VOGONS


486 eBay pickup - new project!

Topic actions

Reply 20 of 34, by chinny22

User metadata
Rank l33t++
Rank
l33t++

For system management (copying files, networking, etc) Win95 is fine and is 100 times more easier then a dos/win3x system.
But for any gaming you'll still be dropping back to dos mode with added bonus of Fat32 and long file name support.

But for that pure nostalgia hit you have to have Windows 3.11 running on top of dos 6.22 and all the limitations it imposes on you 😀

Reply 21 of 34, by Intel486dx33

User metadata
Rank l33t
Rank
l33t

I remember when Win95 first came out back in 1995. I went to the release party at a local Amusement park in Silicon Valley at the time called “Marriotts Great America”. Microsoft and Bill Gates had a BIG Celebration there and were giving away demo CD’s of Win95.
So I picked one up and tried it on my 486dx-33 computer with 4mb ram and 5400rpm iDE hard drive.
Well it loaded but was really slow and took for ever to install. I later upgraded the motherboard to a VLB with 486dx2-50 motherboard
And 8mb of ram But it was still to slow for Win95.

I have a AMD 5x86-133 right now with 32mb ram and CF card and it is running WIn95. It works okay for Win95 but it struggles to play
Some MP3 music or Video.

What runs amazingly well is Win3.11 on a Pentium 200mhz CPU. This is how Win3.11 is suppose to run.
It runs so well and responsive on a Pentium 200 it is amazing.
Infact I remember back in computer education school we use to run WinNT-351 on Pentium CPU computers
Pentium 75 thru 100mhz computers.

In Fact I think this is the best CPU for DOS gaming and Win3.11
For DOS gaming or Win3.11 a Pentium 75 thru 233mhz works bests.

On the 486 computers that I do have I only run DOS and WIn3.11
It performs okay but its like night and day compared to a Pentium CPU.
You really notice the difference.

I would not waste to much money on that 486 computer if you plan on running Win95 on it.
16mb of RAM or 32mb. max should be enough for Win95.
64mb of ram will not help improve performance much .
The 486 CPU is the bottle neck

Its going to run very slow but it probably will work.

I have found that for DOS gaming a 486dx2-66 CPU or higher is best.
256kb of cache is the sweet spot but I would not waste my time and money on cache.
A CPU upgrade will help improve performance best.
16mb of RAM is the sweet spot. Anything more is just excessive for Win3.11

The CPU is really the bottle neck in performance.
Adding more RAM and cache will only minimally help improve performance.
For noticeable performance gains you really need to upgrade the CPU.

Last edited by Intel486dx33 on 2021-07-28, 14:22. Edited 3 times in total.

Reply 24 of 34, by H3nrik V!

User metadata
Rank Oldbie
Rank
Oldbie
Disruptor wrote on 2021-07-28, 13:51:

With 64 MB RAM you have to switch your L2 cache to the slower write through mode when it is not larger than 256 KB.

Isn't that more a question of TAG ram size?

Please use the "quote" option if asking questions to what I write - it will really up the chances of me noticing 😀

Reply 25 of 34, by britain4

User metadata
Rank Member
Rank
Member

I think I’m sticking to 3.1 on it, it’s kind of fun and different working with it and it’s more appropriate to the hardware. At the moment it’s a DX2-66 in there which I think is decent enough considering I’ve got my P233 system for more demanding stuff.

I’ll get some photos of the trace repairs up sure - they’re not the prettiest but I’m totally happy they’re strong repairs that will last.

- P-MMX 200MHZ, PCChips M598LMR, Voodoo
- P-MMX 233MHz, FIC PA2013, S3 ViRGE + Voodoo
- PII 400MHz, MSI MS6119, ATI Rage Pro Turbo + Voodoo2 SLI
- PIII 1400MHz, ECS P6IPAT, Voodoo5 5500
- Toshiba Libretto 110CT, 300MHz, 96MB RAM

Reply 26 of 34, by jakethompson1

User metadata
Rank Oldbie
Rank
Oldbie
britain4 wrote on 2021-07-28, 21:15:

I think I’m sticking to 3.1 on it, it’s kind of fun and different working with it and it’s more appropriate to the hardware. At the moment it’s a DX2-66 in there which I think is decent enough considering I’ve got my P233 system for more demanding stuff.

I’ll get some photos of the trace repairs up sure - they’re not the prettiest but I’m totally happy they’re strong repairs that will last.

Another option is Windows for Workgroups 3.11. Add an ethernet card and you can network the system using the MS TCP32B drivers.
A neat thing about the DX2-66 in particular is that it was the last time Intel's fastest x86 processor was a 486. All the faster models came after the Pentium. And rather than Celeron, Pentium, i3 etc. all existing where there are new models for every market segment at the same time, low end systems simply got the "fast" processor from two or three years earlier.

Reply 27 of 34, by jakethompson1

User metadata
Rank Oldbie
Rank
Oldbie
H3nrik V! wrote on 2021-07-28, 20:52:
Disruptor wrote on 2021-07-28, 13:51:

With 64 MB RAM you have to switch your L2 cache to the slower write through mode when it is not larger than 256 KB.

Isn't that more a question of TAG ram size?

Kind of. There are only as many tags as cache lines. Because the 486 internal cache has lines of 16 bytes, it seems external caches usually use that line size also. So a 256K cache needs 16K tags. A larger tag RAM (like 32K) wouldn't help.

Where you do make a point is that tags don't have to be 7 bits (if a bit is robbed as a dirty bit) or 8 bits (if there is either an external dirty bit, or no dirty bit needed because the cache is write through). A cache could have used 9-bit or 16-bit tags for example. I guess it either wasn't economical, or just didn't matter since these cacheable areas didn't feel limiting at the time. It was like when FAT32 came out, and had a maximum partition size of 2TB ("wow!").

If you break down a memory address you get:
4 bits - for the location in the cache line
14 bits - for which cache line (for a 256K cache)
7 bits - tag for write back cache with robbed dirty bit

That's 25 bits and 2^25 is 32 megabytes. Increase the tags to 8 bits by switching to write-through and you double the cacheable area to 64 megabytes. And besides the 7- or 8-bit tag, the only other thing you have control over in this design is the size of the cache. Every time you double the cache size beyond 256K you increase the number of cache line select bits by one and therefore double the cacheable area.

Reply 28 of 34, by britain4

User metadata
Rank Member
Rank
Member
jakethompson1 wrote on 2021-07-28, 22:50:
britain4 wrote on 2021-07-28, 21:15:

I think I’m sticking to 3.1 on it, it’s kind of fun and different working with it and it’s more appropriate to the hardware. At the moment it’s a DX2-66 in there which I think is decent enough considering I’ve got my P233 system for more demanding stuff.

I’ll get some photos of the trace repairs up sure - they’re not the prettiest but I’m totally happy they’re strong repairs that will last.

Another option is Windows for Workgroups 3.11. Add an ethernet card and you can network the system using the MS TCP32B drivers.
A neat thing about the DX2-66 in particular is that it was the last time Intel's fastest x86 processor was a 486. All the faster models came after the Pentium. And rather than Celeron, Pentium, i3 etc. all existing where there are new models for every market segment at the same time, low end systems simply got the "fast" processor from two or three years earlier.

I’ve never dabbled with networking on my old machines - something I may have to look into. Most of them would need cards adding for a start.

That is quite a cool thing to note, the build date on everything seems to have been ‘94 so most probably a budget-ish system by then

The cache/RAM issues are something I’ll need to look into - the board does have real cache and I THINK it’s 256k. 32MB of RAM might end up working out fine so maybe I’ll only use 2 of the SIMMs after all.

- P-MMX 200MHZ, PCChips M598LMR, Voodoo
- P-MMX 233MHz, FIC PA2013, S3 ViRGE + Voodoo
- PII 400MHz, MSI MS6119, ATI Rage Pro Turbo + Voodoo2 SLI
- PIII 1400MHz, ECS P6IPAT, Voodoo5 5500
- Toshiba Libretto 110CT, 300MHz, 96MB RAM

Reply 29 of 34, by chinny22

User metadata
Rank l33t++
Rank
l33t++

Another vote for WFW 3.11
It has various bug fixes and refinements but looks and acts just as 3.1
The inbuilt networking only installs if you have networking capabilities so wont impact system performance if you don't set this up.

and while I agree 95 is useable on a DX2/66 I also thing Win3x is a better fit.

Reply 31 of 34, by mkarcher

User metadata
Rank l33t
Rank
l33t
jakethompson1 wrote on 2021-07-28, 23:07:

That's 25 bits and 2^25 is 32 megabytes. Increase the tags to 8 bits by switching to write-through and you double the cacheable area to 64 megabytes. And besides the 7- or 8-bit tag, the only other thing you have control over in this design is the size of the cache. Every time you double the cache size beyond 256K you increase the number of cache line select bits by one and therefore double the cacheable area.

If the chipset / mainboards supports it, after all. I have a 486 mainboard that uses 8 bits for tags (I'm unsure at the moment whether it is write-through only, I manually configured it to write-through, or uses an "always dirty" write-back strategy), so it should be able to cache 64M. As that board uses an old chipset, it doesn't map address lines to tag bits within the chipset, but you have to provide "the 8 address lines that are to be used for tag matching" to dedicated address tag bit input pins. The board I am talking about didn't bother to make A25 routable to the tag input pins, so in 256K cache configuration, the board uses just 7 tag bits, without the chipset knowing!. It's not that bad as the board is maxed out at 32MB RAM (8 * 30-pin SIMM), but this board supports memory relocation, and remaps at least the 128KB from A000-BFFF to the top of the address space in default configuration. So the maximum RAM address is beyond 32MB. As the chipset expects all 8 tag bits to be validly provided, it tries to cache both 32MB+16 bytes and 0MB+16 bytes, but the board mixes these addresses up because A25 is missing in the tag value. This causes crashes with 32MB RAM, unless you either disable memory relocation or manually add a noncachable area at 32MB.

So increasing cache size only enables doubling the cachable area, but it depends on the chipset and/or board whether the possibility of doubling the cachable area is actually used. Most PS/2-SIMM based 486 boards have chipsets that are recent enough to provide internal address->tag mapping, so it likely isn't a common limitation in practice.

Reply 32 of 34, by britain4

User metadata
Rank Member
Rank
Member

At the moment still running 4MB RAM - the 64MB kit arrived and when it got here it turned out to be EDO. Tried it out anyway but looks like the board isn't going to play ball with it - checked the ad and I just ordered the wrong thing so now waiting on a non-EDO kit to try. D’oh.

I have a CL 5428 VLB card on the way for it too. The sound card is another thing - it's an SB16 Value which is fine but has no wavetable header for adding a Dreamblaster etc so somewhat limiting there. Might consider one of the "Chill & Phil" MIDI port adapters given the prices of ISA sound cards these days however I do also believe it's one of the cards affected by the hanging note bug - so a new sound card may be a good purchase too.

On further investigation my sound card does have the CT1747 bus interface so is immune to the worst of the hanging note bug and I think it’ll actually be fine. Just as well as I’d probably have to replace the Panasonic interface CD drive if I was swapping out the sound card.

- P-MMX 200MHZ, PCChips M598LMR, Voodoo
- P-MMX 233MHz, FIC PA2013, S3 ViRGE + Voodoo
- PII 400MHz, MSI MS6119, ATI Rage Pro Turbo + Voodoo2 SLI
- PIII 1400MHz, ECS P6IPAT, Voodoo5 5500
- Toshiba Libretto 110CT, 300MHz, 96MB RAM

Reply 33 of 34, by britain4

User metadata
Rank Member
Rank
Member

Ordered a Dreamblaster X2GS for this today - been going back and forth on it and finally decided to splash out on one so very excited for that.

Also ordered the "Chill & Phil" adapter so that it can be plugged into a soundcard gameport and not just a Waveblaster header on the card - this will allow me to use it with the SB16 in the 486 build as well as switch it between other machines if needed, including my AT Socket 7 build which is using onboard audio. Can't swap any sound cards around due to compatibility and I'm not paying eBay prices for another card so until I come across another one - the Yamaha card in my PII build has adequate MIDI capabilities if I need it with the main DOS duties being taken over by the S7 builds and below.

The cache info in here I've found really useful so thanks everyone for contributing to that - looks like this build will be getting 4 8MB (non-EDO!) SIMMs now all being well.

- P-MMX 200MHZ, PCChips M598LMR, Voodoo
- P-MMX 233MHz, FIC PA2013, S3 ViRGE + Voodoo
- PII 400MHz, MSI MS6119, ATI Rage Pro Turbo + Voodoo2 SLI
- PIII 1400MHz, ECS P6IPAT, Voodoo5 5500
- Toshiba Libretto 110CT, 300MHz, 96MB RAM

Reply 34 of 34, by britain4

User metadata
Rank Member
Rank
Member

Have been messing around with my Dreamblaster X2GS and it's absolutely great, the GS bank feels just perfect for a 90s MIDI experience. The Buran bank sounds incredible (for MIDI) in terms of realism but just doesn't seem to fit with retro games as nicely. Yet to play around with any other sound fonts or anything but it's a great little device and pretty convenient using the adapter and sound card line in.

Have now fitted a 3xAAA holder with a 1N4002 diode in series for the BIOS battery. Really need to tidy up the soldering before posting any pics as it's all a bit "mocked up" at the minute to make sure it would work - seems to be a perfectly adequate solution so will tidy it up a little. Still waiting on RAM - fingers crossed it hasn't got lost or anything.

- P-MMX 200MHZ, PCChips M598LMR, Voodoo
- P-MMX 233MHz, FIC PA2013, S3 ViRGE + Voodoo
- PII 400MHz, MSI MS6119, ATI Rage Pro Turbo + Voodoo2 SLI
- PIII 1400MHz, ECS P6IPAT, Voodoo5 5500
- Toshiba Libretto 110CT, 300MHz, 96MB RAM