dr.zeissler wrote:

Multi is much more important. 133x1.5x should be 200mhz. 😀 very dosfriendly!

The ability to adjust the FSB speed on the fly is more important than you would think once you start playing around with disabling the CPU caches. This makes the system speed almost entirely dependent on the RAM / FSB. 3 x 50 Mhz with no L1 or L2 cache has very different performance characteristics than 3 x 100 Mhz with no cache.

A couple of updates:
-I am experimenting with different sound cards in this machine. I can't get the Soundscape to "play nice" no matter what I do. SSINIT hangs when trying to initialize the card about half the time and there is no pattern to it. I decided to put it back in my DX4 100 where it works perfectly rather than keep fighting it here.

-Right now I have a Sound Galaxy Waverider Pro 32-3D PnP (AZT-2320 + Wavetable) installed that I hadn't messed with yet. As others have noted, the PnP configuration for these cards in DOS is difficult. I got all the resources squared away in the Intel PnP utility however and after that I was able to install the card without any issues. It doesn't seem to have any major problems at it sounds decent, but I'm not 100% sold on it yet. I'll probably leave it in for a while and make some recordings before deciding.

-My DVD drive doesn't output audio through the 4-pin connector. I noticed when I looked closely that the 4-pin audio port is labeled "Reserved" and not audio. It was made in 2007 and is probably new enough where they stopped implementing the analog audio out. I need to find a different drive and swap it out.

I'm building a very similar setup to yours but made the mistake of initially buying a Nehalem C3 which unfortunately doesn't seem to scale well, or at all in the 486 speed range. That said, I also can't seem to get the board to work reliably at 133MHz FSB. I'm running the CPU at stock voltage (1.45v) but I'm not sure if its a chipset issue or CPU voltage.

What voltage is your CPU rated at? What are you running it at?

In terms of other problems, I have 3 sound cards in this machine. A Gravis Ultrasound Pnp which is a ressource nightmare, an AWE32 and an Audigy.
The Audigy is for Win98 only and I only have it there because it supports 48+kHz sampling rates.
Still though, after quite a battle, everything seems to work in DOS and Windows 98 with the exception oddly enough of Future Crew's ASM-94 invitation intro which locks up in the space invaders section. I wonder if its an issue with the USB keyboard emulation in DOS.

Cheers

infiniteclouds wrote:

SetMul with the C3s should work in a S370 but software controlled FSB won't.

i can set it only to 3 which is 400Mhz
if i switch from 6 to 3 it freezes.
so i need to switch it slowly.

HW is EPIA 800.

Viaudio.com / viafmtsr woes with l1 and l2 disabled.
simply to slow. leave l1 and l2 enabled it is too fast for wing commander..
So no sound is possible in older games with throttle.. (l1/l2).

My experiences are it's better to use a pentium and disable the caches or throttle down, than using those C3/K6+ etc.

VIA C3 (Ezra-T) has smoother and wider hardware CPU scaling than P54C/ P55C and K6-III+ (on Super Socket 7). It is only beaten by K6-III+ on a turbo-switched Socket 7.

The flexibility for C3 is dependent on the FSB range available. FSB From 50-133MHz or 66-150MHz gives it a smoother scaling range. Not all CPUs are capable of this range - depends on the quality of the CPU and motherboard. A gentle voltage boost can help the C3 (within 10% spec should be fine). Also, low multiplier of 3 is dependent on the quality of the C3 - not all are capable.

Here's a C3 scaling range for an Ezra-T, 1.0 Ghz and 133 MHz FSB specification, on a 440BX board capable of 150 MHz FSB:

Re: WIP 1: Ezra-T slow-down machine, S370 industrial build...

The scaling benchmark index is represented by Speedsys. You can then correlate other benches from this and approximate actual CPU speeds. You won't get this scaling with P54C/ P55C and K6-III+ (on Super Socket 7). Only beaten by K6-III+ on a turbo-switched Socket 7 build.

j^aws wrote:

Here's a C3 scaling range for an Ezra-T, 1.0 Ghz and 133 MHz FSB specification, on a 440BX board capable of 150 MHz FSB

150MHz on a BX chipset? Lucky!!
I can't seem to even do 133MHz reliably.

I use a Nehemiah at ~1466MHz (133MHz FSB) as my main retro system. At that speed, it's roughly as fast as a PIII 733 for demanding 3D games under Windows and usually quite a bit faster in everything else. SETMUL is an amazing tool and you can bring the processor all the way down to just 4x multiplier. Lowering the bus to 66MHz, you can easily hit very low speeds if need be.

I personally have few DOS games that really require that sort of tinkering, but there are many Windows titles with speed issues as well and I just can't go back after first experiencing K6+ CPUs with unlocked multipliers. I happened to need a little bit more power than that however, so when I got the chance to grab that Nehemiah, I did it in an instant and I have been very impressed so far.

Apparently the Ezra core is half as fast when it comes to anything FPU related?

Yes, the Nehemiah is still a very good retro-chip. It should be able to hit (on a 50/66 capable board) all of the points that a K6 can and higher... but still can't hit the fast 486 speeds that a Pentium or Ezra-T can.

One thing is running a generic benchmark like 3dbench 1c and the other is a real game.
3dbench1c delivers ultra smooth and lot's of different values when throtteling a PIII or PIV CPU,
lot's of games do react very different on the throtteling.

dr.zeissler wrote:

One thing is running a generic benchmark like 3dbench 1c and the other is a real game.
3dbench1c delivers ultra smooth and lot's of different values when throtteling a PIII or PIV CPU,
lot's of games do react very different on the throtteling.

When you say 'throtteling', are you referring to using THROTTLE.EXE? If so, what you describe with games doesn't surprise me.

Hardware slowdown using Pentiums, C3s and K6-IIIs by manipulating multipliers, FSBs, various cache disabling (L1, L2, data and instruction), branch prediction disabling and so on, aren't equivalent to using ACPI and THROTTLE.EXE. You can easily test this using a motherboard that can use both a VIA C3 and ACPI for THROTTLE.EXE (e.g. 440BX).

Slowdown the CPU by using both the aforementioned hardware technique and by THROTTLE.EXE, so that you have roughly the same CPU speeds using both methods (test with a benchmark tool), and run some demoscene productions that are speed-sensitive using both slowdown methods. Providing you have a compatible VGA card, you can see graphical corruptions when using THROTTLE.EXE in some productions but not by using hardware slowdown.

Sorry, it has been a while since my last update. I have been having some issues with this thing. Long story short: The BIOS on my GA-6BXC is bricked and I either need to reprogram it our get a new motherboard.

I was trying to use the Intel PnP Utility to reassign resources for the Waverider Pro PnP card I mentioned earlier. I had done this successfully before, but this time I got an error message saying that it couldn't save my settings and another one telling me that I shouldn't use this utility on an EISA system, but to use the ICU instead....

Well, the next time I rebooted I was greeted by a recovery BIOS asking me for a boot floppy. I made a boot floppy, tried to re-flash the BIOS, and it failed. Now the machine won't even boot the floppy disk anymore. Extremely frustrating. Needless to say when I get this fixed I am never touching the Intel PnP Utility again.

tpowell.ca wrote:

I also can't seem to get the board to work reliably at 133MHz FSB

I am starting to wonder about this myself. I have been through several combinations of PC133 RAM sticks and i usually get a couple of failures running memtest86+ on the second pass. If I can't get it to work on the new motherboard I will probably just stick with using a max of 112 mhz FSB.

tpowell.ca wrote:

What voltage is your CPU rated at? What are you running it at?

My Ezra 866 a A350EJ0 stepping and is rated at 1.35V. I never confirmed what the board is actually supplying it come to think of it.

dr.zeissler wrote:

One thing is running a generic benchmark like 3dbench 1c and the other is a real game.
3dbench1c delivers ultra smooth and lot's of different values when throtteling a PIII or PIV CPU,
lot's of games do react very different on the throtteling.

Something I found from messing with this build: You are wasting your time trying to get one of these time machine builds to perform "like a 486 DX2" or "like a Pentium MMX 200". Just find multiplier and FSB settings that work with the specific game you are trying to play. The slowdown / speedup is very smooth, but with much more modern components in general such as PCI or AGP graphics and faster bus speeds the performance characteristics of a system like this will be much different than a "real" older system.

Garrett W wrote:

Apparently the Ezra core is half as fast when it comes to anything FPU related?

Yes. The synthetic benchmarks that I have run put an Ezra running at 1 Ghz in the ballpark of roughly a P3 500 for integer performance and a Pentium 266 for floating point. The FPU is pretty sad, but it still seems to be enough for games like Quake 3 to be very playable.

^^ If you are having trouble hitting 133MHz on a Gigabyte GA-6BXC Rev 2.0, I suggest checking the boards capacitors. This board has a tendency for bad caps, especially around the AGP slot.

Also, I find it quite puzzling that you managed to brick the BIOS with a PnP utility? Sounds more like there are other issues with the board. For reference, I have a stable Gigabyte GA-6BXC Rev 2.0 that hits 133MHz comfortably. Ensure you also have a quality slotket/ slocket with a voltage clamp chip for stability.

j^aws wrote:

I find it quite puzzling that you managed to brick the BIOS with a PnP utility? Sounds more like there are other issues with the board.

The original assumption that I made was that the Intel PnP Utility corrupted the BIOS while it was updating the ESCD. I suppose that given a couple of the other things that I have seen: Persistent memory issues at 133 MHz and I also had my Windows 3.1 system.ini disappear for no apparent reason, you may very well be correct. I don't see any capacitors that are obviously bulging, but that doesn't necessarily mean anything. I ordered a different MB, we'll see how it behaves.

^^ Have you tried reseting the BIOS with the usual reset pins located around the RTC? I've seen failed boots with blank screens fixed again.

j^aws wrote:

^^ Have you tried reseting the BIOS with the usual reset pins located around the RTC? I've seen failed boots with blank screens fixed again.

Yes. Didn't help.

Tenorman wrote:

j^aws wrote:

^^ Have you tried reseting the BIOS with the usual reset pins located around the RTC? I've seen failed boots with blank screens fixed again.

Yes. Didn't help.

One last thing to check/ test is swapping out your PSU. The one you stated using, Coolmax 300W, doesn't seem that familiar, and judging by online reviews, doesn't seem that reliable. Be sure to test it before using another motherboard, as PSUs can be silent killers when all else looks fine.

I swapped out the motherboard with another GA-6BXC Rev 2.0 and we are back in business.

I got a couple of successful passes with Memtest 86 @ 133mhz and the machine hasn't done anything "weird" again so far like files randomly disappearing. Hopefully the stability problems are fixed now.

I am still trying to decide what I want to use for sound cards. I'm not a huge fan of the Aztech Waverider PnP card I was messing with. The MIDI patch set just isn't very good and the DOS software (mixer, etc) seems rather amateurish. Right now I am trying to get a SB 32 CT3670 and an Ensoniq SoundScape Opus PnP card to play nice together. Having some trouble, will make a separate thread for that under Sound.

OK, I think I have all the kinks ironed out and I'm fairly happy with this setup.

The Soundscape Opus works a lot better than the original S2000 in this machine. My theory would be because it is a slightly newer, fully PnP card. The patch set isn't as good, but it still sounds better than the AWE in a lot of games and the MPU-401 functionality is better (can drive many intelligent mode MT-32 games without SOFTMPU).

Final Specs:
-Generic ATX desktop case
-Coolmax CX-300 power supply
-384 MB PC133 RAM
-Gigabyte GA-6BXC Rev 2.0 motherboard (PowerLeap BIOS, older BIOSes will not correctly identify the CPU and have issues with larger hard drives).
-MSI MS-6905 Slocket Rev 2.0
-Via C3 Ezra 866 (A350EJ0)
-Creative 3D Blaster Savage4 Pro 32 MB PCI
-Diamond Monster Voodoo2 8MB PCI
-32 GB Compact Flash (Windows ME / Linux)
-4 GB Compact Flash (DOS 6.22 / Win3.1)
-Realtec 10/100 PCI Ethernet Adapter
-SoundBlaster 32 CT3670 with 2 MB RAM
-Ensoniq Soundscape Opus
-Samsung SM-348 IDE CD-RW/DVD Drive (the LG I started with did not have analog audio out).