gonzo wrote on 2023-02-22, 09:42:

mkarcher wrote on 2023-02-21, 21:34:

The early revision of the 8881 north bridge doesn't support linear bursts.

Good work, mkarcher.
Would you please tell us, which board-revision of the HOT-433 exactly do you use/mean?

The board the current experiments are performed on is a HOT-433 rev.1. The important revision is not that of the mainboard PCB, but the revision of the UMC8881 chip. Information wanted for UMC UM8881F sub-revisions discusses the two revisions that are commonly available: "rev. 01" and "rev. 04" according to the PCI device identification registers. You can also determine the chip revision by looking at the chip: If the first letter after the four-digit date code is "B", you have a "rev. 01" chip, and if that letter is "E", you have a "rev. 04" chip. Only the rev. 04 ("E") chip supports linear bursts.

feipoa wrote on 2023-02-22, 11:07:

mkarcher, as far as I can recall, you should be able to use the Cyrix 5x86 at 120 MHz using fastest timings (2-1-1-1 sram, 0ws/0ws dram) on the MB-8433UUD provided you use 256K double-banked cache. If you are testing in Windows, I assume you are using a Stepping 1, Revision 3 CPU rated at 100 MHz. Sometimes when overclocking the Cx5x86, some of the fancy features, like BTB, FP_FAST, DTE, LSSER, etc may not be fully stable at 120 MHz.

The HOT-433 runs at 2-1-1-1 sram, 0ws/0ws dram for everything except the Cyrix 5x86 at FSB33 and FSB 40 without problem (dual-bank cache, of course). The Cyrix 5x86 refuses to even boot Windows 2000 ("ntoskrnl.exe is missing or damaged") at those timings, but works fine at 2-1-1-1 sram, 1ws/0ws dram. This applies both to FSB33 and FSB40. Possibly the Cyrix 5x86 at 0ws/0ws 120MHz + linear bursts is able to beat the AMD 5x86 at 160MHz in 3DMark 2000.

I got a crash like 3 minutes after power-on at 120MHz with the heatsink at around 45 degrees (estimated from touching it) when I didn't yet have a fan mounted, but it seems to be rock solid with the fan. Obviously, the thermal margin at 120MHz is quite low.

mkarcher wrote on 2023-02-22, 17:49:

feipoa wrote on 2023-02-22, 11:07:

mkarcher, as far as I can recall, you should be able to use the Cyrix 5x86 at 120 MHz using fastest timings (2-1-1-1 sram, 0ws/0ws dram) on the MB-8433UUD provided you use 256K double-banked cache. If you are testing in Windows, I assume you are using a Stepping 1, Revision 3 CPU rated at 100 MHz. Sometimes when overclocking the Cx5x86, some of the fancy features, like BTB, FP_FAST, DTE, LSSER, etc may not be fully stable at 120 MHz.

The HOT-433 runs at 2-1-1-1 sram, 0ws/0ws dram for everything except the Cyrix 5x86 at FSB33 and FSB 40 without problem (dual-bank cache, of course). The Cyrix 5x86 refuses to even boot Windows 2000 ("ntoskrnl.exe is missing or damaged") at those timings, but works fine at 2-1-1-1 sram, 1ws/0ws dram. This applies both to FSB33 and FSB40. Possibly the Cyrix 5x86 at 0ws/0ws 120MHz + linear bursts is able to beat the AMD 5x86 at 160MHz in 3DMark 2000.

I got a crash like 3 minutes after power-on at 120MHz with the heatsink at around 45 degrees (estimated from touching it) when I didn't yet have a fan mounted, but it seems to be rock solid with the fan. Obviously, the thermal margin at 120MHz is quite low.

Little story to add to yours, Sometimes CPU are like that, on the edge.

Back in the day, 1993 to be precise, I was given a brand new Am486DX 40 CPU by a guy who owned a computer shop that I helped him out from time to time as a send off gift before going to college.

Well, Got that little VLB computer built at last, with meager money while in college. Knowing intel 486DX 33 was capable of running without heatsink. Not to be as expected, kept crashing in normal use even at 25MHz. Finally picked up a cooler from computer show fixed that. But that little 486 can do 50MHz not just 40MHz and was a wonderful FPS machine like doom games etc. Eventually bought 2MB hercules dymanite and that was a bliss. 😀

Cheers,

gonzo, from my experience, the the Diamond branded TNT worked a bit better than the Creative TNT on the 486's I tested w.r.t. accelerated 3D. My Creative TNT also has rather noisy output, something I haven't got around to fixing. Does your Creative TNT have noisy output? Maybe they all were like that?

mkarcher, hopefully you can test your Cyrix SrR3 at 120 MHz with your MB-8433UUD. I'm pretty sure I was able to run mine w/W2K at 2-1-1-1, 0ws/0ws. The ET486NT driver works in NT4 and W2K just the same, just mind that I think the units are in DEC.

noshutdown wrote on 2023-02-17, 05:37:

mkarcher wrote on 2023-02-15, 19:24:

To disable the MMX check at the start of 3DMark 2000, you need to patch 3DMark2000.exe . The byte at offset 69626 (hex) is originally 75 (hex) (in the context at that point, this byte instructs the processor to jump if MMX is present) to EB (hex) (which instructs the processor to jump unconditionally. This one is easy.

thanks for the info, thats very interesting.

i also wonder if anyone has tried rendition cards(1000 and 2100) on 486?

I am just coming around to thinking about that, got a Diamond Stealth II kicking around and reading about RV unloads the CPU as well as having reasonably lightweight drivers, thinking it might be a better match for my Cyrix 5x86 than other cards. But gotta get my 486 PCI boards dug out and fixed up. ... also to make certain I actually have 2 Cyrix 5x86

BitWrangler wrote on 2023-02-22, 22:03:

I am just coming around to thinking about that, got a Diamond Stealth II kicking around and reading about RV unloads the CPU as well as having reasonably lightweight drivers, thinking it might be a better match for my Cyrix 5x86 than other cards. But gotta get my 486 PCI boards dug out and fixed up. ... also to make certain I actually have 2 Cyrix 5x86

note that amd5x86, cyrix5x86 and pod are of different compatibility level
amd5x86 is 100% 486
pod is almost 100% pentium compatible
cyrix5x86 and even 6x86 are 586 class, but not 100% pentium compatible
so, pod is most likely to run 3d cards and benchmarks which could run on a complete pentium machine, while amd5x86(a real 486) may fail.
the cyrix5x86 may run in some cases, but has also been reported to fail certain tests that could run on amd5x86. my guess is that it used p5 features which is incompatible.

feipoa wrote on 2023-02-22, 20:35:

gonzo, from my experience, the the Diamond branded TNT worked a bit better than the Creative TNT on the 486's I tested w.r.t. accelerated 3D. My Creative TNT also has rather noisy output, something I haven't got around to fixing. Does your Creative TNT have noisy output? Maybe they all were like that?

I do not have a TNT from Creative, or from Diamond (but I will try to find a Diamond in the future).
My TNT is an Elsa Erazor II, 16 MB.
On the other hand, my Voodoo Banshee is from Creative (3D Blaster Banshee PCI, CT6760) - this VGA works perfectly with the board and the FSB of 60 MHz.
At this time, they are the only VGAs of those models (TNT; Banshee) I have.

With the Banshee and the Voodoo 3, the picture is absolutely clean and clear, they are no noise-problems.

Maybe this thread is the wrong place for this question...

To spare time for the testing of alternative drivers in Windows 9x and newer, they is good to have a backup of the fresh (or still good working) Windows, in case of a later problem.

On my 486-systems, I often have problems to make a backup of "C:" using Northon Ghost or Acronis True Image in DOS-mode.

They are older versions of Acronis like 6.0 and 8.0 using boot diskettes (about 6 to 8 pieces) for DOS. Sadly, this software often freezes the PC or crashes after complete loading of all diskettes and starting itself.

About Norton Ghost, I use the versions 2001 or 2003, as they use only one boot diskette and sometimes they recognize older CD/DVD-burners, too (e.g. LG GSA-4xxx or -8xxx).

The main problem in Northon Ghost I sometimes have:
trying to verify an image just created, Ghost reports an "internal inconsistency" and goes back to DOS. The same happens if I try to restore an existing backup directly (without verifying it).
This problem occurs even if the Windows-HDD is dismounted from the 486-system and connected to another, much more actual/modern system (to avoid possible problems of the 486-system itself, like missing drivers etc.).

How this Ghost-problem can be solved?

Wich backup-software can be used alternatively?

Thanks.

gonzo wrote on 2023-02-23, 10:03:

Maybe this thread is the wrong place for this question...

How this Ghost-problem can be solved?

Sure it is off topic, but may I ask you which compression level you use?
I use fast compression because I noted inconsistencies when using high compression.

Disruptor wrote on 2023-02-23, 12:55:

gonzo wrote on 2023-02-23, 10:03:

Maybe this thread is the wrong place for this question...

How this Ghost-problem can be solved?

Sure it is off topic, but may I ask you which compression level you use?
I use fast compression because I noted inconsistencies when using high compression.

I use different compression levels, from uncompressed to high compressed.
Most common, I use fast/average compressed.
To be honest, until now I didn't pay attention if the compression does affect this problem, or not.
BTW, the problem occurs even by the use of a CD/DVD (not only HDD) for the backup.

mkarcher wrote on 2023-02-22, 17:42:

gonzo wrote on 2023-02-22, 09:42:

mkarcher wrote on 2023-02-21, 21:34:

The early revision of the 8881 north bridge doesn't support linear bursts.

Good work, mkarcher.
Would you please tell us, which board-revision of the HOT-433 exactly do you use/mean?

The board the current experiments are performed on is a HOT-433 rev.1. The important revision is not that of the mainboard PCB, but the revision of the UMC8881 chip. Information wanted for UMC UM8881F sub-revisions discusses the two revisions that are commonly available: "rev. 01" and "rev. 04" according to the PCI device identification registers. You can also determine the chip revision by looking at the chip: If the first letter after the four-digit date code is "B", you have a "rev. 01" chip, and if that letter is "E", you have a "rev. 04" chip. Only the rev. 04 ("E") chip supports linear bursts.

HOT-433 Revision 04 (E) accepts EDO-RAM, too. Right now I am testing this model (same AMD DX5 @ 180 MHz/4V/FSB 60 from above), but not ready with testings yet.
This CPU and a GF 2 MX 400 show until now much lower results in 3DMARK99MAX compared to the boards above with SiS-chipset and Voodoo-VGAs (above 100 fewer points for CPU and VGA).
GLQuake still does not start, as discribed some weeks ago...

gonzo wrote on 2023-02-26, 09:07:

HOT-433 Revision 04 (E) accepts EDO-RAM, too. Right now I am testing this model (same AMD DX5 @ 180 MHz/4V/FSB 60 from above), but not ready with testings yet.
This CPU and a GF 2 MX 400 show until now much lower results in 3DMARK99MAX compared to the boards above with SiS-chipset and Voodoo-VGAs (above 100 fewer points for CPU and VGA).
GLQuake still does not start, as discribed some weeks ago...

Yes, it accepts EDOs. But. Does it benefit from EDOs too?
But it does not fit in my overkill 486 that is populated with 256 MB of DRAM, which needs 1024 KB of L2 cache.

Does your GF2MX have a DVI output?

pshipkov wrote on 2023-02-06, 19:27:

I run GLQuake1 in full screen mode.

@ pshipkov + feipoa:

on the HOT-433 with a GeForce 2 MX 400, there is a just insane result for GLQuake, 800x600, window-mode, of 38,0 FPS!

CPU: AMD 486 180 MHz
FSB: 60 MHz
L2: 1 MB (15 ns), WB-mode
RAM: 128 MB EDO (2x 64, 60 ns)
OS: Win98SE

The test is really very fast (much faster than using the Voodoo 3 on the ZIDA), the picture is perfect, so it can not be a fake-measurement.

My question is now: can this result be compared to yours? 😀

On the Voodoo3-ZIDA-system, I had to choose the window-mode, too, BUT the test starts in full-screen-mode, anyway...
On the GF2-HOT-433-system, until now, I am not able to start at full-screen-mode, too, but the window-mode is a "true" window-mode, as it can be seen in the pictures.

Is it maybe possible, that at the same resolution (e.g. 800x600), the window-mode results in a better score than the full-screen-mode?

BTW, at this time I am using detonator 6.31 - with this one, at least for me, 3DMARK99MAX can not start (it freezes the system).

BTW, the BIOS-settings for the RAM, and the RAM-modules, are maybe still not optimized at 100%.

@ Disruptor: sadly, my GF2 does not have a DVI-connector

thanks for tagging me on your message.
very interesting stuff.
that is windows 95 ocr2 i guess ?
how stable is the system overall ?
asking this because quake is not a great indicator for stability in general.

i am in the 386 zone atm, but will spend time after that to look at Zida 4DPS and hopefully figure out how to make it scale better like you did.
same for drilling into gf2 mx 400.
i remember trying this card before but it was not fully stable.
you will be hearing from me on these things soon.

how are the rest of the usual tests used around here with th gf2 card - doom, q1, etc ?

These are incredible results for a GF2 on a 486 (Am5x86-180)!

GLQUAKE 800x600 = 38.0 fps. What about 1024x768 and 1280x1024?

I think the Voodoo3 won't work on this board, so could you cross compare the results with a Voodoo2?

If you look at page 1 of this thread, you will see that I used Detonator v12.41 w/GF2 and Am5x86. I tried all Detonator revisions and landed on 12.41 for this graphics card and CPU. However, at 160 MHz, I'm only reporting 26.7 fps.

Are you also able to cross compare your HOT-433 results with the MB-8433UUD and list the memory, cache, and PCI timings/speeds you are using on both systems?

Were you able to get any Direct3D games working with your GF2 and Am5x86 in Win9x? If so, which games, which DirectX version, and which Detonator driver version?

Your results make me want to pull out my HOT-433 again, but to do so would be to not finish other projects.

feipoa wrote on 2023-02-27, 02:16:

These are incredible results for a GF2 on a 486 (Am5x86-180)!
Your results make me want to pull out my HOT-433 again, but to do so would be to not finish other projects.

Yes, feipoa, that is my problem, too 😀
If I understand you and pshipkov right, the scores of GLQuake are comparable each other at the same resolution, regardless of "window" or "full screen" mode (or, in case of the Voodoo 3, "windowed fullscreen" mode...).

At the moment, I am still trying alternative drivers for the GF2, so this could take some time for finishing.
After that, I will do all "usual" DOS- and Windows-tests to complete the benchmarks, as well I will try some fine-tunings in the BIOS.

A Voodoo 3 does not boot with this board.
About a Voodoo 2: to be honest, I am not sure if my Voodoo 2 works properly, but I will try it, too.

About other retro-games: I do not have (and do not use, to be honest) much of them, except of Quake I+II and Doom, and of course 3DMARK.

BTW, my HOT-system is still very "empty" (still no audio card installed etc.), maybe that's the reason it's very fast 😀 We will see...

BTW, I do not have a 8433UUD-board.

pshipkov wrote on 2023-02-26, 22:52:

thanks for tagging me on your message. very interesting stuff. that is windows 95 ocr2 i guess ? how stable is the system overal […]
Show full quote

thanks for tagging me on your message.
very interesting stuff.
that is windows 95 ocr2 i guess ?
how stable is the system overall ?
asking this because quake is not a great indicator for stability in general.

It's Win98SE at this time.
Until now, there is no stability for 3D with DirectX.
Maybe I will try 95, too, if the problems persist.

feipoa wrote on 2023-02-27, 02:16:

What about 1024x768 and 1280x1024?

feipoa, I see, (not only) you will test your own HOT-433-system soon again 😀

Now the testing of Win98SE + GF2 MX 400 is over (no BIOS-optimization yet):
- the only fast detonator-driver for GLquake is for me 6.31 (8.05 and 12.41 work both, but they score at 800x600 about 10,0-10,5 FPS -> this is really a shame for this system...)
- with 6.31, for GLquake they are: 640x480 ->38,2 FPS; 1024x768 ->31,1 FPS; 1280x1024 -> 24,7 FPS; 800x600 -> 38,4 FPS (the fastest one result; no picture of it)
- with 6.31, Quake II scores about 11,0 FPS -> too slow, I think...
- no start of 3DMARK99MAX with any detonator-driver above (even freeze of the rotating cube in the Dx-6.1-interface after 1-2 seconds during the VGA-test)...

BTW, in 2D (desktop) I use 1280x1024dpi in True Color.

Non-working D3D was the issue for me as well. If you cannot use D3D, then get a few more FPS using NT4 as the operating system. For me, 28.4 fps in NT4 vs. 26.7 fps in w95.

Looks like 800x600 is fairly optimal.

What I don't understand is why your results with Detonator 12.41 are only 10 fps. When I tested w/12.41, I received 26.7 fps at 640x480 w/Am5x86-160.

feipoa wrote on 2023-02-27, 02:16:

Were you able to get any Direct3D games working with your GF2 and Am5x86 in Win9x? If so, which games, which DirectX version, and which Detonator driver version?

Now it's the 2-nd time for me experiencing a big CPU-performarce-crash during a RUNNING Windows (on-the-fly) with a HOT-433.

At this time I am using a Rev. 1 with EDO-support, but I saw exactly the same problem some months ago with a Rev. 4!

The same AMD DX5 @ 180/4V works fine and absolutely stable without problems on the KM-S4-1 shown above, so it's not a CPU-problem.

Still not sure, but maybe the HOT-433 (regardless of the revision) is not stable at FSB of 60 MHz.

In Windows 95C and 98SE, the benchmark-scores of GLQuake/Quake II are dropping down from one test to the next (in about 10 seconds). After the problem occurs, both SiSoft Sandra and CPU-Z show the same phenomenon: an extremly big loss of CPU-performance!! That's really strange and very, very scarry...

The same problem happens using WinME, as I was able to get 3DMARK99MAX working with detonator-versions 30.82, Omega 14072 and 44.04 (can be found on Phil's Computerlab).
In WinME with these drivers, GLQuake and Quake II does not start anymore.

BTW, on Win95C/98SE, the combination of above drivers +3DMARK99MAX still does not work.

In fact, at least for me, either GLQuake, or 3DMARK99MAX is possible at once. For 3DMARK, WinME is necessary.

In the next step, I will test the same CPU (180 MHz/4V) with an ECS UM8810P-AIO (setting for FSB 60 MHz is possible).