the first one fell off.

So your stock IBM blue heatsink fell off. Why not just thermal epoxy it back on and put on a fan?

How's the progress on the chipset removal and resolder?

Seems strange that you were able to run this same IBM 5x86c CPU on a Shuttle HOT-433 rev. 1-3 at 150 MHz, but not on rev. 4. Try removing one stick of RAM. I find the 66 MHz x 2 operation works best with a single stick of FPM (gold contacts).

Excellent, I would love to run my 5x86 at 133mhz if my board supported that setting.

Is that speedsys with the cyrix enhancements?

On a side note, have you run many games with the 3dfx card? Ive tried a few, GLquake runs a little bit slow with my 5x86-120, Outlaws almost runs good, their must be some games that will run well on a cyrix 5x86

So your stock IBM blue heatsink fell off. Why not just thermal epoxy it back on and put on a fan? How's the progress on the ch […]
Show full quote

So your stock IBM blue heatsink fell off. Why not just thermal epoxy it back on and put on a fan?

How's the progress on the chipset removal and resolder?

Seems strange that you were able to run this same IBM 5x86c CPU on a Shuttle HOT-433 rev. 1-3 at 150 MHz, but not on rev. 4. Try removing one stick of RAM. I find the 66 MHz x 2 operation works best with a single stick of FPM (gold contacts).

I think there was a misunderstanding, I've never been able to use a rev 1-3, only rev 4. It was on my first Shuttle HOT-433 Rev. 4 that I was able to do such numbers... these two new ones that I got are not as capable as that one was at 150MHz, despite being the same board model... I'm going to check for differences in chips soon.

I don't really care for epoxy, and because the heatsink/fan combo seems to work better than the old blue heatsink, I've just packed the old one away and kept it this way.

I did run into an issue with this setup, and I suspected the RAM at first (2 sticks of FPM vs 1 stick), but I always try to remember the basics first, and discovered that my PS/2 keyboard that I was using with an adapter shorted out and now I'm stuck using a 25 year old Zenith Data Systems AT keyboard. I guess this is a lesson in respecting AT design.

As far as the progress on resoldering goes... well... I'm still not sure how I'm going to pull it off, but I have most of the materials for the easier parts and just starting to trust myself with it 🤣.

Excellent, I would love to run my 5x86 at 133mhz if my board supported that setting. Is that speedsys with the cyrix enhanceme […]
Show full quote

Excellent, I would love to run my 5x86 at 133mhz if my board supported that setting.

Is that speedsys with the cyrix enhancements?

On a side note, have you run many games with the 3dfx card? Ive tried a few, GLquake runs a little bit slow with my 5x86-120, Outlaws almost runs good, their must be some games that will run well on a cyrix 5x86

I haven't really benched games yet, as I am still messing around with other settings (i.e. sound). I do have a small list of games to bench later, though.

I'm running:
Lsser=off
rstk=on
loop=on
fp_fast=on
mem_byp=on
dte=on.

EDIT I am running into an issue with the Voodoo 1 card. The screen goes blank when I run Tomb Raider 1. I have to reboot the computer and swap the video plug back to the main card for a second for the display to start working... I'm wondering if it's because I'm using an LCD monitor and the resolution is too low.

Can you try that HOT-433 rev4 board with 512 KB of cache? 4 of those 1024 KB, 10 ns modules should work if the board has jumper settings for 512 KB cache.

I also haven't tried the new UMC chipsets yet. I have become very busy all of a sudden. 686 Benchmark Comparison is also on hold.

If you turn on branch prediction (BTB), you're speedsys score will jump to 75.4 (but you also need to turn off RSTK and LOOP).

It is odd how this motherboard won't run at 3x50 MHz with the Cyrix. Test 150 MHz with an AMD X5 to rule out memory, cache, chipset issues. I have found these UMC boards to be very RAM particular with different FSB's and BIOS timings.

Nice 2x 66MHz build, but the PCI-Performance is pretty low for an UMC based system. So the graphics performance may be the bottleneck in your system. You should have a PCI memory performance from 18 - 22 MB/s with a common S3 card but not 11 MB/s.

FGB wrote:

Nice 2x 66MHz build, but the PCI-Performance is pretty low for an UMC based system. So the graphics performance may be the bottleneck in your system. You should have a PCI memory performance from 18 - 22 MB/s with a common S3 card but not 11 MB/s.

I think I can explain the low performance of the PCI video card. I purposely hobbled the PCI performance (as well as the cache timings, etc...) to ensure a boot could be made and to test it over time at the lower speed settings.

I've pulled the Voodoo 1 video card for now until I can sort it out and while doing that, I observed the codes on the northbridge and southbridges.
UM8886BF
9645-FXT
KMBQ05
---------
UM8881F
9645-EYA
MA0505

My first HOT-433 had:

UM8886BF
9625-FXA
KM4M62
---------
UM8881F
9631-EYA
R62703

It would appear that the major difference between the boards is their manufacturing date (if I'm reading this correctly).

Can you try that HOT-433 rev4 board with 512 KB of cache? 4 of those 1024 KB, 10 ns modules should work if the board has jumper settings for 512 KB cache.

It does have the jumper settings for 512KB... I'll have to work on my soldering skills faster, it would seem.

If you turn on branch prediction (BTB), you're speedsys score will jump to 75.4 (but you also need to turn off RSTK and LOOP).

Very close... but not correct. I posted a jump to 75.52, not 75.4. Also my 3dbench score jumped from ~75 all the way up to 83.3 at the same MHz.

It is odd how this motherboard won't run at 3x50 MHz with the Cyrix. Test 150 MHz with an AMD X5 to rule out memory, cache, chipset issues. I have found these UMC boards to be very RAM particular with different FSB's and BIOS timings.

ADZ-133 OC'd to 150MHz very easily at 3.6v. I'm just trying to imagine how high my speedsys score would register with the 5x86C @ 150 with BTB on. (without setting it up properly, I was in the low 80s range before).

With which BIOS settings did you hobble the PCI performance? My IBM 5x86c-133 (2x66) gets 47.2 MB/s for the Speedsys video test using a Matrox Millennium G200 and a Biostar MB-8433UUD v2.0.

I think the important identifiers in the chipsets are the bolded entries, as follows,

UM8886BF
9645-FXT
---------
UM8881F
9645-EYA

I beleive the bolded items affect chipset functionality. I'm not yet sure what the last identifier is for, but perhaps it also exhibits some minimal change in functionality? When searching for these chipsets, most suppliers will list the product, for example, as UM8886BF - FX or UM8881F - EY. They placed the importance on those first two characters only.

So all of your HOT-433 v4 boards have soldered cache? Did you get the proper DIP sockets to put in?

Try 3DBench with a Matrox Millennium G200. My score was greater than 99.9 and I had to use 3Dbench2.

EDIT: By the way, that is a very interesting 3.5" and 5.25" combo drive you have. Where did you get it and are there more? Do both drives work?

feipoa wrote:

With which BIOS settings did you hobble the PCI performance? My IBM 5x86c-133 (2x66) gets 47.2 MB/s for the Speedsys video test […]
Show full quote

With which BIOS settings did you hobble the PCI performance? My IBM 5x86c-133 (2x66) gets 47.2 MB/s for the Speedsys video test using a Matrox Millennium G200 and a Biostar MB-8433UUD v2.0.

I think the important identifiers in the chipsets are the bolded entries, as follows,

UM8886BF
9645-FXT
---------
UM8881F
9645-EYA

I beleive the bolded items affect chipset functionality. I'm not yet sure what the last identifier is for, but perhaps it also exhibits some minimal change in functionality? When searching for these chipsets, most suppliers will list the product, for example, as UM8886BF - FX or UM8881F - EY. They placed the importance on those first two characters only.

So all of your HOT-433 v4 boards have soldered cache? Did you get the proper DIP sockets to put in?

Try 3DBench with a Matrox Millennium G200. My score was greater than 99.9 and I had to use 3Dbench2.

EDIT: By the way, that is a very interesting 3.5" and 5.25" combo drive you have. Where did you get it and are there more? Do both drives work?

I just set all the PCI bus settings to slowest... but now I'm wondering if that's really the issue. I tested several video cards that I have (sorry I don't have the G200) and I got very similar results with all of them. I'm really starting to wonder if Shuttle just plain has poor PCI performance. Oh, well. I'm keeping my eye's peeled for a G200.

I've noticed the same thing on the chipset markings... but I like to include all numbers incase there's something more to it.

Yes, all Rev. 4 boards seem to have soldered sockets. I have DIP sockets that I'm hopping will work... they're 16pin, but I was told that if I place them together, it'll work just the same, because the pins don't touch each other. I hope that's the case.

Btw I did have a card combo that yielded a 3dbench(2) score over 100 once, but I can't remember what the combo was... might have been my Soyo 4saw2 with my other Trio card.

I loved the combo drives from way back when we had one given to us... I got that one a number of years ago from eBay to replace the one that was given away... yes it's fully functional and can copy from A: to B: and vice versa. The first drive I used was an Epson and that one actually did start to have issues after a while, but the guy we gave it to claimed he knew how to fix it. The other brands that I know of are Cannon and TEAC (what I'm using now). Most of the drives I see now are on eBay and overpriced by a long shot... also they usually are USA shipping only. 🙁

Did the combo floppy drives come in white/beige rather than brown?

What happens if you set the PCI performance to the best BIOS settings? You should be able to leave the PCI performance on its best setting since you are using a 1/2 FSB-to-PCI divider in the BIOS.

feipoa wrote:

Did the combo floppy drives come in white/beige rather than brown?

What happens if you set the PCI performance to the best BIOS settings? You should be able to leave the PCI performance on its best setting since you are using a 1/2 FSB-to-PCI divider in the BIOS.

...Brown? I missed something, bro. I've only seen them in gray (the one I have), beige, white, and once a black one.

With it running at it's best, the PCI performance doesn't seem to change a bit, much to my surprise. Also the Speedsys score is fluxuating now. It used to be a solid 75.52 and now it's anywhere from almost 76 down to just over 75 (I like to run it at least once every day).

In beige and white? I'd like to get me one of those. The grey/brown colour that you have would only match my 386 tower.

My IBM 5x86c-133 scores in Speedsys never fluctuate. I'm not sure what is going on with your PCI performance. Trying different graphics cards, different PCI slots (try not to use the shared PCI/ISA location), and seeing how 3Dbench, PcpBench, Quake1, Doom, and Landmark do compared to my IBM 5x86c-133. Perhaps just Speedsys is having issues with your HOT.

For 150 MHz operation (3x50), did you try the cache at the slowest setting (3-2-2-2) and the RAM at 2/2? For a few select RAM modules I have, I can run 50 MHz FSB at 0/0, but other modules sometimes require 1/0, 1/1, or 2/1. For 66 MHz FSB, my good RAM modules work at 1/0.

For cache at 50 MHz, you can sometimes get away with 2-2-2-2, but 3-2-2-2 is a good initial setting.

Something else you might want to try is changing the SIMM slot that the RAM is in. Sometimes the more frequently used SIMM slots don't make as good of contact, especially at higher FSB speeds.

I'll keep on the lookout for a cheap beige or white combo drive wihle I'm on the hunt for a decent functioning 8433UUD. I think I understand what you ment by brown now. (Adding to my shopping list of Matrox PCI, better RAM, and perhaps a CPU capable of 150+ MHz under the Cyrix design).

I realized that I had accidently moved my RAM timings to 3/3 so I moved them to 2/2 and locked the speedsys score to 75.51 now. The other settings I left at the slower rate 3-2-3 and PCI wait at 1. It's kinda funny, if I go too slow it fluxuates more than if I go too fast.
When I ran the 50MHz FSB, I used the slowest timings as a rule, mostly because my RAM isn't extremely efficient.

Swapping the SIMM slot didn't make a difference at all with the performance.

I'm still concerned about Speedsys on the HOT-433 because it's posting the speeds at 125 and 126 MHz when it's at 133... it used to post correct speeds before I blew the old motherboard.

Changing the PCI slots made no difference, but I did discover that my ATI Rage 128 didn't work on one HOT-433 though it worked in the past on my other one.

Thanks for all the help and support so far guys. There are so many great minds here.

Your experience reinforces my beleif that the HOT-433 was a buggy motherboard. In my experience, the Biostar MB-8433UUD is the best overall UMC-chipsetted, PCI-supporting, 486 motherboard. The only drawback being the 512 KB L2 cache limit and the inability to set the L2 cache into Write-through mode to increase the cacheable RAM area.

I may consider trading an MB-8433UUD for a fully functional PCI-SiS board of my interest. I would really like to compare one of these later SiS boards with the MB-8433UUD.

[quote="feipoa"
I may consider trading an MB-8433UUD for a fully functional PCI-SiS board of my interest. I would really like to compare one of these later SiS boards with the MB-8433UUD.[/quote]

I can run some tests if your interested, in speedsys my pci bandwidth with the the Voodoo3 is only around 16,480kb/s, im hoping to get an ALI PCI chipset board working that will be an interesting comparision.

feipoa wrote:

Your experience reinforces my beleif that the HOT-433 was a buggy motherboard. In my experience, the Biostar MB-8433UUD is the best overall UMC-chipsetted, PCI-supporting, 486 motherboard. The only drawback being the 512 KB L2 cache limit and the inability to set the L2 cache into Write-through mode to increase the cacheable RAM area.

I may consider trading an MB-8433UUD for a fully functional PCI-SiS board of my interest. I would really like to compare one of these later SiS boards with the MB-8433UUD.

What's special about the SiS board apart from the fact that you don't have one? I don't believe there are any hidden features to be unlocked on those like on the Biostar boards.

sliderider wrote:

What's special about the SiS board apart from the fact that you don't have one?

There have been successful reports of some SiS PCI boards with a working Voodoo3 graphics card. Some SiS boards also seem to work with 60/66 MHz front-side buses. The screens of UMC-based PCI boards won't turn on with a Voodoo3, Banshee, or TNT installed. The issue with using a UMC board and a Voodoo1 or 2 is that a second graphics card is needed for 2D graphics. On 3-PCI motherboards, that leaves only one extra PCI slot. I prefer to run systems with a PCI SCSI card and a PCI network card, which isn't possible with a Voodoo 1/2, however it would be possible with a Voodoo3 (as it has 2D/3D on one card).

While I am happy with running a Matrox Millennium G200 for 2D/3D on my UMC-based 486, I am wondering if a Voodoo3 3000 PCI could squeeze out any added performance, and compatability, when mated with an IBM 5x86c-133/2x. I beleive the SiS boards at least have a 1:1/2 FSB-to-PCI divider, which would make using a 66 MHz FSB a possibility.

For those with a good 486 SiS PCI board, would you mind running the following benchmarks using a Voodoo3 300 PCI, 512 KB WB cache, 64 MB RAM (or 256 KB WB cache and 32 MB RAM), and either an IBM 5x86c-100/120/133, AMD 5x86-133/160, Intel DX4-120/133, or POD-83/100:

DOS
Symantec Sysinfo v8.0/8.1, PiDOS (75k digits), Landmark v2.0, Bytemark v2-32-bit, Roy Longbottom Dhrystone v1.1 (dhry1od), Roy Longbottom Whetstone (whetcod), Speedsys v4.78, Cachechk v7.0, 3Dbench1, 3Dbench2 ver 1.0c, Doom 1.9s timedemo 1, Pcpbench VESA Modus 100 640x400, Quake1 Timedemo 1 (320x200 & 640x480, console off)

Win98SE
SuperPi v1.1 (128k digits), Ziff-Davis Winbench96 CPUMark32 and Graphics Winmark, Ziff-Davis Winbench99 FPU WinMark99 and CPUMark99 (Stand-alone program), Ziff-Davis 3DWinbench97, WinTune98, Sandra99 (CPU, Multimedia, Memory), PassMark v4.0 (2D Graphics Mark, Memory Mark, Math Mark), 3DMark99Max, Final Reality v1.01, MDK Performance Direct3D/Software, and Quake2 (640x480 - Software).

The reason for listin some non-graphic type tests is to get a good idea for how well your SiS system is performing on the CPU, RAM, and cache side of things. For example, if our CPU, RAM, and cache performances for the SiS board are on par with those of my UMC board (given the same CPU), but your SiS/Voodoo3 system does much better in Quake and other graphic tests, then the Voodoo3 might provide some benefit over the Millennium G200, even though both systems are CPU-limited. This is my motivation anyway. I'd also like to know if the SiS board can handle a TNT2.

dirkmirk wrote:

I can run some tests if your interested, in speedsys my pci bandwidth with the the Voodoo3 is only around 16,480kb/s...

Some forum members have questioned the accuracy of Speedsys' video performance test, so I think it is a good idea to get a sample pool of benchmark results (list above).

dirkmirk wrote:

I can run some tests if your interested

Yes please!

dirkmirk wrote:

Im hoping to get an ALI PCI chipset board working that will be an interesting comparision.

My one encourter with an ALi PCI 486 board wasn't the best, but that's just one board. I look forward to your results, perferably with the above listed ensuite of benchmark programs.

feipoa wrote:

If you turn on branch prediction (BTB), you're speedsys score will jump to 75.4 (but you also need to turn off RSTK and LOOP).

Correction: That should be 76.43 with branch prediction enabled, not 75.4. Image attached.
Double Correction: I believe, both, BTB and LOOP were enabled for that Speedsys capture.