Thank you! I was just reading about WPCREdit 😀

Your post saves me lots of time/research. I'm going to give PCISET a try.

-Brian

That worked!!

For anyone else stuck using AMD K7 with SETMUL, download this:
https://www.uwe-sieber.de/files/pciset.zip

Then use the following written by mockingbird:
pciset 0305 1106 55 00000100 00000100

I recommend putting this in a BAT file.

EDIT: I just wanted to point out that this register is specific to the KT133 series chipset. A different register will likely be necessary for another chipset.

-Brian

I have confirmed this works on KTA133A -based Epox 8KTA3+ non-pro.
Can now clock my Mobile XP (Barton) from 300 (3*100) to 1866 (14*133) in SETMUL.
Surprisingly I still have to unlock higher multipliers, I thought they should already be unlocked on mobile CPUs.
Oh well, that's not much of an issue.

After long time I've tested the PCIset command.

Setmul works with AXMH2500FQQ4C and now I can have 3x multiplier. Finally! CPU set at 300 MHz!!!

Throttle and everything working great!

Don't know why, but DOS losses access to the hard disk after disabling the L1 cache of my Cyrix MediaGX 166MHz CPU via SetMul. My motherboard is an AVT INDUSTRIAL, LTD. 586 GX LITE V2. Beside that I cannot even disable the "CPU Internal Cache" in BIOS. The value "Enabled" is white-colored and cannot be changed to "Disabled".

plain plain DOS no DMA driver loaded?

It turns out that there are two options in BIOS, namely Read Prefetch and Write Buffering, affect the result. After disabling them the L1 cache can be disabled without any problem.

I have recently purchased two Via C3 CPUs, both the 800A, one built in 2001 and it's the 1.35V version, the other was built in 2004 and it's the 1.25V version.

The first is an Ezra (or T) core and the second probably Nehemiah?

JSO wrote on 2023-09-29, 09:43:

I have recently purchased two Via C3 CPUs, both the 800A, one built in 2001 and it's the 1.35V version, the other was built in 2004 and it's the 1.25V version.

The first is an Ezra (or T) core and the second probably Nehemiah?

Pictures would help.
A Family model stepping readout would be the definitive answer. For example C3 Nehemiah+ = 6.9.8

I created this post as a thread, then realised this thread existed and thought it fits better here. TL:DR - I bought a C3 500Mhz for use with SetMul and DOS and it's made a promising start.

A little while ago I built a Socket 370 system for 90s and DOS gaming, it’s been super for mid to late 90s titles; especially Quake, SWAT 3, Falcon 4.0.

But the Pentium III 866MHz ‘Coppermine’ is not the most flexible CPU and runs many older titles far too quickly.

Every man and his dog advised me to get a Via C3 CPU, particularly an Ezra or Nehemiah if I can find one. Along with SetMul they have a wide range of performance options. I’m still yet to get my hands on one of those, but I did snag a cheap C3 Samuel 500MHz chip and have herein lies my review!

For these tests, no discrete graphics card is involved, onboard only. The other specs of the PC are: Biostar ML6VB motherboard, 256MB Samsung PC133 RAM, 64GB SSD (via IDE adaptor) and a Soundblaster CT4180 Vibra 16.

First, a straight fight between the outgoing Pentium III and the C3 at stock.

These results surprised me, we knew the C3 Samuel would be slower…but the difference here is rather staggering. That said, it ain’t here to be fast…and at this stage, I was feeling bullish we had a solid foundation for some DOS emulation.

So, down to business; I chose four benchmarks from the very handy dosbench suite from Phil's Computer Lab, and headed into DOS 6.22. For all of these tests, Quake was run at the lowest resolution, Doom was run at the highest settings.

First, a modest change of multiplier, from the usual 5x to 4x - the system now running at 400MHz. The results scaled as expected, still too fast for DOS gaming.

SPEEDSYS score:98
SPEEDSYS CPU equivalent: P6-200
Doom: 57 FPS
Chris's 3D benchmark: 88
Quake: 42 FPS

Next I tested at 3.5x and again, the performance losses were very moderate and scaled with the above. Then at 3x multiplier the system hung immediately. 3.5 is the lowest we can go here, so it's time to try out some of the other options. SetMul comes with a toggle to disable L2 cache, I-Cache and BPD or 'branch prediction' - I wasn't certain what the last option did, but I through it into my testing.

First, 3.5 multi with L2 cache off - the difference was immediate and tangible

SPEEDSYS score:15.72
SPEEDSYS CPU equivalent: Just below 486DX2-50
Doom: Unstable, kept crashing at outset.
Chris's 3D benchmark: 14.5
Quake: Wouldn't start.

Next, 3.5 multi with just i-cache off, still slower, but not as pronounced as with L2 cache disabled

SPEEDSYS score:33.46
SPEEDSYS CPU equivalent: Between 486DX-50 and Pentium 133
Doom: 11 FPS
Chris's 3D benchmark: 34.5
Quake: 5.2 FPS

Finally I went all in, I turned off both of the cache settings, set the multiplier at 3.5x and disabled BPD. For this test I didn't measure Doom or Quake, really there was no point, we knew they'd run painfully slow. I was more interested in the raw numbers from Chris's 3D benchmark and Speedsys.

SPEEDSYS score:13.56
SPEEDSYS CPU equivalent: Between 386DX-40 and 486DX-50
Chris's 3D benchmark: 13.8

This is the spot, well hopefully. It puts us around a strong PC from 1991; in theory. Time for some real-life testing; I loaded up a game of Wing Commander, then Pirates! from 1987.

Wing Commander was okay, it was much like playing it on a 486SX as I did in the mid 90s, smooth most of the time; occasional slowdowns when too many other ships are nearby. Could probably tweak the settings to speed this one up a bit. Pirates! was perfect, ran at exactly the kind of pace I'd hoped for, it was slow during loading and transitions but I suspect this is more a limitation of its 80s programming. Anyhow, for now I think it's safe to say there's great promise and we have somewhat tamed the beast (lol, if you can call this chip a beast) - both of these titles are no longer unplayable, frames are similar to how I remeber them on period hardware.

Now the caveats: I'm yet to play Swat 3 which is probably the most demanding title I have for this PC. At stock, the Via C3-500 benches like a PII-233, which is on the lighter side forlate 90s and early 2000s titles. So we'll see. I did try an overclock with SetMul but couldn't get it stable. My BIOS has no such options unfortunately.

Also, I have VIAFSB, but, I'm struggling to get it working. If I could adjust the FSB along with the multi, I could potentially go lower. But really, I don't think I need to for now. So, to conclude, 24 years after it was universally panned by the computing world for being far too slow for its time, the Via C3 Samuel 500 has found a welcome home.

Update regarding VIA C3 'Samuel' 500MHz - at stock it can't run SWAT 3 properly, the performance is abysmal. Trounced by a Celeron Mendocino.

To conclude; with SetMul, the CPU is decent for emulating the low end; most of the DOS era will be fine with this CPU. However unlike most other Socket 370 CPUs, it's no good for the later, more intensive early 2000s Windows titles.

I can confirm that the much touted Ezra or Nehemiah are the ones to have for the best performance range on this platform.

Very interesting. My build is a PIII 600Mhz. I really don't want to build another slower system. From the above, are you advising that not much can be done with a PIII? I would loved to play WC1 with it 😀

Apparently those via c3 chips I hear are a continuation of Winchip technology. Those winchips weren’t the greatest either, they were touted as easy upgrades to 200, 233, 266 mhz but if I recall, basically a p166 was better than them.

It’s a shame, because the c3 was almost a cyrix joshua. Which would have been a far more interesting and better cpu.

Hi! i have a question, im using softfsb to change multiplier of my cyrix 5x86 120, and let me do 1x or 3x (normal speed), no problem about this, is amazing to change from 120 to 40 with just one command

My problem is that softfsb randomly hangs, sometimes just first time i use to go to 1x, sometimes when i want to go back to 3x, and sometimes after 3 or 4 fsb changes... but always hangs in some change.. this is normal?

bytesaber wrote on 2024-05-16, 03:13:

Very interesting. My build is a PIII 600Mhz. I really don't want to build another slower system. From the above, are you advising that not much can be done with a PIII? I would loved to play WC1 with it 😀

Besides toggling the L1 and L2 caches, not much other than resorting to ACPI throttling (which is a mixed bag) thru Throttle or CPUSPD.

Changing the FSB speed could help as well, but it is less effective on faster CPUs lacking hardware options to slow it down. 😒

I've been playing with Ezra and Nehemiah on couple of projects. Some questions emerged:

- (Ezra) SetMul reports 3-12 as available range of multipliers, but up from default 9 only accepts 10 (works fine) and 12 (locks). It says 11x is not supported, are locked multipliers inside the supported range to be expected ?

- (Ezra) SetMul reports 9 x 100, but NSSI reports 8 x 100 (if we increase the multiplier to 10x with SetMul, then NSSI reports 9 x 100, e.g. always 1 multiplier behind), is this the expected behavior in NSSI detection with Ezra ? Nehemiah multiplier is detected correctly

- (Ezra/Nehemiah) Do VIA C3 with higher default voltage (e.g. 1.4 vs 1.35) overclock proportionally better (at stock voltage) ?

- (Ezra/Nehemiah) For motherboards with no BIOS settings to set the FSB, nor jumpers, we can use VIAFSB (on supported VIA mobos) to fine-tune the stock FSB of the Ezra or Nehemiah
It's not the same as setting via BIOS/jumpers, since if we go at run-time from 133 to 66, then we need to disable sound, etc. or we can expect it to lock
Is there a way to "fiddle" with the pins of Ezra and Nehemiah, to force it to boot at certain FSB (different than chip default), i.e. to replicate the behavior of FSB jumpers ?

Thanks!

In theory, the FSB straps should be the same as for Intel CPUs.

Hey,

sunmax wrote on 2024-08-13, 05:20:

- (Ezra) SetMul reports 3-12 as available range of multipliers, but up from default 9 only accepts 10 (works fine) and 12 (locks). It says 11x is not supported, are locked multipliers inside the supported range to be expected ?

The set and read table for Ezra has a gap after 10x. The gap of unavailable settings is 10.5x, then 11x, then 11.5x. After that only 12x is available again. 12.5 x and up are not available with an Ezra.
I don't remember how I determined the table contents exactly, I looked at datasheets and the other programs mentioned in the credits. Also I tested lots of settings by just trying, then see what the CPU does. That would include settings I had to mark as unsupported, later.

sunmax wrote on 2024-08-13, 05:20:

- (Ezra) SetMul reports 9 x 100, but NSSI reports 8 x 100 (if we increase the multiplier to 10x with SetMul, then NSSI reports 9 x 100, e.g. always 1 multiplier behind), is this the expected behavior in NSSI detection with Ezra ? Nehemiah multiplier is detected correctly

The total speed measurement in MHz is always the verification. There may be an error in NSSI's readback table.

sunmax wrote on 2024-08-13, 05:20:

- (Ezra/Nehemiah) Do VIA C3 with higher default voltage (e.g. 1.4 vs 1.35) overclock proportionally better (at stock voltage) ?

Don't know. Normally I would expect lower voltage parts to be better specimens, unless they are clocked lower as the higher voltage variant.

sunmax wrote on 2024-08-13, 05:20:

- (Ezra/Nehemiah) For motherboards with no BIOS settings to set the FSB, nor jumpers, we can use VIAFSB (on supported VIA mobos) to fine-tune the stock FSB of the Ezra or Nehemiah
It's not the same as setting via BIOS/jumpers, since if we go at run-time from 133 to 66, then we need to disable sound, etc. or we can expect it to lock
Is there a way to "fiddle" with the pins of Ezra and Nehemiah, to force it to boot at certain FSB (different than chip default), i.e. to replicate the behavior of FSB jumpers ?

The bootup multiplier is determined with the row of bridges on the backside of the C3 CPU. Between the black die-cap and the pins. The amount and logic of these bridges differs among C3 variants. Below some notes I gathered around Ezra, from looking at examples. Last number on each line is the bootup FSB request. The 3rd to 6th bridge are likely the ones used for selecting the multiplier.

1xx Ix II II II   =  5.5x 133
2xx xI xx II II   =  6.0x 133
3xx xx xx II II   =  6.5x 133
4xx xI Ix II II   =  7.0x 133
5Ix xx Ix II Ix   =  7.5x 100
6   ^^ ^^

It is easy to connect a bridge with a pencil stroke or something. But it is quite hard to cut one. I don't think I ever managed to cut one. Did not care much about it, so did not push my luck.

Thanks Gerwin! Very insightful 😀

Looking at the sequence of the bridges in your notes, may we assume they also control the FSB ?

I notice "1-2" and "9-10" have a different configuration for 133 and 100. I read in the C3 specs (re: Bus Frequency Selection) when comparing 133 Mhz and 100 Mhz types: "Processors from these speed bins indicate this by shorting the BSEL[1] pin to ground internal to the package." - is this achieved via the bridges ?

Looking at this picture from Vogons: download/file.php?id=9307

It seems there are couple to the left too, and the ones on the top have 3 connections point.

Can you help matching the bridges numerically to your sequence ?

Thanks again!

sunmax wrote on 2024-08-15, 15:59:

Looking at the sequence of the bridges in your notes, may we assume they also control the FSB ?

I suppose yes. But I have to take a better look at my notes...I was actually mostly tinkering with Ezra-T and Nehemiah, which have different bridges. Sometimes I could easily find the logic in the patterns, other times not so much.

sunmax wrote on 2024-08-15, 15:59:

Looking at this picture from Vogons: download/file.php?id=9307
It seems there are couple to the left too, and the ones on the top have 3 connections point.
Can you help matching the bridges numerically to your sequence ?

Is that your Ezra specimen, or one with the same configuration as yours?