Jump to content
HWBOT Community Forums

SPI 32 M @ i486 - oldschool OC


ludek

Recommended Posts

Hi. I'm going to run SPI on something slower than my slowest CPU so far: AMD K5-PR75 LINK (it took a bit more than 21).

 

There are some problems.

1. Operating system,

2. Stability(extremely long counting operations eg. ~1-2 weeks? or more?),

3. Performance :) - memory issues.

 

 

The biggest is the operating system. Most comfortable way and (top performance) is Windows XP, but as far as I know, this system can't be used with that old CPU. Why? There are answers found on the internet:

There are several reasons why Windows XP does not run on a 486:

1. Most 486 computers simply cannot support enough memory to run Windows XP. 64 MB is pretty much the maximum you'd ever find in a 486; 8 to 16 MB is most common.

 

2. The kernel in Windows XP is SMP-enabled by default. To support SMP, Windows XP uses an instruction called CMPXCHG8B, which is only found in Pentium and later processors.

 

3. In order to allow for optimizations on-the-fly, such as using MMX or SSE instructions, Windows XP also requires the CPUID instruction, which only the later 486 processors had.

 

 

 

1. I know that XP can be booted with 64 meg of ram, it does work on TUSL2, PIII. I bought two special, still-working, high-performance, high-end socket 3 motherboards with "ultra fast" EDO-Ram (72pin). I think, these can work with at least 64 MB memory. Both have 4 mem slots. I have also 4x32 MB (high end) non-ecc memories, will try if they work. One of these motherboards have also 256 kB of cache, which is pretty good amount, but this one has only ISA ports. Manual says also about 128-meg support. Second motherboard has got few PCI ports and it's not bad, but has less amount of cache (don't know yet, how much).

2. Is anyone able to explain this? ??

3. MMX and SSE are supported by Win XP, but are not required to run. A prove is my 21 hours SPI score on AMD K5, which have not built-in MMX feature.

 

 

I'm thinking about NT 4.0, which can work with SPI 1.5XS. Possible problems: device drivers, maybe less stability.

 

 

 

OK, so next "big" problem is the stability:

1. A/C power stability,

2. System stability.

 

The first one can be solved by UPS with a huge accumulator.

 

Second is system stability.

I was making some tests on "oldschool" platforms. This is very important and stability depends on it. The Pagefile. When I was making SPI32 using Windows XP, Pentium MMX and 128 MB ram, system was crashing after 4-6 hours... I have done some research and it turned out, the Pagefile was turned off by me, because of it's location on slow HDD. When a platform has 512 MB of memory, missing pagefile was not problem anymore. Everything was rock stable.

 

So, when I was counting SPI 32 MB on AMD K5 PR-75, there was a need to connect 3 hard drives to achieve best performance. The only motherboard working with that old AMD K5 could support max 128 MB of memory, and so pagefile must be on. One disk for the system (4 GB because of problems with MB support), one for SPI and one for Pagefile.sys.

 

No matter if I choose XP or NT4.0, this solution would give me a best system performance possible and will provide stability, but there are other problems...

 

 

 

 

Hardware problems:

1. no integrated IDE controller

2. drivers, drivers, drivers...

 

 

I have SOYO motherboard with 256 kB, very well motherboard, with very well clock generator (25;33;40;50;66;75;100 MHz support ! ). But it only have ISA ports and no IDE. I have an old ISA IDE controller, but it is quiet slow for shure and don't know if I will find any drivers.

 

Second motherboard has IDE controller but I'm not shure if is there possible to find any drivers, but it has got PCI and I can put external IDE controller... but it don't wanted to boot from external so called "external scsi adapter", there is no such-option in Bios, but I have drivers to this.

 

 

So if anyone has anything to add or has got any advice, please tell me :)

 

To be honest i have also not much expirience with very old platforms, I had once 486DX2 notebook when I was in the middle of elementary school, there was no problems with audio or IDE or modem or any other devices, no IRQ settings and things like this :P

So... wish me luck :D

Edited by ludek
Link to comment
Share on other sites

As far as I know, SPI 1.5 XS can't run on Win 98. Few days ago I made a performance comparison between Win NT4, XP and Win 7 - using virtual machine.

Win 7 has got the top score because of virtual machine program was working slower.

 

SPI on each system was working separately, XP, then NT4, then win 7.

 

 

 

systemcompare9pcjp.jpg

 

 

 

 

System spec: I7 3770K @ 4.3 GHz, DDR3-1600 @ 7-8-8-24 (Elpida BBSE), Asus Z77

Edited by ludek
Link to comment
Share on other sites

Socket 5 CPU can run on socket 7 motherboard. I have eg. AMD K5 or non-IHS versions of non-mmx pentium 1 (P54C) they are oficially on Socket 5 platform, but my Socket 7 mobos are working with those :)

 

I use Soyo 025N2 and PL4600.

 

 

I have tested, both working with 128 MB ram :)

 

There are lots of things I don't understand, eg when connect a hard disk to my Win 7 PC, then it can't work back in the old 486 motherboard (windows error "inaccessible boot device"). Installed 3 times NT4.0 yet :P

 

 

Edit:

Nope, NT4.0 will be installed 4th time :rolleyes:

 

 

I have installed NT4.0 on Ultimate-epic-performance Socket 7+Pentium MMX hyper-fast motherboard, added IDE PCI controller, installed drivers, changed "C" drive IDE-cable to this-one and it was booting... but another experiments with UMC IDE controller ruined my system data :( Ehh...

Edited by ludek
Link to comment
Share on other sites

  • Crew

ludek111, WinNT should be your OS of choice, though Win2K should run too. Like it's said, XP requires CMPXCHG8B support but it has nothing to do with SMP (CMP stands for compare, not multiprocessing). It was implemented in Pentium family only.

I haven't benched 486 yet, but I've read good enough. Feel free to ask.

Link to comment
Share on other sites

New problems on my way,

1.destroyed partition, reinstall (6th time),

2.SOYO motherboard is not stable(memory problems, bluescreen while installation, black screen (no VGA signal)),

 

 

Just repaired my old ET4000AX graphics card with 1MB memory :) 92year :D

PHOTO - ICS 9214's DIP socket destroyed by motherboard battery leakage, cleaned DIP of "MUSIC" chip, replaced crystal 14.3 MHz and rusty legs of memory cleaned also. Now it's working :)

 

Photos of my motherboards:

SOYO 1

SOYO 2

 

PL4600 1

PL4600 2

 

Working Soyo

Disk errors

 

 

It turned out that PL4600 motherboard has no L2 cache, I don't know why I didn't noticed this. PL4600 and Am486DX4-100 = SPI 1MB in something like 50 minute because of no cache and pagefile problems (as always:( ).

 

 

I have to make Soyo more stable first, then I will back to this project. :)

Link to comment
Share on other sites

  • 1 year later...

Resurrecting this thread

 

Did you manage to run 32M on a socket 3 system?

 

I have tried to run Spi 32M on a socket 3 system with a Am5x86-p75 and 96MB EDO and WinNT4. I have a 512MB CF-card as system disk and another disk with 1GB FAT16 partition which I run Spi from. Page file is 200MB on system disk.

 

The 32M run seems to run fine for about 24 or so hours, just before the end (I havn't seen this live yet, always ends at night or when I'm at work), and when I turn the screen on, SuperPI process is terminated and there is a Dr.Watson exception. My theory is that it crashes when the calculation is done and SuperPi is finishing up.

 

Could the problem be the page file? Should I put it on another partition? Enlarge it? What about partition sizes? NTSF vs FAT16?

 

I have not installed any service pack yet since WinNT4 did not start after trying to install SP6a. If I put the page file on another partition there is enough disk space so that I could try to install all service packs one by one and see if that works.

 

 

 

 

New problems on my way,

1.destroyed partition, reinstall (6th time),

2.SOYO motherboard is not stable(memory problems, bluescreen while installation, black screen (no VGA signal)),

 

 

Just repaired my old ET4000AX graphics card with 1MB memory :) 92year :D

PHOTO - ICS 9214's DIP socket destroyed by motherboard battery leakage, cleaned DIP of "MUSIC" chip, replaced crystal 14.3 MHz and rusty legs of memory cleaned also. Now it's working :)

 

Photos of my motherboards:

SOYO 1

SOYO 2

 

PL4600 1

PL4600 2

 

Working Soyo

Disk errors

 

 

It turned out that PL4600 motherboard has no L2 cache, I don't know why I didn't noticed this. PL4600 and Am486DX4-100 = SPI 1MB in something like 50 minute because of no cache and pagefile problems (as always:( ).

 

 

I have to make Soyo more stable first, then I will back to this project. :)

Link to comment
Share on other sites

First, my system is unstable. Its memory issues or CF card is hmm... too fast? I don't know, wasted so much time. Maybe will try again.

 

NTFS format appears to work better and faster, but causes problems if I connect CF card to another PC with modern OS. Then it destroys filesystem and I'm forced to instal NT again :( :(

FAT is slower but less problems with copying data.

 

Please take a look to any SPI 32 MB score. In most (?every) case it takes 256 MB of memory. Disk usage is 8 times than that or more. Note that if running 1 mb spi, it writes 8 MB of data in SPI folder. I think that amount of data written on hard drive is 8 times of choosen sample or more. I don't remember exacly but maybe 128 MB x 8 is written to complete SPI32M calculations. 1GB CF card was able to handle this when I was doing SPI32 on AMD K5 (in my HWB scores).

 

 

Connect another hard drive just for pagefile, another for SPI and it might be good if another is for system. 3 discs all shall be good.

 

 

I don't know if any service pack is giving something more to performance. I have enough of NT4.0 but I must say it's fast for SPI and takes tiny amount of system resources.

 

Keep pushing!! :)

Link to comment
Share on other sites

SuperPi 32M test is heavy on HDD on it's finish. You need to finetune your machine to have stable I/O operations to pass this. It crashed for me at least 10 times for many about ~20h runs. It is hardcore, but it is managable...

 

Mostly PCI latency helps (from default 32 to 42 it was need to pass SPI 32M test on 125MHz P90 on TXP4-X), but changing components change the game. Using ATI Rage XL completely change the need settings over a S3 Trio64, for example.

Link to comment
Share on other sites

I'm fighting with Win 2K now. It's possible to install this on 486 platform and it's a way better than NT4.0 (wow USB support). After ~5-6 days of struggling, trial and crosses... I discovered that W2K SP4 is not able to be installed because of "sp4.cab" file:

"The following value in the .SIG file used by setup is

corrupt or missing:

 

Value 0 on the line in section [ SourceDisksFile ]

with key "sp4.cab."

 

Setup cannot continue. To quit setup, press F3.

"

 

Clean w2k is installing nicely on this platform and maybe I'll be able to run CPUz :) WinNT was so annoying. Win 2K has caused even more problems but they was more obvious and solvable. Drivers missing, too small partition size, unstable system, non-ACPI devices and more others, but I choosed W2K as a good old-school system. It takes more than well-cutted XP, but it's good enough.

Link to comment
Share on other sites

Well, if that does conform you a bit, then I tell you, that todays for the 4 or 5th time SuperPi 32M run crashed for me on the AMD K5 PR75 of mine at the end, when saving the score... I gut WinNT to bones to get good results, but it always crash at the end and I find no reliable way to restore the UDMA2 HDD access, witch is also key to speed.

 

It is either 20MB/sec with little CPU usage (mind that this is K5 PR75 @ 90MHz, so not exactly fast CPU by any standards: http://postimg.org/image/qx7kmwknx/ ) or around 2MB/sec with 85+% of CPU usage. That suxx.

 

Quess what impact that have on the score.

 

...

 

So I sort of give up and try a XP run. While WinNT gives me under 1h per loop, WinXP need 1h 10+min per loop... and it probably crash again...

Link to comment
Share on other sites

I just want to let you know that this thread is still alive :D

 

LINK

 

Extremely slow Num-Lock response video. Warning! Action starts at 2:10 min :D

It took almost 2 minutes to react to pressed numlock button... horrible

 

I give you advice. If running such slow system it's good to clean all slots eg. EDO ram slots, mems connections, CPU pin grid array and the socket and power connector also. Here i call it "contact spray". It's the best and I don't see any high frequency issues when I do not clean cleaner-residues on cpu. It's stable as a table.

 

do not give up and wish me luck when taking screen shot. I forgot connect UPS. Hope that power station does not fail... Hardcore begins

Edited by ludek
Link to comment
Share on other sites

Well........ Guess what just hapened?! I was looking from time to time at the 486 display and... Blank colour of desktop was the thing I seen. I was so happy to see this:

 

img_4842z5uid.jpg

 

But just 1-2 hours later it crashed :( :(

 

WHYYYY?!?!? Is there a bug in windows guts? It the aplication - CPU time limitation to 24 hrs or something? I do not understand this crap. Enough time waisting for this month. At least I did SPI1M and pifast benchmarks :)

Link to comment
Share on other sites

Just 1-2 hours later it crashed :( :(

 

WHYYYY?!?!? Is there a bug in windows guts? It the aplication - CPU time limitation to 24 hrs or something? I do not understand this crap. Enough time waisting for this month. At least I did SPI1M and pifast benchmarks :)

 

Interesting thought, is there any submissions at all thats longer then 24 hours posted from superpi?

 

Got my thinking about some info from this post (that I shorted down to only include the relevant part)

 

Depends on the hardware tested, but yes, running Spi32/W1024 on Pentium 1 will set you for 30 hours+. Do remember Wprime has an issue when a run takes over 24hrs, make sure you revert the OS time back every 23ish hours.

 

 

Could it be that superpi got the same problem as Wprime with 24h+ runs?

Link to comment
Share on other sites

Folks, unfortunately I'm not able to offer help, but I want to say I honor the effort you put into benching these old machines. :ws:

The oldest PC I've ever used had a 286 CPU inside, but it was DOS only. At least for me it's very interesting and I'm crossing my fingers for your success. :)

 

While waiting for the next button to respond you can have a watch at this video. Those guys will tell you how incredibly fast a 486 actually is. :D At 6:37 for example you can see how they were benchmarking hardware in the 80s.

 

Link to comment
Share on other sites

Interesting thought, is there any submissions at all thats longer then 24 hours posted from superpi?

 

Got my thinking about some info from this post (that I shorted down to only include the relevant part)

 

""Depends on the hardware tested, but yes, running Spi32/W1024 on Pentium 1 will set you for 30 hours+. Do remember Wprime has an issue when a run takes over 24hrs, make sure you revert the OS time back every 23ish hours.""

 

 

Could it be that superpi got the same problem as Wprime with 24h+ runs?

 

 

I will try this. My AMD K5 PR75 submission was <24 hrs, so it makes sense. Thanks for interest and help.

 

 

@UP

Thanks :) Yes I have got more time will watch this, look funny.

Edited by ludek
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...