feipoa wrote:

For synchronous operation, the PCI clock typically runs at 41.7 MHz (when the FSB is set to 83.3 MHz). Wasn't 66 MHz operation added in PCI revision 2.1?

Thats a very good point. Although I'm not sure if ultimately the PCI bus, to support simultaneous 33MHz and 66MHz modes, incorperated an onboard clock multiplyer (akin to a DX2) or not... Also, even if PCI 2.1 defines 66MHz operation, does that mean all PCI 2.1 compliant cards run at 66MHz or 33MHz to suit their needs, or adapt to whatever speed needed to be the most compatible. I would imagine though that if a PCI 2.1 card runs stock at 66MHz, attaching it to a Syncronus Bus running at 83.3 might overclock that card to 83.3 unintentionally. Plus, what would that do to the ISA bus if it's PCI CLK/4? Honestly, I can't find much information online about the subject, seems that most of these younger kids today have moved onto some new fangled thing called PCIe. 😜

If that causes an issue though, i don't know. I would imagine that most enthusiast-grade cards that were made in the day when Cyrix CPUs with 83MHz Buses were king probably have some kind of failsafe mode or special tolerances for safe operation with those systems.

That said, if what you say is true and it would work, I might try and put a K6-2 (not +) 500 in my TX97-E and set it to 6 x 83, benchmark it and see if she's stable.

maddmaxstar wrote:

That said, if what you say is true and it would work, I might try and put a K6-2 (not +) 500 in my TX97-E and set it to 6 x 83, benchmark it and see if she's stable.

Expect trouble with the onboard IDE controller at 83 MHz FSB clock. Also, many NICs stop working properly at PCI speeds above 40 MHz.

There seems to be some confusion. I am refering to the PCI expansion cards themselves supporting 66 MHz operation, not the motherboard. While wiki mentions 66 MHz support being added in PCI 2.1, I cannot say for certain if all PCI 2.1-era expansion cards support 66 MHz operation. I am pretty sure that all PCI 2.2 expansion cards must be able to work at, both, 33 and 66 MHz.

If you are using an 83 MHz FSB, your PCI bus is generally run at 42 MHz, not 83 MHz.

I am pretty sure the ISA bus uses a seperate, or fixed, clock frequency. I'll measure it next time I have my socket 7 on the bench. I personally had no issue running a Cyrix MII with an 83 MHz FSB. I had a Matrox G200 PCI and an AWE64Gold ISA in the system.

My AMD K6-III+ had no issue running at 6 x 83 MHz in my TX socket 7 board with modified BIOS.

5u3 pointed out onboard IDE issues at 83 MHz. My tests were done using a PCI SCSI card, so I cannot comment on this issue. In terms of the network cards which run at 42 MHz, you'll have to play around with this. I would suspect that PCI 2.2 cards should work. I seem to recall some 3Com 10/100 network cards being PCI 2.2 compliant. An Intel PRO/100 S may also be 2.2 compliant, but I haven't tested these yet in my socket 7.

I'll try it and post my results. I have a 3Com 3C905-TX in the TX97-E system, so that should be fine. It's the Voodoo2 (And the accompanying Trident 9440) I'm worried about. Plus if theres issues with the onboard IDE controller, there may be issues there as well, though I believe the Motherboard manual and documentation did clearly advertize 83MHz mode.

Gah! I just popped the cover off of the K6-2/400, turns out it's not a an Asus TX97-E board, but a TXP4. It's still running a hacked BIOS I put on it long ago to support the 6x multiplyer, but the FSB only goes up as far as 75MHz. So far it seems to be running fine at 450 (6x75) using an udnerclocked K6-2/550 chip, but I wanted to run it at 83. It might yet have that mode, I'm gonna have to read into it further.

You can systematically try all FSB jumper combinations in search of the 83 MHz setting.

feipoa wrote:

You can systematically try all FSB jumper combinations in search of the 83 MHz setting.

Yup, finding undocumented jumper settings is fun! Especially on a 486 board 😁

But still, how can one actually use 83Mhz FSB if it's so unstable?

My initial tests at 83 MHz were stable (on the TX board). Sometime between now and the next 10 years, I'll so some long-term stability testing at 83 MHz with a Cyrix.

I'm still trying to source a suitable candidate to test out my UMC board at 83 MHz. I'd need a CPU which recognised the 2X multiplier setting as 2X and could run at 166 MHz (83x2). From the other post, these AMD DX4-100-WB16 chips might be good candidates for test, but I don't have any. Perhaps an AMD DX4-120 would suffice?

I found the hidden jumper combo on the Asus TXP4 for the 83MHz FSB!

For future reference, the jumper settings are:
FS0: 2-3
FS1: 1-2
FS2: 1-2

It appears to be up and running and stable, though I haven't done anything with it. I'll put the Voodoo2 (8mb STB) back in and do some tests, try some games, etc.

As for UMC boards, 83MHz... 486... do you mean a UMC 486PCI board? I have two UMC boards, though my DX4-120 isn't a 16k Cache version.

Yes, the UMC 486 boards seem to have an 83 MHz setting. I discovered it with my osciliscope. The signal swing isn't very large though.

Are you sure the AMD DX4-120 runs at 4X on the 2X setting? I'd be really surprised if that was the case. I'm counting on the AMD DX4-120 actually recognising 2X as 2X for 2x83 operation.

I'll pull apart that system and try tinkering with the settings. I'm kinda curious at seeing how the thing fares at running at 4x33 anyway.

I just tested my AMD DX4-120 using the 2x/4x setting, and it is definately only recognising 2X, not 4X.

I'll really have to check it now. I'm quite sure that I set the 2/4x jumped and she posted at 160, it was only a month ago. I'll check it later today and post a picture.

What's the model number off your Am486?

I have three of these chips with slightly different date codes, but the particular piece I tested was labeled as:

Am486 DX4-120
A80486DX4-120SV8B
C 9541GPE
3 VOLT
25398 N

Here's a closeup of my DX4-120.
Date code shows 9th week of 96.

And here's what I got when I Set the bus to 33MHz and enabled the 2x/4x jumper:

System appears to be running stable at these specs, I may just leave it this way. The motherboard is an Arstoria AS496, some documentation and manual scans can be found here: http://motherboards.mbarron.net/models/486pci/as496.htm . The documentation has it clearly labeled that the 2x jumper doubles as the 4x jumper. It's weird that the BIOS or Chip recognized it as a 4x multiplyer and is running sound at 133. I may try it in my other 486PCI system to see if the same happens.

The weird part is that PC-Config Identified the chip as having 16k L1 Cache, but I have doubts that it's actually 16k instead of 8k. Is there an alternative DOS or Win9x based utility that I can measure the L1 Cache with?

SpeedSys also identified that you have 16 KB of WB cache. You have what appears to be an AMD X5-133 labeled as an AMD DX4-120. The CPUID is one in which I have not seen before: 04E4.

Your L1/L2/Memory timings in SpeedSys all seem really slow. Please refer to the Ultimate 486 Benchmark Comparison, Appendix 1, Column J for what an optimised system running an AMD X5-133 would be running at. Refer to Column K for an AMD DX4-120.

I was not aware that AMD would relabel something as a 120SV8B, the 8 here refering to 8 KB cache, when it really had 16 KB. It seems so unlike AMD.

One possible slowdown issue is that I had to disable the Writeback jumper on the motherboard a while back as it wasn't running stable at normal speeds. I might have tuned down the memory timings as well during troubleshooting and didn't set them back. I'll check.

I tweaked the Memory settings in the BIOS to the following below:

And reran SpeedSys:

The speeds improved slightly, but still slow. As I mentioned earlier, the performance might be comprimised due to the machine flaking out when the Writeback jumper is enabled(although it's set to enabled in the BIOS). When I reenable the setting on the board, I get this:

It freezes just after the POST screen.

To be sure the Cache being reported by PC-Config and Speedsys 4.78 are correct, I changed the jumpers and set the board to 3x33, it booted as an Enhanced DX4-100, and still shows as having 16k L1.

And to rule out it being any kind of weird screwy motherboard BIOS misreporting, I tried it in my other 486, which runs a Lucky Star LS486E Rev C1. Again, when set to run on a 66MHz FSB with Enhanced features enabled, it still shows that it's running 16k L1, plus with Writeback properly enabled and running OK, the speeds are much closer to normal:

And the BIOS settings for the Lucky Star are set to pretty well the same as the Arstoria AS496 board.

I played a few games on it as well, and it seems to be quite stable.

Very strange. I wonder if the chip is mislabeled. In any case, it seems to be running just fine as a 5x86-P75, so I may just leave it this way.

Any reason why you have CPU Burst Write disabled? Is it not stable?

What voltage are you running the CPU at? You may have an X5-133 relabelled as a DX4-120 because it failed characterisation at 133 MHz. You might want to run it at 3.6 V if your motherboard has a setting for 3.6 V. The CPUID is odd to me. I'd hold onto this CPU as "evidence".

I'm not sure why that first motherboard is so slow. In my 486 testing experience, I have definately concluded that some board inherently have poor throughput, irregardless of what setting you input into the BIOS.

I think the board has a setting for 3.45v, which I believe is what it's set at. And I didn't know what the CPU Burst Write was, just assumed it was a Cyrix thing. I enabled it, seems stable, no performance increase, same test results.

I might try swapping Cache and TAG chips between boards to see if that makes a difference. If not, I might just swap parts around so the faster board is in my main 486. With both boards using UMC496/497 chipsets, I shouldn't need to reinstall Win95.