I skimmed the datasheet and I didn't see any pins for voltage ID, so I'd say the voltage is whatever the motherboard is set to.
Either that has jumpers, or it's fixed and you should measure it.

Measurement of the Vcore was a very good idea!

As the datasheet offers only the topview of the pin-layout, AND as there are two different kinds of Voltage (separately for CPU and for I/O), more attention is required for the measurement on the bottom side (see picture).

For some possible CPU-frequencies tested (changing of the CPU-ratio for having 225 to 375 MHz, according to the increased FSB of 37,5 MHz), AND for measurements at the shown bottom-pins AS WELL at the voltage-generator itself (AME1084), the Vcore is always 2,9 Volt! Constantly!

This answers at least 3 questions:

1. It seems to be so, that at least some GX1-chips still accept the "old" higher voltage (2,9 V) of the GXm-generation (similar to my experience with the Cx6x86L shown in this thread, as I was able to run it like the 6x86-non-L stable at 3,5 Vcore instead of 2,8 Vcore). This is maybe similar to some CPUs AMD-486-DX5-133 accepting 5,0 V instead of 3,45 V (and so they are able to reach 200 MHz).

2. Why my CPU can run at 375 MHz (due to the higher voltage).

3. This board can not adjust the Vcore automatically at any other voltage (it's fixed, as "myne" means it, too).

Maybe you should test, if your particular CPUs can accept higher voltage than e.g. 2,0V or 2,2V (if a changing of the Vcore is possible on your mainboards, or if a CPU does accept 2,9 Vcore like mine).
So this would be the real true luck for overclocking them.

I know, SiSoft-Sandra is not a very representative benchmark.
Anyway, I would show the influence of the CTCHIP34-batch shown above once again (activation of RSTK, LOOP and FP_FAST).
The screen-shots are from Sandra 2004 SP1, CPU-Multimedia-benchmark.

Another stability-test for CPU@375 MHz/ RAM@125 MHz: Memtest
The CPU is recognized as Pentium MMX

Have you looked up the exact data sheet for that particular processor and checked what it says the tolerances are on the Vcore? If you're exceeding the allowable vCore range you could be drastically accelerating the time to failure for that processor, if I'm not mistaken.

Paralel wrote on 2025-02-16, 00:42:

Have you looked up the exact data sheet for that particular processor and checked what it says the tolerances are on the Vcore? If you're exceeding the allowable vCore range you could be drastically accelerating the time to failure for that processor, if I'm not mistaken.

But yes, of course this Vcore of 2,9 V is much more than the maximum shown in the GX1-datasheet (2,31 V) 😀
My system is not running all the time "24h/7d a week" - it's just running for some more hours, so I don't see any problems about this (as this particular CPU already has proven its stability).
I just don't have any other MediaGX-mainboard (even less one with adjustable Vcore), so for me there is no option to change the Vcore for this CPU (and so test if it can still do 375 MHz at a lower voltage).
This system is meant just to show the power of a MediaGX-chip -it's not for everyday use. This is the same principle as having an AMD 486-DX5-133 (3,45 V) running at 200 MHz (5,0 V).
Sometimes we need to take some risk to reach more power 😉

gonzo wrote on 2025-02-16, 11:05:

But yes, of course this Vcore of 2,9 V is much more than the maximum shown in the GX1-datasheet (2,31 V) 😀

2.31? Ha. I love it. That is fantastically specific.

Also, I completely agree with your assessment. It is truly fascinating that it is able to overclock so well and so stably. It shows the true capabilities of these processors which I have absolutely underestimated.

Paralel wrote on 2025-02-16, 19:13:

gonzo wrote on 2025-02-16, 11:05:

But yes, of course this Vcore of 2,9 V is much more than the maximum shown in the GX1-datasheet (2,31 V) 😀

It shows the true capabilities of these processors.

Well, in fact the MediaGX is (more or less) a better Cyrix 5x86 (not even 6x86), made at much finer tecnology (and so having a much lower power-consuption, espetially the GX1), with added MMX, build-in RAM-controller and PCI-controller, but not accepting L2-cache. And maybe added Pentium-identification for the Operating System and some software.
Sadly it was on the market about 5 years later than the 5x86, in a completely new era of CPU-performance compared to the other CPU-models made by other producers in 1999/2000.

Therefore, it performs per MHz (even overclocked) at the level of a Pentium I 233 MMX from 1996/1997. So for me this system is just a kind of "nice to have it", as it is too slow for the the time it was made.

Does somebody know, are they some advantages of this CPU compared to e.g. a Pentium I MMX, just because it's a Cyrix-CPU?

OK, so you are running Vcore at 2.9 V. Is I/O voltage also at 2.9 V?

Does your system have a CPU temperature sensor? If so, what does it read after 2 hours of uptime at full load, 375 MHz?

I'm not sure if I'd want to run a ceramic GX1 at 2.9 V. I think the BGA chips have quicker heat dissipation.

It would be nice if I could run at 333 MHz. I might play around with some higher voltages.

Yes, the MediaGX is like a Cyrix 5x86 with MMX and higher clocks. This is why I was drawn to the CPU. It is unfortunate that they didn't get branch prediction working. The core revision seems to follow the S0R5 from the Cyrix 5x86 series, rather than the S1R3 revision.

gonzo wrote on 2025-02-17, 09:06:

Does somebody know, are they some advantages of this CPU compared to e.g. a Pentium I MMX, just because it's a Cyrix-CPU?

I believe the only debatable advantage is that Cyrix, before they sold to Via, faced with the glaring reality that performance wasn't going to pay the bills, had attempted to pivot to low power. So in theory, you can jam it in a sealed box in a dusty factory and have it reliably do the needful for decades on low wattage with nothing but a sad little extruded lump of aluminium to spread the heat around the sealed box a bit better.

feipoa wrote on 2025-02-17, 09:12:

OK, so you are running Vcore at 2.9 V. Is I/O voltage also at 2.9 V?
Does your system have a CPU temperature sensor? If so, what does it read after 2 hours of uptime at full load, 375 MHz?

The I/O voltage is 3,3 V.
There is no temperature sensor onboard, and it seems to be no sensor build-in in the CPU.
There is no info about the temperature in BIOS and in SiSoft Sandra (the pictures are from Sanrda 2004 SP1; BTW, the Vcore of 2,2 V shown there is wrong - they are 2,9 V).

What I can tell you, is, runnig at 375 MHz/2,9 V, and using a fast medium sized fan, the cooler (it is a bigger one from a socket 462, aluminium only, no copper inside) becomes not very hot (maybe about 35 °C).
So I do not need to take "special care" about the cooling.

Are you running it on the CPU in a closed case?

feipoa wrote on 2025-02-17, 20:37:

Are you running it on the CPU in a closed case?

Yes. No problem.

Does anybody have experience with SET6X86.EXE ?

The only version I can find requires EMX.EXE in the same directory to work properly in pure DOS, but the only version of EMX.EXE I can find seems to want OS/2 to work properly.

EDIT: found the pure DOS version on this very forum, D'OH!

Link including attachment in OP for anybody else who stumbles across this looking for the same thing.

TLDR: Why would modifications to a 6x86 CPU register remain in place after a long power down including disconnection from the wall?

Okay so this kind of insane.

Used SET6X86 to modify a register with the aim of tweaking LFB performance on my motherboard's onboard GPU (ATI Rage Pro 3D) - a bit like a DIY FastVid because FastVid won't work on this system.

Original VSPEED results:

~46 MB/s (banked) / 52 MB/s (LFB)

After register tweak (C2 0x82 >> 0x83)

~53 MB/s (banked) / 60 MB/s (LFB)

Quake timedemo jumped from 28.1 FPS to 32.4FPS.

The weird part:

The fix has stayed in place even AFTER I powered off the PC and left it unplugged from the wall for several hours.

The register I modified has returned to its original state (as expected following a power cycle) - but the performance improvement remains the same, confirmed with both VSPEED and Quake timedemo.

Just...how?! Does this suggest that the motherboard/chipset or video card retained the optimized LFB behavior? Would be very curious to hear from anybody who knows how that could happen. I expected to have to add the tweak to a batch file, but...it's somehow permanent?

Specs:

IBM 6x86L PR166
MS-5182 motherboard
ALi M1541 and M1543C
ATI Rage Pro 3D 8MB onboard
128MB SD RAM
6GB hard drive

Maybe there is a part of the program that you used that sets it again after boot maybe it’s not showing that it is set though

Well, I can't complain - means I don't have to fiddle with batch files or autoexec 😁

The programme may have written to the motherboard's NVRAM for a particular CPU register setting. This would explain it not being cleared at reset. If you reset the CMOS or swap motherboards, is the setting reverted to default?

AMISETUP, for example, writes to NVRAM for CMOS hidden and non-hidden settings, which get retained in a similar fashion [in NVRAM] until your CMOS battery dies or you use the reset jumper.

Dan386DX wrote on 2025-11-02, 00:29:

Used SET6X86 to modify a register with the aim of tweaking LFB performance on my motherboard's onboard GPU (ATI Rage Pro 3D) - a bit like a DIY FastVid because FastVid won't work on this system.

I am new do this LFB and FastVid. Is it possible to have LFB performance boost to other CPU-s as well?

FastVid is for Intel
K6INIT is for AMD
CYRIX has no tool?

What about other less known CPU-s, like Transmeta Crusoe?

feipoa wrote on 2025-11-02, 03:33:

The programme may have written to the motherboard's NVRAM for a particular CPU register setting. This would explain it not being cleared at reset. If you reset the CMOS or swap motherboards, is the setting reverted to default?

AMISETUP, for example, writes to NVRAM for CMOS hidden and non-hidden settings, which get retained in a similar fashion [in NVRAM] until your CMOS battery dies or you use the reset jumper.

Thanks for the reply. Good call about the NVRAM writes, surprisingly even a successful CMOS clear did not get rid of the enhanced LFB. So at this stage, I'd say one of three things is at play

1) I'm mistaken. Possible, but my hand written notes for the pre-set6x86 surgery clearly say "46 banked, 52 LFB. 28FPS Quake"
2) The conditions I ran the original tests were somehow different; although I've always used the same DOS config, I've even been trying other modes and underclocks to try and replicate the early lower results and can't.
3) The changes are written somewhere else, like the chipset or VGA (are these even non volatile on platforms this old?)