gerwin wrote:

SetMul allows disabling the L2 cache on the K6 Mobile and VIA C3.

Awesome! Thank you for that!

You're welcome.

Most of the SetMul functionality is for the VIA C3. But the user base for this series of processors is very small. We had a discussion about it here: Cyrix III (i.e. VIA C3) for vintage gaming?. Not all things there reflect my recent experience with VIA C3. Which goes like this:

The C3 is a further development of the Winchip, with VendorID 'CentaurHauls'. With the right slotket any VIA C3 processor ran on all four of the i440BX based boards that I tried, from Aopen, Gigabyte, Soyo and Shuttle. The right slotket being the 'MS-6905 Master' for any C3, and the 'Upgradeware Slot-T' for any Tualatin socket version: Ezra-T and Nehemiah.
All VIA C3s ignore the normal Intel multiplier jumper signals, but they do support software adjustment. Later C3s have bridges on the CPU package that can be modded to some degree, to change the default multiplier.

The VIA C3 Samuel 1, Samuel 2, Ezra and Ezra-T are quite similar. I could not get their cache and memory throughput up to Intel level. I don't know why that is. This is about the fastest of this first series, on a GA-6BXC board, LFB write combining enabled:
VIA C3 Ezra-T (C5N) at 9.0x133=1200MHz:
-Speedsys=626,28 pts.
-Quake I - 640x480=38,1 FPS.
-PC Player Bench=50,5 FPS.
With certain caches disabled it does make a good 286/386/486/586 emulator. On a BX, We did not have that before.
Edit: Benchmarks by Kamerat, and some compatibility tests

The next and final C3 version behaves differently, it makes the previous versions seem unfinished. Cache and memory throughput are sufficient now:
VIA C3 Nehemiah (C5XL) at 9.0x133=1200MHz:
-Speedsys=793,93 pts.
-Quake I - 640x480=71,5 FPS.
-PC Player Bench=106,4 FPS.
This version is not as good in 486 emulation though; It does up to a 486DX-40 at most. Still it does better in that regard then the Intel Pentium II/IIIs.

Great work, thanks for releasing this!

I tested it with my K6-3+ last night. With a 2x multiplier and 66Mhz bus Speedsys rates it as a P133.

I was hoping that disabling the caches too would give me something slower than a 386-40 but interestingly multiplier doesn't seem to make any difference to the Speedsys score with both caches disabled.

Also, I noticed a difference in Speedsys score when disabling both caches through the BIOS as compared to disabling them using SetMul.

L1 & L2 disabled through BIOS = ~10
L1 & L2 disabled through SetMul = ~19

Perhaps the BIOS disables both L2 & L3, whereas SetMul only disables L2?

bjt wrote:

L1 & L2 disabled through BIOS = ~10
L1 & L2 disabled through SetMul = ~19

That's very interesting indeed!

I have to check this out one day and see what performance points you can hit with a SS7 machine 😀

Great job! Thank you very much!

bjt wrote:

Great work, thanks for releasing this! I tested it with my K6-3+ last night. With a 2x multiplier and 66Mhz bus Speedsys rates i […]
Show full quote

Great work, thanks for releasing this!
I tested it with my K6-3+ last night. With a 2x multiplier and 66Mhz bus Speedsys rates it as a P133.
I was hoping that disabling the caches too would give me something slower than a 386-40 but interestingly multiplier doesn't seem to make any difference to the Speedsys score with both caches disabled.
Also, I noticed a difference in Speedsys score when disabling both caches through the BIOS as compared to disabling them using SetMul.
L1 & L2 disabled through BIOS = ~10
L1 & L2 disabled through SetMul = ~19
Perhaps the BIOS disables both L2 & L3, whereas SetMul only disables L2?

When you use an AMD K6 with on-Die L2 cache, the motherboard external cache changes from L2 to L3. I found that when I disable the L1 cache, the L2 seems to become non-functional too. Giving me 20,87 points in Speedsys at 2.0x60=120MHz. This is with the L3 cache still working. If the L3 prevents the system from getting slow enough you should try with the external cache disabled in the BIOS. SetMul only toggles CPU on-Die cache. I don't know what the resulting speed will be.

When left without sufficient cache, data must travel to and from RAM all the time: the FSB speed becomes even more important then it already was, while the clock multiplier matters less. Unfortuntely I do not have much success in changing the FSB speed on an ALi Aladdin V board without reboot. The ALi Southbridge is not well supported, while for i440BX boards there are several tools to do just that (SoftFSB, SetFSB, CPUFSB, RayeR's SMB for DOS).

In Speedsys you can see the raised memory speed ranges in the lower right graph. The three different levels indicate that L1+L2+L3 cache is at work there. One can see what size these caches have, even without reading the text in the graph.

gerwin wrote:

If the L3 prevents the system from getting slow enough you should try with the external cache disabled in the BIOS. SetMul only toggles CPU on-Die cache. I don't know what the resulting speed will be.

That explains my results. I got a SpeedSys rating of ~10 with all caches disabled, regardless of multiplier. This is roughly equivalent to a 386-40. I'll have to try one of those software FSB programs to see if I can slow things down further (TX chipset).

Just so you know, If the FSB software is not written for both a particular Southbridge and a particular PLL clock generator chip, it won't do anything. And usually all FSB options of the PLL are already available through jumpers or the BIOS.

Regarding Socket 7 slowdown in general: The Cyrix 6x86L, using a 1.0x multiplier / 60MHz / no caches, still scores 10,84 in Speedsys. Now that you mentioned it, It seems hard to get below the 10 points on Socket 7.

gerwin wrote:

Just so you know, If the FSB software is not written for both a particular Southbridge and a particular PLL clock generator chip, it won't do anything. And usually all FSB options of the PLL are already available through jumpers or the BIOS.

FSB change through software for Super Socket 7 would be awesome. I understand it's a complicated task though.

Some EPoX MVP3 socket 7 boards are supported by SoftFSB, to name something. You can find it in the file SoftFSB.dat.

Just checked the VIA C3 Nehemiah with the SetMul options for Cache and such, and setting FSB speeds from 50 to 133MHz. Lowest SpeedSys score I could get is 3,77 points (~386SX-20), this could be ramped up gradually to 14,46 points (~486DX-40). When enabling all L1 cache again it jumps to 127,30 points at least.
I could also use Throttle to get even lower Speedsys scores on that machine. But that ACPI slowdown method does not always work well with games.

hello gerwin,

could we mirror your version of Ali V Throttle and SetMul10 at K6plus.com ? It would be nice to have these in the utilities section...

Yes, no problem. This place you mean: K6Plus.com Utilities Section

Though maybe you better wait a little bit, as I have SetMul version 1.1 ready. Just that the readme needs to be updated.
- AMD K6: Exclusively disable L1 cache, leaving L2 functional
- Pentium P54C test register options. (wished intel retained this on the Pentium MMX and the pentium II/III...)
- Disable L2 cache on a Pentium II/III
- Little support for Winchip C6
- Fixed protection fault when running it on a 486 system

I did not mention the modified Throttle in this topic before, so I will do that now:
Throttle for ALi Aladdin 5 Socket 7 boards.

Thanks a lot!

that is great news, let us know when you have finished the SetMul v1.1 package.
I will send the Ali V Throttle to the site admin in the meantime.

These utils definitely should be hosted at vogonsdrivers by as well. IMHO their usefulness is comparable to SoftMPU only (among recent developments for time machines)

Thanks for the appreciation.

v1.1 is available. 😀 See section "NEW SINCE UPDATE v1.1" in the readme.

I wanted to add better CPUID support detection too. But that requires a custom error handler and I have no good DJGPP example of how to do this, yet.

Given the the target CPUs work, it is only a cosmestic issue, I suppose...
One thing I do not get:

Fixed protection fault when running SetMul on a 486 system

how could one run Socket 7 or later processors on a Socket 3 board ? Or is it just precaution for dizzy midnight time-travellers who left the KVM switch on the wrong position...? 🤣

I fixed it because I wanted SetMul to disable the L1 cache on my 486. That and programs crashing is never a nice thing, If possible it should at least give a message like '...not supported'.

doh.
That did not cross my mind... So the utility can be useful as a replacement of ICE and ICD, which work in real mode only.

Thanks for this latest version. I'm using it to enable/disable L1 cache on my 486 DX4-75 laptop, which doesn't have cache control options in the BIOS. I'm seeing slow 486 (SX25 or SX33) performance with L1 disabled.

Since this week a Cyrix 5x86 100GP S1R3 was added to my inventory. The CPU is working fine in a cheap 486 VLB motherboard, with a modded BIOS. The CPU can be jumpered for multiplier 2.0x or 3.0x.
I have modified SetMul to tinker with the multiplier options which are mentioned in the 5x86 Datasheet. The good news is that one can toggle between the bootup multiplier and 1.0x anytime, but seemingly that is about all it can do. It won't do any switching between 2.0x, 3.0x or 4.0x. There is no observed effect of the half clock speed bit.
Will release it once I add a proper CPU detection routine for this one. If someone wants to try something I would be happy to mail it the way it is now.