null

The Serpent Rider wrote:

3.2v is not enough to kill that CPU. I would start worrying after 3.5v.

If it runs stable at 3.2v I’ll keep it for now. This motherboard does not do 0.1v increments. Next step would immediately be 3.5v.

Got a super socket 7 mainboard that does 0.1v increments and that hopefully supports some more sensors like temperature. cpu-z does crash if I just swap the mainboard so will try a fresh install.

Anders- wrote:

Yep, one need to remove a bit of heatsink material.
Thin sensors like these helps with the fitting: http://i1270.photobucket.com/albums/jj620/dyk … 982/sfdss-4.jpg

No idea about whether this cpu features some sort of internal measure of temperature, haven't heard of it at least.

Yeah that’s exactly how the temp probe looks like (even a thinner wire like the one on the pictures)
Will give it a shot thx .... for now trying a super socket 7 mainboard that perhaps has some more features in the sensor department.

I stumbled on this when looking for any discussions on IBM 6x86MX vs Cyrix MII processor overclock (or undervolt) performance, but don't really see any on that topic.

I haven't had a large enough sampling to fully compare, but the National Semiconductor produced CPUs all seem to have some overclocking headroom (note NS produced MIIs started with 250 nm, so all the 2.9V parts ate 250 nm). IBM built 6x86MXs and MIIs were 350 or 250 nm depending on the vintage (and Cyrix branded ones can be determined by the manufacturing code), but I have an even smaller sampling of IBM 250 nm chips and none of the fastest/last ofs to work with (ie IBM manufactured PR333 3.5x75 MHz) or any of the rare 250 nm low voltage mobile parts.

However, the 250 nm normal 2.9V rated NS chips seem to overclock well and undervolt well, though it may also depend on the motherboard/chipset. My very first MII 366 (2.5x100) tends to run fine at 2.2V 250 MHz and does 3x300 at 2.9V well enough if it's cooled well (though probably not at its factory rated max temps) and probably better with a bit more voltage.

I'm not sure about the 350nm 6x86MX/MII parts, but the 6x86L CPUs overclock well with overvolting in my experience. At 3.3-3.5V all my PR-200s run at 200+ MHz (68x3 is the best I've gotten, and they don't like 100 MHz FSB, so no 2x100 or 2x105). The processor manufacturing codes on some of the later dated 6x86 CPUs (with 3.3 or 3.5V rating) seem to use the same 350nm process as the 6x86L as well, so I'd think they'd be similarly voltage tolerant. However, there's always the possibility of different ratings used on SMD components that might pop. But, assuming the 350 nm 6x86MX CPUs are built like the 6x86L, they should tolerate 3.5V and not die, but bear in mind heat and current limits of a board and PSU.

Also beware: 250 ns process from one fab might not be similarly tolerant to another and/or SMD components external to the die may be rated differently, so the fact Cyrix/IBM/NS CPUs seem to tolerate crazy high voltage for 250 nm parts doesn't mean others will. I discovered by sad accident that 250 nm K6/K6-2 (or K6-III 400 in this case) CPUs will die instantly at 3.4 or 3.5 volts (I accidentally switched a Socket 7 board to single-rail mode when troubleshooting). I sacrificed a common/spare K6-2 I had (I think a 300) to confirm this. It could be some SMD caps or resistors failing open or short or such and not the microchip itself going bad (sans any damage a shorted capacitor or resistor would cause) but the CPUs are dead as-is. (potentially a project in delidding and SMD soldering at some point, I'm keeping those dead CPUs around in any case)

I'd heard anecdotes of K6-2s dying at 2.8V or higher when overclocked, but never instant death (usually ran for at least a little while and then died), though 2.6V seems to be the max for sustainable overclocks with the 250 nm AMD CPUs. I've heard even less about overvolting Slot 1 or Socket 370 CPUs. (like Mendocino Celerons, 250 nm 2.0V rated parts at 2.8V Klamath Pentium II voltages, or the rather rare boards supporting adjustable voltages between 2.0 and 2.8V)

Also that said, I'm not sure about the 180 nm parts for either NS or AMD in this regard. Cyrix's 180 nm parts default to a rather high 2.2V already while AMD's maxed at 2.0V and ranged down to 1.6V For the K6 2+ and III+ CPUs. It's not usually recommended to go above 2.2V on 2+/III+ CPUs, but given the weirdness of NS's 2.9V rated 250 nm CPUs, I'm not sure if the 180 nm ones are similarly overvoltage tolerant and I'm not sure I'd want to risk killing one of my white-top 2.2V PR-333s to find out. (that and I'm not 100% sure all of those are 180nm, though it seems like the 4x multiplier wasn't added until the 180 nm switch)

Refreshing my memory with datasheets (and looking over some I hadn't before), Cyrix, IBM, and NSs split-rail voltage Socket 7 parts (from 350 nm all the way out to the 180 nm PR433) share the same absolute maximum ratings on the datasheets:

4.0V for VCC3 (that would be the I/O voltage)

3.3V for VCC2 (that's Vcore)

Recommended operational voltage=
2.1-2.3V for VCC3

That might be a tad conservative for the 350 nm parts, particularly the 6x86L, but is certainly generous by the time you get out to 180 nm. Though again, the node size isn't the only limiting factor of a process or overall construction of a CPU. (still I doubt 250 and 180 nm parts would tolerate 3.5V like 6x86L CPUs might, and again I was under the impression that late model 6x86s with 3.3 or 3.5V ratings used the same 350 nm manufacturing process as the Ls, so as far as the silicon goes it should've been similar)

The earlier single-rail 6x86s (and possibly even the late model 350 nm variants of those) have ratings from IBM of
4.0V max
3.15-3.6V recommended (for 3.3V rated parts)
3.4-3.6V (for 3.5V rated parts)

And Cyrix's single-rail PR-200+ at 150 MHz
4.0V max
3.15-3.7V recommended

I also haven't seen any Socket 5 or 7 boards that support over 3.6V (usually 3.52V max) so that 3.7V operating range is interesting. I haven't seen overclock-oriented socket 5 boards supporting higher voltage either, but perhaps those are just rather uncommon and/or DIY voltage mods were more common for 1995-1996 enthusiasts. (given overvolting the early socket 5 chips should've been relatively similar to doing the same for late gen Socket 3 parts: ie DX4s and 5x86s ... aside from actually running them at 5V) Plus NexGen uses 4.0V standard for their CPUs.

Cyrix and IBM datasheets:
http://datasheets.chipdb.org/Cyrix/M2/mii_433.pdf
https://datasheet.datasheetarchive.com/origin … AHI00047955.pdf
http://datasheets.chipdb.org/IBM/x86/6x86MX/mx_full.pdf
https://datasheet.datasheetarchive.com/origin … /DSA0057496.pdf
https://www.ardent-tool.com/CPU/docs/Cyrix/6x … 6MX/94329_4.pdf
https://www.ardent-tool.com/CPU/docs/IBM/6x86/ds_add.pdf
http://datasheets.chipdb.org/IBM/x86/6x86/6X_4-1.PDF

Meanwhile, Intel's 350 nm Pentium MMX has absolute maximum of
4.6V I/O
3.7V core

And the 250 nm Celerons (and presumably Tilamook MMXs) are rated for Vcore +1.0V or nominally 3.0V for absolute maximum Vcore.
So Celeron 266 to 533 PPGA
Vcc (core) = 3.0V abs max
2.0V recommended

While Coppermine celerons are (across the board)
Vcc (core) = 2.1V absolute max

Celeron:
http://datasheets.chipdb.org/Intel/x86/Celeron/24365803.pdf
https://www.dexsilicium.com/Intel_Celeron.pdf

P55C
https://datasheet.octopart.com/PENTIUM-MMX-23 … eet-7279168.pdf

But Socket 5/7 AMD CPUs seem to be consistently lower rated in terms of absolute maximum ratings.

The K5 is listed as 3.8V max for a 3.52V operational voltage CPU.

I couldn't find datasheets for the 350 nm K6s but they may be similar to the K5

And the K6-2 at 250 nm has an absolute maximum rating of just 2.6V

The 2+ at 180 nm an absolute max of 2.2V

So compared to Intel chips with 3.0V or IBM/NS manufactured Cyrix parts at 3.3V that's rather low for 250 nm (plus the 250 nm IDT Winchip 2 chips running at 3.52V).

Then there's 2.2V 180 nm which isn't unusual compared to Intel's 2.1V and that leaves NS's 180 nm process Cyrix MII CPUs in the quite unusual boat of still specifying 3.3V absolute maximum and 2.3V within the nominal operational range.

http://datasheets.chipdb.org/upload/Unzlbunzl … 2F%20AMD-K5.pdf
https://www.amd-k6.com/wp-content/uploads/201 … e_Datasheet.pdf
https://www.amd-k6.com/wp-content/uploads/201 … e_Datasheet.pdf

So there might be weird/interesting overvoltage potential for the Cyrix IBM/NS parts, or they might just tolerate the voltage and not benefit in any way (run hotter but no faster). Though given how long Cyrix/IBM/NS kept pushing CPUs into the 2.9V range, it seems like high voltage generally benefited the 6x86 family. (interesting food for thought for how the updated 5x86 core of the MediaGX would've performed in a Socket 7 variant, especially with larger 64kB Cache like the 6x86MX/MII or Winchip Family while having the nominally slower FPU clocked much higher for a given PR rating)

Many motherboards that had a large bank of jumpers to set vcore and vio where normally only one jumper at a time was to be installed could unofficially be set beyond 3.5 volts by installing multiple jumpers.

My PCCHIPS m571 could be unofficially set as high as 4 volts by installing too many jumpers to the voltage jumpers

I used that trick to overclock my k5-pr133 to pr200 :0