I have a few Intel 486DX4-100 SK096 and will try them out what is the highest OC they can manage - I think at least one would do 150 😉

kixs wrote:

I have a few Intel 486DX4-100 SK096 and will try them out what is the highest OC they can manage - I think at least one would do 150 😉

I have 10 of them 😁 (went a bit overkill). The three I tested wouldn't do 133 stable at 4v.
Then I moved to other projects 😁
I'll get back to that. The 486 I want as a daily one will be a more basic intel VLB one @ 120, but I'd really like to get a 133MHz one at 2x66 stable on my MB8433 😀

Hmm... I have at least 20 😉 Hopely I'll find that special one 😁

Yes, time is not on our side... so many cool things to do, to test... 🤣

Its possible its as easy (or as hard) to get an Intel DX4-100 to run at 3x50 MHz as it is to get it running at 2x66 Mhz as high FSB often limits overclocks somewhat. Im pretty sure 3x44.5 MHz would be really easy with 3.6V or 4V.

Its pity very few boards support FSB speeds between 40 and 50 MHz, I think some revisions of the Asus PVI-486SP3 can do 48 MHz but the PCI clock will still be high enough to mess with the built in I/O controllers, 44.5 MHz would be more useful. Its easier with the older ISA and VLB motherboards with a crystal that controls the FSB.

My Lucky Star LS486E boards only support 1/1 and 1/2 PCI dividers which makes the 50 MHz FSB setting more or less useless. I have another very interesting PCI board I need to tinker with though which could have the 2/3 divider.

Skyscraper wrote:

Its possible its as easy (or as hard) to get an Intel DX4-100 to run at 3x50 MHz as it is to get it running at 2x66 Mhz as high FSB often limits overclocks somewhat. Im pretty sure 3x44.5 MHz would be really easy with 3.6V or 4V.

Hadn't thought FSB could be a problem, since my PCI is still 33MHz.
I think I was able to complete PCPBench and 3DBench, but never doom (and quake is not even conceivable if doom doesn't go to the end)
Ok that topic made my 486 fiber tickle, I'll take out the test bench soon 😁

sunaiac wrote:

Hadn't thought FSB could be a problem, since my PCI is still 33MHz. I think I was able to complete PCPBench and 3DBench, but nev […]
Show full quote

Skyscraper wrote:

Its possible its as easy (or as hard) to get an Intel DX4-100 to run at 3x50 MHz as it is to get it running at 2x66 Mhz as high FSB often limits overclocks somewhat. Im pretty sure 3x44.5 MHz would be really easy with 3.6V or 4V.

Hadn't thought FSB could be a problem, since my PCI is still 33MHz.
I think I was able to complete PCPBench and 3DBench, but never doom (and quake is not even conceivable if doom doesn't go to the end)
Ok that topic made my 486 fiber tickle, I'll take out the test bench soon 😁

My experiences with high FSB limiting the top overclocks are with Socket-7 and newer platforms but it would not be stange if its the same with Socket-3.

Here are some more benchmarks with the Intel DX4-100@120

Cachechk

PCPbench VGA 320*200 8bpp: 23.6 FPS

PCPbench Mode 100: 10.6 FPS, Not too bad..."Klasse 4 Pentium 90 System mindestens 10,0"

I have some issues with running Quake, its not an issue with CPU stability but something having to do with the motherboard as it just wont run Quake v1.06 or newer with any CPU. I will copy and paste my post I wrote in the "What retro activity did you get up to today?" thread one and a half week ago. Since then I have even tried changing the cache, first to 4x512Kbit chips and then to another known to be good 8x256Kbit set which didnt help and now I obviously tried using a different HDD and DOS 7.10.

I diddnt think this issue would hinder me from benching Quake as older versions run fine but I just noticed that they do not have the timedemo function. Quake 1.01 accepts the "timedemo demo1" command but only plays the normal demos without doing any benching and I just read that its the same for v1.02 - v1.05.

From "What retro activity did you get up to today?" 2015-10-11

______________________________________________________

"Im trying to figure out why my Asus PVI-486SP3 board wont run Quake 1.06

Trying to run Quake 1.06 in DOS 6.22 always results in a page fault.

I have tried switching all hardware, CPUs (Intel DX2-66, Intel DX4-100 and AMD DX4-120), memory (lots of different FPM) and video cards (ISA, PCI and VLB cards). I have also tried with and without memory managers and with different BIOS settings but the result is always the same.

I think I have finally given up, its not like this system will be running Quake anyhow and everything else runs fine. The same Quake installation works fine using my Lucky Star LS486E board. The only thing left to test is using another HDD as the Asus board dosnt identify my 2.5GB test drive correctly but that will have to wait."

_______________________________________________________

Even if the PCI bus is still at 33MHz, there is a bunch of other stuff attached to the FSB. The northbridge, memory and L2 cache might easily limit overclocks.

sunaiac wrote:

I have 10 of them :D (went a bit overkill). The three I tested wouldn't do 133 stable at 4v. Then I moved to other projects :D I […]
Show full quote

kixs wrote:

I have a few Intel 486DX4-100 SK096 and will try them out what is the highest OC they can manage - I think at least one would do 150 😉

I have 10 of them 😁 (went a bit overkill). The three I tested wouldn't do 133 stable at 4v.
Then I moved to other projects 😁
I'll get back to that. The 486 I want as a daily one will be a more basic intel VLB one @ 120, but I'd really like to get a 133MHz one at 2x66 stable on my MB8433 😀

Got home and tried four SK096. All work at 3x40=120, only one boots at 3x50=150 but freezes in DOS. This one works fine at 2x66=133. At FSB 66MHz the cache and dram timings are the slowest possible and this takes a hit. It's not much faster than 3x40 with all settings to the max. I'll try the rest of my SK096 when I get the time.

Also tried four AMD 5x86-P75. Two ADZ and two ADW. Only one ADZ works fine at 3x60=180. The other posts at 180 but won't boot. ADWs max is 160MHz.

This is all on Gigabyte GA486AM/S V2.21 with 1MB cache. I'll post screenshots later. AMD @180 gets 103.0 in 3DBENCH 1.0c 😁

I've reach that :

I propose we create a new thread insteand of hacking that poor one 😁

Skyscraper:
Your PCPBench results are interesting.

While 320x200 is "slow", mode 100 is really fast. My 486-100WB running at 2x66=133MHz gets 31 in 320x200 and "only" 10.7 in mode 10. This is with "heavy" tweaks - I'm sure it's not even 95% stable 😉

More here (latest results are not yet published):
486 max

I think running the Intel DX3-100 at 120 MHz is a fantastic idea on your Asus PVI-486SP3. Due to the VL-to-PCI bridge on this board, the PCI graphics throughput tends to be a little slower than on some other PCI-based socket 3 boards. Running the PCI bus at 40 MHz helps to bridge that gap. I have several Intel DX3-100 chips with WB L1 cache, all of which will run at 120 MHz reliably. For Quake to run, you may need to up the voltage to your CPU. What voltage are you running? I beleive this motherboard has a 3.6 voltage jumper setting. Try that. If not, try 4.0 V. You might also want to ensure that your DRAM RAS to CAS Delay = 3 and that your DRAM RAS to MA Delay =2. If you are using L2 in WB mode, ensure that L2 TAG is set to 7. If using L2 in WT mode, ensure L2 TAG is set to 8.

kixs, I wouldn't go so far as to say that ADW chips are maxed at 160 MHz. The one stable example of an Am5x86-133 chip running at 200 MHz was an ADW chip.

I only told my experience with my four AMD 5x86-133 and didn't do any generalization about ADW.

feipoa wrote:

I think running the Intel DX3-100 at 120 MHz is a fantastic idea on your Asus PVI-486SP3. Due to the VL-to-PCI bridge on this board, the PCI graphics throughput tends to be a little slower than on some other PCI-based socket 3 boards. Running the PCI bus at 40 MHz helps to bridge that gap. I have several Intel DX3-100 chips with WB L1 cache, all of which will run at 120 MHz reliably. For Quake to run, you may need to up the voltage to your CPU. What voltage are you running? I beleive this motherboard has a 3.6 voltage jumper setting. Try that. If not, try 4.0 V. You might also want to ensure that your DRAM RAS to CAS Delay = 3 and that your DRAM RAS to MA Delay =2. If you are using L2 in WB mode, ensure that L2 TAG is set to 7. If using L2 in WT mode, ensure L2 TAG is set to 8.

kixs, I wouldn't go so far as to say that ADW chips are maxed at 160 MHz. The one stable example of an Am5x86-133 chip running at 200 MHz was an ADW chip.

As I wrote the issue with Quake has nothing to do with the CPU. I have tried with a stock Intel DX2-66, a stock AMD DX4-120, the Intel DX4-100 at stock and all these CPUs with both 3.45V and 3.6V (no 4V support). The result is the same, Quake 1.06 and Quake 1.08 dosnt run, Quake 1.01 works perfectly fine though.

If you read everything I wrote you will see that I have tried changing every piece of hardware including the cache and tag and tested different BIOS settings. This motherboard just wont work with Quake 1.06 or Quake 1.08 if the error isnt some strange BIOS setting I missed. My Intel DX4-100 SK096 is VERY stable at 120 MHz with 3.45V, I looped the game Blood ínside Windows 95 for 4 hours yesterday 😀.

I will test the more relaxed memory settings you suggested again but I think I already tried most combinations of settings. I also have a hard time understanding how the system can be 100% stable in Windows during extended periods of heavy load at 40 MHz FSB but not run Quake (1.06/1.08) in DOS (or Windows) at 33 MHz FSB.

My board cant use the newer BIOS versions because of my flash chip so Im using the latest supported one. This is another reason why Im using the Intel DX4 as the AMD 5x86 isnt supported or correctly identified by the BIOS Im using (It runs fine but will hang at soft reboots with 4x multiplier). As I ran a more or less identical system with a DX4 back in the 90s I feel that the DX4 is enough otherwise I would upgrade the flash chip to be able to run newer BIOS versions and CPUs.

The situation with Quake is really strange as my Lucky Star LS486E boards have no issue with running Quake 1.06 using the exact same CPUs and other hardware including 4x 512Kbit cache chips I have tried with the Asus PVI-486SP3 when trying to solve the issue.

Skyscraper wrote:

I thought that it would be a good idea to start out with a benchmark that puts alot of stress on the system... for a very long t […]
Show full quote

I thought that it would be a good idea to start out with a benchmark that puts alot of stress on the system... for a very long time! 😁

SuperPi 1M: 6087 seconds, also knows as 1h 43m 27s.

Intel_DX4-100@120_Asus_PVI-486SP3_SuperPi1M_trimmed.jpg

For performance references see this thread.
Post your retro computer superpi results here!

It seems member Tetriums Intel DX4-100 system using the same chipset spent 10960 seconds calculating 1M.

🤣! Yup, I seem to have an edge in the retrocomputer-community for building systems that perform exceptionally sloooooow 😁

kixs wrote:

Not the same CPU. Yours is WB (if its the same as my), 20% faster cpu clock and FSB. But even with all this there should be other differences as your result is like 40% faster. Maybe different BIOS optimizations.

Actually, iirc the CPU I used was indeed a WT chip and I think I suspected not all RAM was being cached (had 4x16MB FPM in it), but never tweaked it to perform faster as I reckoned that if I needed a faster rig, I could just build me a Socket 7 one 😜

Also the graphics card I used was in hindsight a rather poor choice (Rendition 2100 made by Diamond), but hey, it was the first retro rig I ever completed 😀

Some updates to the Quake 1.06/1.08 issue.

As this is the computer I first dug out of storage after my prolonged summer break from retro computing I can now report that Quake 1.08 works perfectly, it has probably always worked perfecty.

I later figured out what the difference was between this system and the other 486 class systems where Quake 1.08 worked, this system did not have the sound configured in DOS and I never tried Quake in Windows 95 with this system. I diddn't bother with DOS sound with this build as I was swapping sound cards and focusing on MPEG cards in Windows 95 and other Windows stuff.

I never imagined not configuring the sound card would hinder the Quake 1.06/1.08 installs I copied from my bench disk from running in DOS until I experienced the same thing with another system. The page fault error message diddn't make much sense and Quake 1.01 worked fine. I wasted so much time trying to track down this issue but I only swapped hardware and tried different BIOS settings.

So here it comes, better late than never. The BIOS settings are as they were pictured on the last page except "Slow DRAM Refresh (1:4) which is now active as it improved memory performance in Speedsys by a tiny bit.

Quake 1.08 timedemo demo1 with sound (CT2760) enabled. 11.7 FPS in DOS and 11.6 FPS in Windows 95 (quake.exe). In DOS Himem is loaded but not EMM386, mouse-driver, CD-driver and some other crap are installed.

The picture is showing the Windows 95 run.

Lets make (the) compatibility great again!

The last hour or so I have messed with UMBs trying to load everything high without using vide-cdd.sys or shsucdx.exe.

With the "standard" "Device=c:\...\emm386.exe RAM /MIN=0 I=B000-B7FF" line I got what seemed to be more than enough UMBs. For some reason it diddnt matter how much I reshuffled the drivers between UMB region 1, 2 and 3 something would always end up in the conventional memory although there were plenty of UMB memory left. I tried manually specifying more free memory blocks to use wich lead to some exciting error messages so no luck there but for once including "highscan" worked and added another 32kB to UMB region 3 and suddenly everything played nice.

It's a bit odd that although config.sys and autoexec.bat still are specifying that both atapicd.sys and mscdex.exe should be loaded into UMB region 2 I'm sure one or the other gets loaded into UMB region 3. I will not investigate this further though as it works but I might change ctmouse.com for mouse.com just to keep everything 100% vanilla as I still have almost 29kB left in UMB region 3.

Edit

Not tinkering any more... yea right...

I figured out that DOS 7 diddn't seem to like me excluding memory with EMM386 X=ranges to exclude. When I only used I=ranges to use and manually specified where the EMS buffer should go with frame=adress I could map out all parts of the upper memory area that should be available for UMBs without error messages. The odd thing is that I ended up with less memory than before so I went back to using "RAM /MIN=0 highscan I=B000-B7FF"

/Edit

Edit2

For some reason Quake gained a little performance from clocking the ISA bus higher.

Quake 1.08 in DOS with sound, 12.0 FPS. The ISA bus is running at 13.3 MHz instead of 7.159 MHz and EMM386 is loaded but the latter dosn't seem to affect the performance at all.

/Edit2

Today this build has made a leap forward but it wasn't the intention when I started out, I was just going to make a small improvement.

I was a bit tierd of having to listen to the somewhat loud sound the Maxtor 71080AP (1080MB) HDD I use in this system make and I also wanted some more space for games and stuff.

With the newest non Plug and Play BIOS the Asus PVI-486SP3 does not like larger HDDs, thankfully it isn't stuck at the 504MB limit but anything over 1GB is hit and miss and when testing all my spare suitable sub 8GB HDDs the largest one the BIOS would detect correctly was a Quantum Fireball 1700AT (1707MB). This HDD is reasonable quiet and in good working order, I made a surface scan in the Asus PVI-486SP3 system just to be sure and everything was fine.

These are the drives in question.

The next issue was transferring the system to this HDD with my Norton Ghost software. I could not get my main system with an EVGA SR-2 motherboard to boot with an old AT drive connected to the on board VIA controller handling the onboard IDE so I had to resort to this low quality IDE to USB thingymabob I found in one of my "random crap" boxes.

The IDE to USB thingymabob.

The thingymabob would only detect the HDD if I first connected power to the HDD, then really fast connected the thingymabob to the HDD and then the USB cable to the computer. If I tried to do it in any other order the HDDs would not even spin up or the HDD would not get detected. In any case this seemed to work, at least when I got the timing right. I could create an image of the Maxtor drive with Norton Ghost and write the image to the Quantum drive.

I connected the Quantum drive to the Asus PVI-486SP3, booted with a floppy and did a "sys c:" without issues. I could boot from the drive and everything seemed to work but when using the system I started to get read errors and when doing a surface scan I found that the drive was now littered with bad sectors. That's what I get for trying to image old HDDs with a cheap USB to IDE thingymabob.

As I do not own another HDD larger than the 1080MB Maxtor drive that will work with the latest BIOS supported by the old flash chip with 12V programming voltage my board is equipped with I was getting a bit irritated by this point. By changing the flash chip to a SST PH29EE010 or Winbond W29EE01 chip with 5V programming voltage a newer PNP BIOS that won't work the the 12V chip because of some strange incompatibility (not the BIOS size) can be used. This newer BIOS should with luck support larger drives.

I did a quick Ebay search and found that I would have to pay $12 with shipping for the cheapest SST PH29EE010 as the Chinese diddn't seem to have any and I could not find any Winbond W29EE01 chips at all. Spending $12 and having to wait days or even weeks sucks so instead I checked some motherboards in the first random box with motherboards I happend to stumble on in one of my storage units. I found this Shuttle? i440LX board with a Winbond W29EE011 chip. Only a single digit difference, close enough!

The donor i440LX board, probably a Shuttle HOT something.

Next I tried to find my "flash chip programmer", a i430VX motherboard with the BIOS chip easy accessible at a corner making it easy to hotflash. It turned out that I have no clue where that motherboard is, it's at least not where it should be. I decided to use the Asus PVI-486SP3 to program the new BIOS chip. The manual would not tell me the correct position to put JP32 and JP33 to get 5V flash programming voltage but trusty old TH99 listed the jumper settings (JP32 1-2, JP33 2-3).

"Hotflashing" is always exciting, even more so when the motherboard is mounted in a case. Luckily everything went perfectly fine and my Asus PVI-486SP3 now sports the 0307 PNP BIOS with Y2K fixes. I hope this BIOS will support larger HDDs but this BIOS will also let me run an AMD 5x86 P75 CPU! I dont know if I will upgrade the CPU permanently but I will at least do some benching with the 5x86 P75. 😀

The new BIOS chip in place without the magic smoke escaping!

So far so good!

😀

Edit

Now a "quiet" Western Digital Caviar 22500 (2559.8 MB) HDD gets detected correctly! I think I will transfer the system to the new HDD without fancy ghost software and USB to IDE thingymabobs this time, lesson learned! 😀

/Edit

I have tranfered the system and the rest of the stuff to the new drive. I disabled the swapfile and copied everything in Windows using the 486 system, not as fancy but at least it works. I have also installed an AMD 5x86-P75 and done some benchmarking.

The AMD 5x86 running as an AMD 486DX4-120 (16kB WB) gets 11.5 FPS in Quake 1.08 with sound enabled so it's 0.5 FPS slower than the Intel DX4 @120.

In speedsys the AMD CPU is also slower than the Intel DX4.

The Intel DX4 does not support a 4X multiplier while the AMD 5x86 does, when running as a "5x86-P90" at 160 MHz the AMD CPU gets 14.3 FPS in Quake with sound enabled.

It's also faster in Speedsys CPU test by a large margin, when it comes to memory performance the Intel DX4 @120 and the AMD 5x86 @160 are neck and neck.

Pcpbench performes exactly the same in "VESA Modus 100" (640x400 8bpp) with the AMD 5x86 @160 as with the Intel DX4 @120. When benching "VGA modus" (320*200 8bpp) the AMD 5x86 @160 is even 0.3 FPS slower than the Intel DX4 @120. This is somewhat odd, perhaps it's the new PNP BIOS that isn't as fast as the old non PNP BIOS or perhaps Pcpbench really likes Intel 486 CPUs.

To sum up the results the Asus PVI-486SP3s performance does not seem that great in Quake, I have got better scores with an AMD 5x86 @160 MHz using other motherboards. It could be the PCI Matrox Millennium II that isn't a top performer in Quake, perhaps another PCI card would be faster although from what I have seen of member feipoas benchmarking I reckon the PVI-486SP3 only performs at its best with a VLB video card.

At least playing MP3s with Winplay3 now works flawlessly with the AMD 5x86 @160MHz! With the Intel DX4 @120MHz MP3s started stuttering a bit after playing for about 60s if I didn't use the half frequency output option.

I believe the Trio64 may be faster in Quake compared to the Matrox Millennium II. Try a Virge or Trio64 and see what you get. I do not own a Millennium II. Your Quake scores are low. To determine if they are too low, run with a Trio64. Read this thread. I tested this board and compared it to others? Performance comparison of 486 motherboards with VLB-only, PCI-only, and PCI+VLB

Although the SiS 496 chipset on the PVI-486SP3S is a fast chipset for the PCI 486 platform, both the PCI and VLB speeds are slow on this motherboard. Refer to the above thread. Also keep in mind that if using Phil's VGA Benchmark Suite, that the Quake scores come out lower than running Quake 1.06 demo on its own.

I sorta recall an Intel DX4-120 being able to just play Mp3s without dropping the resolution in half. Are you using a write-back version? And maybe you can try Winamp v2.05 with the spectrum analyzer off and turn off scrolling the song name.

feipoa wrote:

I believe the Trio64 may be faster in Quake compared to the Matrox Millennium II. Try a Virge or Trio64 and see what you get. I do not own a Millennium II. Your Quake scores are low. To determine if they are too low, run with a Trio64. Read this thread. I tested this board and compared it to others? Performance comparison of 486 motherboards with VLB-only, PCI-only, and PCI+VLB

Although the SiS 496 chipset on the PVI-486SP3S is a fast chipset for the PCI 486 platform, both the PCI and VLB speeds are slow on this motherboard. Refer to the above thread. Also keep in mind that if using Phil's VGA Benchmark Suite, that the Quake scores come out lower than running Quake 1.06 demo on its own.

I sorta recall an Intel DX4-120 being able to just play Mp3s without dropping the resolution in half. Are you using a write-back version? And maybe you can try Winamp v2.05 with the spectrum analyzer off and turn off scrolling the song name.

I tested Quake with a PCI Diamond S3 Trio64 video card and got the same 14.3 FPS. Luckily it dosn't matter that this computer is a bit slower in Quake than it should be. This build is more of a nostalgic build as it's sort of a rebuild of the first PC I built on my own.

I think the AMD 5x86-P75 (AMD 486 DX5-133) CPU will stay in this build as I have loads of these CPUs while I do not have more than a few Intel DX4-100 CPUs. I might find this Intel DX4 handy for benching in other motherboards as it has Write Back L1 cache and I know for sure that it's capable of handling 120 MHz.