It's not really worth putting a Obsidian2 in there. Even a Voodoo1 will be starved by a 486. They really just do not have the oomph to do 3D of that caliber.

A Pentium Overdrive would be as best as can be, because it has a much better FPU, but even it would still be super constricted by the 32-bit RAM bus. I have one of those that will do 100 MHz and it could barely pull 15-20 fps in Jedi Knight with a Voodoo3! I had to run through a bunch of drivers before the Voodoo3 would work with a Am5x86, btw, because the drivers must use Pentium-specific instructions in later revisions. Not sure if that would affect Voodoo2.

As for that chipset, well here's a site I found after a quick bit of Google Groups searching. Info about EDO and lots of other stuff. The mobo uses the UM8881 chipset.
http://web.archive.org/web/19981202033238/htt … om/sdnd/vip.htm

I think you must be the only person who ever liked POD83. From what I hear even when it is overclocked to 100MHz the am5x86 and cx5x86 have no problem knocking its ass around the block for integer performance. Though, POD83 supposedly has the best FPU of them all it would seem that not too many 486 games would use it. One of the other huge downsides to the POD83 is that it doesn't work properly in a majority of the motherboards designed for it. In the ones it can work well, it will always force the L2 cache to be turned off. It has a large 32k L1, but it seems it isn't enough to compensate for the loss. I'll likely stick with the am5x86. Though, I hear that the 0.35u AMD dx/2s have a good shot at reaching 120 or 133MHz in clock doubled mode, so I'm going to give it a shot.

I also completely agree about the Obsidian2 being overkill for a 486. But, I don't have any other PCI systems to put it in, so it just goes in because it's available. I'd like to be able to use it on GTA if it is possible to patch it for Voodoo2. I also have a 4MB Voodoo1 if that is more practical.

Oh I know that POD83 is a pain, but it is still a very interesting little CPU. It actually has the same amount of L1 cache as a Pentium MMX.

I had a go with Am5x86 160, Cx5x86 120, and POD83 @ 100 a couple of years ago just for kicks. In FPU-heavy games (I tested Jedi Knight and Quake), POD83 is clearly superior to the 486s. Even at 83 MHz. I also ran some pure synthetic FPU benches with Sandra and Everest and POD83 just outclasses even a 160 MHz Am5x86. The Pentium FPU is a lot better than a 486's and some games definitely do take advantage of it (Quake is basically built around it).

However, Am5x86 actually made the snappiest system for all other tasks. It was basically even with POD83 @ 100 for integer tests but Windows simply felt quicker. POD83 seems to not work properly with the L2 cache on the board I have (and probably others) even though it is officially supported. I suspect this is why the performance fell behind in everything other than FPU-heavy apps. (late model MSI 4144 486 PCI board).

Most pre-3D era DOS games are primarily integer arithmetic-based so they perform very well on Am5x86 and Cx5x86. So a Pentium Overdrive in a 486 is sort of pointless because its better architecture still can't make 3D games perform on the 486 platform and its compatibility sucks. Integer performance isn't superior to the Am5x86 @ 160. And wow was it ever priced sky high back in its time! A rip-off just like every other Intel Overdrive.

I think it's pretty interesting. I found a post you made in 2004 or 2005 in which you were really complaining about POD83, and how it is the worst upgrade chip ever. I really wonder what happened to make you fall in love with such a terrible device. 😎

Oh Ok.

Well I just recapped the thread I'm sure you found at Cpumuseum.de and told you right at the end that I think it was a waste of money in its time. Hell, it was the biggest rip off CPU in consumer history at, I believe, around $500. Good thing I got it free. It is indisputable though that it is the best FPU you can put in a Socket 3.

How can a Pentium reworked for a 486 bus and with double the L1 cache not be interesting? Did you know that it clocks itself down if its little integrated fan dies, as a means of thermal protection? This is ~1995 we're talking about. AMD couldn't hack something like that together until Opteron!

If you want to rip on Intel, you ought to look at Cyrix too. Their 5x86 speed claims relied on a number of features that are disabled by default in the CPU because most BIOSs don't support them. And actually the touted branch prediction feature doesn't work right at all on early revisions. Buggy CPU there.

Am5x86 is by far the best performing and most reliable 486 option, outside of raw FPU speed. Good thing most DOS games don't really use the FPU because half of the 486 market didn't have one.

After doing some extensive reading, I have come to the conclusion there is no such thing as PCI 2.1 support on 486 system boards. There are four main late generation PCI chipsets for 486 systems:

-Intel Aries & Saturn
-SIS497
-ALi M1487
-UMC UM8881

I have read some documentation on all of them, and each one is just PCI 2.0 compliant. The ALi and the later revisions of the UMC support EDO DRAM, and I have seen people claim the SiS does too, but the one on my Asus PVI SP3 does not. Perhaps it was also included in a later revision. From what I understand, the key to a good PCI 486 motherboard is to make sure that the PCI bus doesn't hang off the VL bus. Supposedly this limits the speed to 8mb/sec and causes incompatibility with busmastering cards, so it's best to avoid boards that have both VLB and PCI slots.

I think what is rather disappointing about these chipsets is that they all seem to have silly caching restrictions. I have heard that the ALi is limited to caching just 32mb, and the UMC just 64mb (possibly more with writeback disabled). If you have a board with 512k or 1024k cache ram you should be able to cache 128mb and 256mb respectively, so this is just a load of crap.

Overall, I believe UM8881F is the best choice as it has 60 and possibly 66MHz bus option with divide by 2 for PCI. I think no other chipset supports this option. I think it would interesting to try running both Cyrix 5x86 and 0.35u DX/2 at 2x66MHz with PCI cards in spec.

Another fifth possible chipset is one made by "ALD". I don't know much about it because I can't find any documentation. It looks like it supports pipelined burst L2 cache, and possibly has PCI 2.1 support since it was likely designed for MediaGX. Though, I am affraid to try ALD because I feel they are a small company and there are probably many bugs in their product.

Finally, I think booting from Firewire is definetly out of the question. Even many modern systems don't seem to do this. But, can anyone recommend a good Firewire/USB 1.1 adapter that works in PCI 2.0 slots?

Just remember that RAM was stupidly expensive back then. 32MB in a 486 is a lot of RAM. 😀 I've no doubt that they gimped these chipsets a bit, considering that even some Pentium chipsets have limited caching (430FX/MX/VX/TX and others I'm sure).

I have a MSI 4144 mobo that's based on SiS497/496. It has 256K cache. I believe it caches 32MB. I've used a Promise Ultra66 card on it without trouble, so it seems busmastering isn't a problem with it. I believe I benched the PCI bus at around 20 MB/s.... Remember too here though that a 486 mobo's RAM bandwidth is pathetic. Probably well under 100MB/s practical bandwidth. The cards can't DMA faster than the RAM. 😎 I've had a Voodoo3, Verite, SoundBlaster Live!, and Philips Seismic Edge in my board. PCI 2.1 support may not be as critical as you think.

I've managed to boot CD-ROMs on 486s by using a boot floppy with a tiny freeware app on it called Smart Boot Manager, btw. I doubt it can do anything with Firewire though.

Wow, a 486 PCI thread, very interesting 😉

I have some experience with Intel Saturn and SIS 496/497:

Intel Saturn/Aries:
Fast PCI performance (up to ~60 MB/s), but make sure you get the bugfixed chipset revision (rev. 04 for Saturn, rev. 02 for Aries), older ones have trouble with busmastering and address decoding.
Saturn has memory interleaving and must be equipped with two PS/2 SIMM modules, Aries can perform the same trick, but also works (slower) with just one module installed.
The downside: FSB limited to 33 MHz, timing problems with AMD CPUs, no support for Cyrix features (what do you expect, its an Intel chipset 🤣).

SIS 496/497:
I'm still not sure whether this is a genuine PCI host design or just has a PCI bridge on a VLB system. Most info on the web and in old magazines says it's the former, but the lame PCI performance and the integrated CMD640 VLB IDE controller suggest the latter.
This is a very versatile chipset. Worked with all PCI 2.0 and VLB cards I've tried so far. However, it's not suited for overclocking experiments (no PCI divider), anything above 40MHz FSB crashes the system if you have PCI cards installed. The L2 cacheable area is 32 MB (WB) / 64 MB (WT).
The on-chip CMD640 IDE controller is buggy on all chipset revisions older than B4 (SiS 496NV+497NU).

swaaye wrote:

Remember too here though that a 486 mobo's RAM bandwidth is pathetic. Probably well under 100MB/s practical bandwidth.

Correct. Common memory throughput rates for late 486 boards:
Main memory: ~ 7-11 MB/s
L2 cache: ~ 25-40 MB/s

Anonymous Coward wrote:

But, can anyone recommend a good Firewire/USB 1.1 adapter that works in PCI 2.0 slots?

Apart from the PCI compatibility issue, these won't work under DOS/Win3.11.

Anonymous Coward wrote:

PCI slot 3: AHA2940U2W (needs PCI 2.1)

Don't know about this one, but the plain AHA 2940 should work in 486 boards, same with the NCR 810.

I am pretty sure that my Aires board is one of the buggy ones. It was relatively stable, but it just felt really slow overall. My Saturn board (I think Saturn II) may be one of the better ones. Actually, I never got around to testing it, butit looks nice though since it has on-board SCSI. I believe the Saturn was supposed to be superior to the Aires in a few ways since they were both available during the same time frame. It's just a shame that neither of these boards have any goodies to play with. Maybe a POD-83 would like one of them? I find my SiS497 (PVI SP3) board to be a little better than the Intel boards overall since they have better CPU support and a newer BIOS, but the PCI bus on this board really stinks. It doesn't feel as slow as my Aires, but the card compatibility is awful (especially with network and SCSI adapters).

For me the caching limitations on late 486 board and many Pentium boards is really insulting. I have a lot of older 486 boards that can cache 64meg, 128meg and even 256megs of RAM. Yes, anything over 32 megabytes is overkill for a 486, but if your board supports more than 128kb cache, it had better damn well cache more than 32megs of RAM or else it's just there as a decoration. Why would you have 1MB cache RAM if you can only cache 32mb? Don't even get me started on the Intel TX and i815 chipsets.

Getting back to my UM8881F board...

if the 2940U2W won't work I have a 2930U to try out. Do you think an ATA66 controller would be more practical? I have a feeling that I won't be hitting the 20mb/sec cap anyway. I haven't really had good luck with promise controllers on later systems, and I also think the CPU overhead on SCSI is likely better anyway (not to mention external device support).

I'm also not 100% about USB and Firewire, but I have heard there are actually some things you can do with them in DOS. I have seen drivers for both that support USB sticks and removable media. If I could at least get a USB mouse going I'd be pretty happy. It's not easy getting the optical mouse working over a serial port.

Also, where do you get your numbers for 486 memory bandwidth? I have quite a lot of 486 motherboards, and the bandwidth can vary greatly. I have been using WinTune 2.0 for testing, but it may not be a very accurate tool.

-opti495SX w/ti486SXL 40MHz reports ~14mb/sec
-opti EISA VL am5x86 160MHz reports ~18mb/sec
-SiS EISA VL am5x86 160MHz reports ~25mb/sec

I was under the impression that the average transfer rate for decent 486VL system was about 17mb/sec. Aren't the 486 PCI boards supposed to have better performance, especially with EDO RAM installed?

Anonymous Coward wrote:

For me the caching limitations on late 486 board and many Pentium boards is really insulting. I have a lot of older 486 boards that can cache 64meg, 128meg and even 256megs of RAM. Yes, anything over 32 megabytes is overkill for a 486, but if your board supports more than 128kb cache, it had better damn well cache more than 32megs of RAM or else it's just there as a decoration. Why would you have 1MB cache RAM if you can only cache 32mb? Don't even get me started on the Intel TX and i815 chipsets.

Well, we're talking about consumer hardware here. True, there were some nice EISA workstation boards with fewer limitations before, but when the first 486 PCI boards came out, the professional market had already moved on. Besides, people who had the money for 1 MB SRAM and 256 MB DRAM in 1995 bought workstations based on other platforms, not this weedy x86 stuff... 😉

Anonymous Coward wrote:

Do you think an ATA66 controller would be more practical? I have a feeling that I won't be hitting the 20mb/sec cap anyway. I haven't really had good luck with promise controllers on later systems, and I also think the CPU overhead on SCSI is likely better anyway (not to mention external device support).

Yeah, chances are better with a SCSI controller, because the onboard EIDE solutions back then were downright awful, and I don't think modern PCI ATA controllers will work on these boards.

Anonymous Coward wrote:

Also, where do you get your numbers for 486 memory bandwidth? I have quite a lot of 486 motherboards, and the bandwidth can vary greatly. I have been using WinTune 2.0 for testing, but it may not be a very accurate tool.

I tried to find a reliable download link for WinTune 2.0, but didn't find any. However, after reading a description of this software on the net, I think I'll pass on this one.

The tool I use for measuring memory/bus bandwidth is called CTCM, written by C't magazine editor Andreas Stiller. It is a DOS program with very detailed output (documentation unfortunately only available in German).

Anonymous Coward wrote:

Also, where do you get your numbers for 486 memory bandwidth? I have quite a lot of 486 motherboards, and the bandwidth can vary […]
Show full quote

Also, where do you get your numbers for 486 memory bandwidth? I have quite a lot of 486 motherboards, and the bandwidth can vary greatly. I have been using WinTune 2.0 for testing, but it may not be a very accurate tool.

-opti495SX w/ti486SXL 40MHz reports ~14mb/sec
-opti EISA VL am5x86 160MHz reports ~18mb/sec
-SiS EISA VL am5x86 160MHz reports ~25mb/sec

I was under the impression that the average transfer rate for decent 486VL system was about 17mb/sec. Aren't the 486 PCI boards supposed to have better performance, especially with EDO RAM installed?

You have a 33MHz 32-bit system bus, which gives 132MB/s bandwidth theoretically (ooh, just like PCI 🤣). The system bus is shared between all the devices, the cache, and the RAM. I'm also sure these boards weren't very efficient in general, and that even EDO has some ugly latencies. So, I don't think it's surprising that these systems have very poor practical bandwidth.

They certainly can vary though. I've seen between 10-40MB/s in cachechk posts on Usenet. PCI is obviously never going to show all it can do on these boards no matter what.

Anonymous Coward wrote:

Do you think an ATA66 controller would be more practical?

I'm also not 100% about USB and Firewire, but I have heard there are actually some things you can do with them in DOS.

I think a modern UltraATA card will do just as well as a SCSI card from those days. Even UltraDMA/33 offers more bandwidth than the system can ever consume. 😀 I just remembered though that I do not think busmastered DMA was working with the Ultra66 because when I did a HDTach test, CPU usage was 100%. It was still quite fast though. I have almost zero SCSI experience, strangely enough, so I couldn't tell you which to go with. If you have both, try both.

DOS and USB don't really mix outside a select few devices like keyboards and mice. And those only work if you have BIOS emulation for them, I believe.