Cyrix M2s

[RoadWarrior]

Hi guys,

I have overclocked a few socket 7 chips before, but just got my hands on a Cyrix M2 (for *real* cheap) that I am gonna have in the kids PC. I know cyrix chips are dreadful overclockers, but I'd still like to find any info that is out there on them. I have an MII 366GP 2.9V core 3.3V i/o 250Mhz actual. The board I have to work with is an oldish Abit SM5A. it can do 2.5v-2.9v and 3.2v-3.52 core and i/o.

I would like to get it to the PR400 level, which is officially 95Mhz x 3 I think. It is multiplier flexible, I only have up to 3.5 and 83Mhz. I have the board running great at 83Mhz, no problems with a P133 oc 166 @2x83.

So, does this CPU tend to oc better when you lower the core volts as some are known to do? I was thinking of trying 2.8v core and upping the i/o to 3.38 from 3.3 to try and keep it crisp.

Also does lapping affect the oc ability of these chips much at all? I was thinking of lapping it and making a square aluminum shim to fill in around the cap of the chip for better heat transfer.

I have used a try it and see approach with pretty much everything else up to now, but as we all know, Cyrixes ain't so robust.

This chip is used, it was a pull from a major OEM machine I think, does that auger well for oc ability at all? (Presuming cyrix wanted to impress major OEMs and supplied them with the best parts)

I guess you are wondering why I am bothering for the kids machine. Well I'd really like it to say P400+ at boot, because we have a K6@300 and a K6-2@450 (only showing 400 in BIOS) and would like them to be conviced they got an equal machine. Also every little helps, would like it to have a chance of running some 3D games okay for some 3 player link ups, want to try and put it just inside the PII-233 required game bracket.

thanks all for any tips,

Road Warrior

[skywalker[TSG]]

man im sorry to say this but

GET OUT OF THERE ITS GONA BLOW !!

but seriously i would not recomend that you try to overclock it

those cyrex cpu´s generatre too much heat for their own good

[RoadWarrior]

Thank you for your most elucidating and eloquent insight. If it was easy I wouldn't be asking.

[Peter M]

Erm, no overclocking headroom on 2.9V MII parts, sorry. Especially on the older parts made by IBM or NatSemi. The current VIA made ones (labelled MIIv) might have a little headroom, but don't expect any miracles. Chances are that your old board's voltage regulators are maxed out at stock settings already, those old MIIs are HUNGRY (the MIIv not so).

Forget about lapping them too, their heat spreader surface is machined to be perfectly flat from the factory already.

Run it 3.0x 83 MHz instead of its native 2.5x 100 MHz for the same 250 MHz (and "333" performance rating instead of "366"). And cool it really well, apply thermal paste and use a reliable and thermally good fan/heatsink combination.

"400" performance rating would be 95 MHz 3.0x - there are 2.9V MII parts at that speed, but I'm afraid your board won't be able to do that. 83 MHz is stretching the old SiS chipset already.

If you want it to play 3D games OK, then you'd need to use a graphics card with its own T&L computing, Geforce or Radeon series.

Regards, Peter

[RoadWarrior]

Thanks Peter,

I know cyrix did a lower bus speed 250 actual Mhz one than this one, but I thought that was .35 micron technology, whereas this version is tweaked and reduced to .25 micron. At least that's all the info I kind find. I think this is the only speed grade at .25 micron, because they sampled out the PR400s and PR433s and nothing more, before shrinking the die again for the core that VIA is now using. So since the PR is based on CPU32 and CPU16 integer benchmarks in winstone I think, it should be PR366 whatever bus and multi gets me to 250. Although I understand that performance in other areas is hurt by the bus drop.

It's an Intel VX chipset on this board. Have read of people having K6 2.9v CPUs on it (200,233) oc'ed to the 290 mark, so hopefully the regs will cope.

I found in another forum a guy saying he had got 300Mhx @3x100 on this chip and was running it at 2.6V, complaining that Compaq was "trying to kill it" how it was set at 2.9. This kind of makes it sound like the voltage should have been reduced when they went from .35 to .25 micron and they didn't for compatibility reasons or whatever.

I think this one came out of a compaq, was yarning with the grrrl I bought it off for a while though, so kinda hard to remember the whole conversation.

Also I should have the 3.5 multiplier on this chip, and I have 3.5x on the board. So 3.5x83 should give 290ish, not 3x95, I know what SiS chipset you are talking about I think, the one that claims 100 but can't.

Where's all the good socket 7 oc stuff and tech details that used to be at places like, brotherhood of the cpu, alternative cpu, and altx86? they have all died :-(

If this is a .25 micron part, well intel uses that for 400Mhz or so, so hopefully there is a bit of life in the thing, .35 micron was maxing out at 300 of course. No reason why it shouldn't be happier with a bit of a lower core voltage if that is the case either.

Anyway, at least I have the hope that since it came from the same source as one reported to do 3x100 it might give me 290.

Road Warrior

[Peter M]

Erm, no, not quite.

M2 core history.

First public one was called 6x86MX, being M2 core revision 3 on .35 micron, 133 and 150 MHz. No major change until revision 8, which is a bit faster per MHz and was renamed to MII. This yielded up to 262 MHz, was still .35ish micron, consumed a lot of power and was hard to cool. These were the last ones IBM made, some of them were packaged with flip-chip technology (those with the mushroom style heat spreaders that stand off the package) to make it to 100 MHz bus.

Then came the switch to NatSemi manufacturing, with not much actual output ever seen. During that time, the chip was revised again (revisions 13h and up) to have 4.0x multiplier, and again be a little faster per clock.

What we now have on the shelves is VIA made "MIIv" stuff, most of them being slightly shrunk .30 micron-ish revision 14h stuff (up to 285 MHz "400" speed grade) that still is 2.9V but has about 20 percent lower power consumption than the IBM made ones, and revision 20h or higher .25 micron 2.2V stuff ("400" and "433") that consume half as much and are reported to have ridiculously much overclocking headroom - the 3.0x 100 MHz "433" chip has been reported to run 4.0x 100 without any extra effort ...

VIA apparently doesn't make the "366" speed grade, so what you have there should be one of those IBM made flip-chip powermongers.

With Intel's VX chipset, your best stable bet would be to run it 3.5x 66 MHz actually, for 233 MHz and "300" performance. 75 or 83 MHz overclocks your PCI bus and IDE HDD, with data corruption very likely to happen.

To know exactly, check the CPU revision. It's software readable - the usual suspects of diagnostics software packages will help you.

Regards, Peter

PS: I was talking about the SiS 5581/82/97/98 75 MHz PCI chipset that later made it to 83 MHz officially, not the not-quite-super-7 "5591" 83 MHz AGP chipset.

PPS: Running an MIIv-333 3.0x 83 MHz (rev. 14h 2.9V part) on a PC-Chips M560 as we speak http://www.sysopt.com/forum/smile.gif

[This message has been edited by Peter Missel (edited 03-28-2001).]

[RoadWarrior]

Chip geek says the latest M2s from Via are .18 micron, and lists the 366 at .25 micron the rest at .35 micron. <A HREF="http://www.sandpile.org" TARGET=_blank>www.sandpile.org</A> seems not to have been updated since via took over, it lists .35 and .25 fabs without giving what chip is which. Power consumption figures point out the Nat semi model must've been smaller than .35 and not as small as the planned 400 and 433.....
250 MHz: 27.6 W (Cyrix), 23.9 W (IBM Wire Bond)
250 MHz: 27.9 W (IBM C4 Flip-Chip)
250 MHz: 23.3 W (NS 2.9V)
250 MHz: 8.8 W (NS 2.2V)
285 MHz: 12.5 W (NS 2.2V)
300 MHz: 13.7 W (NS 2.2V)

When you say "mushroom style" heat spreaders do you mean like on a K6, 'coz it isn't one of those. It's a gold square on top apart from lettering looks exactly like the ones on the VIA site.

I could ID the core revision etc with wcpuid or something, but I don't have a listing of the cores anywhere to compare to. From the pics at sandpile, it looks somewhere between 04h and 08h, or later than the 08h if they went back to 04h style package again. this is the muchroom one from IBM? <A HREF="http://www.sandpile.org/impl/pics/cyrix/m2/08_top.jpg" TARGET=_blank>http://www.sandpile.org/impl/pics/cyrix/m2/08_top.jpg</A>
it's not that.

serial number on the bottom is v7sp891obd and it also says
(c)(m)1995-1998
CYRIX
above that.

The IBM ones seem to have (C)(M)1995-1997 on

My Vx board is running 83x2 as we speak with PCI cards that can cope with it. Overclocked P133 to P166. More stable than the PR166+ I had in another board in this box. However, cyrix support on that board was marginal and I suspect bus reflection issues with the L2 cache socket.

So thanks for your input, is there a list of the core revisions somewhere?

Oh by the way , what's good for finding the revision etc from a DOS prompt? could really use something I can get on a boot floppy, because it is going to be a couple of weeks before I get the board out of here etc, I can test it on an Asus T2P4 I have out of a system at the moment.

Road Warrior

[This message has been edited by RoadWarrior (edited 03-28-2001).]

[Peter M]

So, more history. The pic at sandpile shows an IBM made, Cyrix-packaged flip-chip one, must be rev. 08h then. If you'd look at one of those from the side, you'd see that there is a gap between the ceramic body and the heat spreader. The latter is attached directly to the silicon die, and the ceramic body is as thin as a K6.

Normal wire-bond packaged 6x86MX/MII have a thick and heavy ceramic body, the heat spreader attached to it with no gap.

Regarding the revisions, the light shrink from .35 to .30ish process (somewhere between revision 08h and 13h) came when Cyrix was at NatSemi, hence the lower power consumption on the 2.9V parts. The 2.2V parts are on a .25 micron or smaller process.

You can easily get the Cyrix CPU core revision by reading its DIR1 configuration register - in DOS, start debug.exe and type

o 22 FF
i 23

and there you have it.

Short summary of revisions, collected from the various documents:

03-07 6x86MX 2.9V
08 MII 2.9V (faster per-clock operation)
13-1F MII 2.9 post-shrink (supports 4.0x multiplier, faster again)
20-3F MII desktop 2.9V, mobile 2.2V
40- MII 2.2V

Someone with a 2.2V part please fill us in with their revision number ...

Regards, Peter

[NRG]

I built a computer for my mother so she can use the internet in her house and i scavenged around my workshop and found a cpu when i dusted it off...lol ir said it was an IBM 6x86 pr233 now this little IBM cpu is running at 300mhz no problem because i used a standard athlon hsf it's very stable too...so if you use good cooling i can't see why you should not be able to do it...but if it goes pop! it is at your own risk cause i don't usually play around with chips which are that unpredictable but go have a look on the overclockin sites to see if anybody has achieved what you want to clock it to...


NRG

[RoadWarrior]

Thanks for the meaty stuff Peter,

Now, have it in the Asus board, haven't a clue what speed I had it running at because I think the current BIOS in that maxes out at PR266 for detection. The manual says cyrix cpus are set at 3X for 4X speed and 1.5X for 3.5X speed. Anyhow, set it on 66x4 for the first go and it booted, set it as 75x4 and it booted again, could feel the heatsink warming up a bit though, nothing bad, just enough to feel. Wondered about the PR266 display and reset the multi to 3.5? and it was still showing PR266, so I guess the 3.5 multiplier definitely isn't 1.5x on this board. Can't say much more, this is just an old DOS hdd with a few simple diags on it. It "bent the needle" on two benchmarks, snooper and sysinfo (old dos version), hit the stops I think.

Anyway, did the debug thing and it gave me "14" so this must be revision 14h or did it give me that in decimal?

So reading all what you told me, I think I have a little bit of "headroom" there if it is a National Semi part, post shrink .30 micron. I may have had it running at 300 already, I may not have.

I have a Thermalloy high flow biggish heatsink on it at the minute, neat things, they come with a patch of heatsink transfer compound already on. I bought this a while ago, but I think it was one of those listed okay for socket 370 high temp applications. So hopefully it's got enough cooling "grunt" on it to fail gracefully (i.e. crash) instead of catastrophically (i.e. die).

Thanks very much for your help Peter.

I am going to go on a leeching spree now to see if I can find any good, small diags and benchmarks for DOS that recognise this thing, to see actually how fast I am running it.

hmmmm, odd how the links screw up when you edit a post, they were fine first time I posted on that last one of mine.

I was unaware that the mulipliers for these may be different than other CPUs, so I am kind of wondering how my ABIT board with the software menu is going to handle this CPU. Whether the multipliers are pre-translated for Cyrix or whether I have to translate them. Hmmmm, if I have a 4x multiplier, what's the PR of 333Mhz??? ;-)

Road Warrior

[RoadWarrior]

Hokey dokey,
here's what's happening so far on the ASUS board 1.5x=3x 2x=2.5x 2.5x=2.5x and 3x=3x so so far I have only had it up to 3x83 for 250Mhz. Found a nice DOS util called HwInfo at tweakfiles which is telling me the actual speed.

Does the M2 (Or THIS M2 rev 14h) have the extended multiplier jumper on pin BF2 like the K6 does? If I jumper it, how does it modify my clock multiplier settings? Like on a K6 it modifies 2x-4x 2.5-4.5 3-5 3.5-5.5

Since all the other multipliers are a bit crazy I don't want to jumper it then discover it sets it to 6x for all of 5 seconds.

Oh, it seems to be running stable on 2.7 volts by the way. Boots DOS 6.22 runs doom is all the testing so far.

Road Warrior

[This message has been edited by RoadWarrior (edited 03-29-2001).]

[RoadWarrior]

grounded BF2 but it didn't work, thought I'd try it since I found a 50mhz clock that should've been safe. Wouldn't boot, tried all regular jumper combos in conjunction with the grounded BF2 pin. Didn't kill it.

[RoadWarrior]

Hmmmm been reading ... http://www.via.com.tw/pdf/cyrix/102ap.pdf
this only shows clock multipliers up to 3.5 available. BF0 and BF1 pins are called clkmul0 and clkmul1 and BF2 is nc/reserved. This is odd, I should be getting 3.5x and I'm not. Also 2x only works with all the jumpers pulled off.

[Peter M]

Hi again,

Debug.exe works in hex, so yours is rev. 14h. Thus it uses the BF2 pin, however only 4.0x multiplier is documented to exist, no higher.

btw, all the multiplying is internal to the CPUs. There's not really a point in saying "this mainboard supports those multipliers", since all the board does is provide jumpers to feed a setting number into the CPU - the CPU then does whatever it was designed to do at that setting.

If you want to find the actual multiplier your Cyrix M2 is currently running at, you can read the DIR0 register from DOS debug.exe ...

o 22 FE
i 23

The first digit is always 5 for M2 core, the second digit encodes the multiplier. So determine your actual speed from your board's bus speed and the multiplier read from the CPU. Table of DIR0 value vs. multiplier (all hex):

50 or 58: 1x
51 or 59: 2x
52 or 5A: 2.5x
53 or 5B: 3x
54 or 5C: 3.5x
55 or 5D: 4x

1x, you might ask, how do you do that? Well, the original Pentium, as well as the Cyrix M1 and M2, have an undocumented Deturbo pin Y33. Ground that, and you get 1x on those processors. My PC-Chips M560 mainboard has an equally undocumented turbo switch connector that kicks my M2 back to 83 MHz from its jumpered 250 ...

333/83 or 333/95 would be "466" according to the Cyrix M2 BIOS writer's guide. 292/83 would be "400".

Regards, Peter

[Peter M]

oh, erm, the BIOS writer's guide is the AP118 document. However, the VIA web site has an old revision of that document online.

And, regarding the multipliers, old boards sometimes make the stupid mistake of not being able to pull a BF pin UP (under the assumption that the CPU internally does that). Now while that was correct for the original Pentium, defaulting to 1.5x where BF0 and BF1 both are High "1" level, it isn't for the P-MMX and M2. These default internally to 2.0x "01" combination. On those old boards, you won't get anything but 2.0x and 2.5x (and 4.0x and 4.5x if you use the BF2 pin for a CPU that cares about it).

Regards, Peter

[RoadWarrior]

Many thanks again Peter for the detailed info, I tried that debug thing with the jumpers set for 3.5x and it reported 3x like HWinfo was doing, so I assume I am having a problem with the pull ups or pull downs.

When I tried BF2 I dropped a thin strand of wire in the socket to jumper it to a nearby ground, pulling it low. This works great with a K6 on the board. However as I mentioned it did not boot the M2 like this, does the M2 need this pulled high? or was it possibly because of an invalid combination with the other jumpers/BF pins because they are getting pulled wrong?

By the way, this chart of Cyrix CPUs seems to indicate the 366GP is .25 micron.... http://users.erols.com/chare/elec.htm
very very good resource at these pages just found it main page here.... http://users.erols.com/chare/main.htm

Going to sit down with the socket pinout now and find out how to jumper the BF pins direct, hopefully there's grounds and vccs near enough that I don't have to solder, just drop in wires. Hmmmm until I just started with this, it never occurred to me that I could maybe get a 2.5x intel classic pentium into a socket 5 with some socket wiring chicanery.

I began by using the info here.... http://www1.jump.net/~lcs/kalle/adding.htm
unfortunately the cyrix links off Kalle's site are broken.

Hmmmm, possibly I am not wise to spend so much time tweaking this thing on the Asus T2P4 board, since I am definitely going to put it in the Abit with the soft menu, which is going to be a whole different kettle of fish probably.

Road Warrior

[RoadWarrior]

Okay, **NOW** it's overclocked!

Running 3.5x83 at last, hooked bf0 and bf1 into a vcc3 hole (3.3v) with a couple of strands of wire. Booted and played doom on 2.9v, booted and played doom on 2.7v, not getting too warm at all, I'd say around 30-35c at the side of the sink. This is not what you might call extensive testing though.

Hwinfo now thinks it's a MII PR433, odd benchmarks in that, don't understand them, CPU 16 and FPU 16 bit are scoring around the PII-400 mark, whereas the CPU 32 and MMX are scoring around the PII-233 mark. Now what seems odd is that the FPU is so good and the 32Bit is so bad. The 32 bit integer test is meant to be P-Pro optimised, and I thought from the outset 6x86 had P-Pro optimisations. Anyway, seems to be a "feature" of the benchmark, because it seemed inline-ish with the other 6x86 scores they had.

Guess I made my PII-233 minimum though.

Next up, let's nail that 4x multiplier down somehow.....

Road Warrior

[RoadWarrior]

Found an old hdd with win 95a on and plugged it in for a test, @3.5x83 it hung on 2.7v, runs fine on 2.8v, downloaded "CPU stability test" and ran the 5 minute benchmark which warmed it up some, scored 124 which was marginally better than someones PII-300 result.

Still no luck with 4x, beginning to think that BF2 isn't working, unless there's a trick to it, would think the default is unconnected on the board or pulled high.

[This message has been edited by RoadWarrior (edited 03-31-2001).]

[RoadWarrior]

Took the M2 off the board and put the K6 back on it and started playing around, there's definite problems with the multiplier pins. Either due to manufacturers not handling all cases of pull up or pull down on the BF0 pin, or I've got a jumper go flaky from too much fiddling. Intel describes a problem like the former here... http://channel.intel.com/business/ibp/processor/techpage.htm#Bus Fraction

[Peter M]

Told you about the BF pins ... a proper mainboard should be able to pull them UP or DOWN.

Some can only pull DOWN or leave open, relying on the CPU to pull up by default.

Now Pentium MMX, K6 and M1/M2 pull BF0 DOWN by default, with such a stupid old board leaving you with no means to pull it back up.

I can't see why your M2 doesn't do 4.0x - you should have BF0 and BF1 in the position for 2.0x and then pull BF2 low to get to 4.0x. BF2 is pin Y35.

btw, the M2 core does have PentiumPro instruction set extensions, but code arranged to be optimized for how the PPro works will not be optimally fast on the M2.

... and finally, you should use resistors, 1KOhm typical, not wires, to pull pins up or down - just to get some current limitation. A resistor soldered to the socket pins on the back side of the board works best.

Regards, Peter

[RoadWarrior]

Yes I think the M2 probably does not like something about me using wire links instead of proper pullups or pulldowns. Is much easier to drop a tiny loop of wire down the pins though than do a complete mod.

Anyway, decided to do a quick lobotomy on this PC with the Abit SM5 in it, managed to set it at 3.5x83, and it booted win 95, however, after running SiSoft Sandra 99 benchmarks it crashed explorer. It did this on 2.7-2.9v and 3.3-3.45V io. On the ASUS at 3.5x83 it ran a stability test all night with no errors, didn't run Sandra on it in that though. So for the moment I have it set back to 3x83. Going to try a BIOS flash on this to see if it improves. Gives an error on boot for CPU speed, detects as PR300.

Flashed the Asus with the K6 now on it, from v205 to v207 BIOS and it now gives BIOS checksum errors everytime I try to set the CPU past 2.3V, this is extremely annoying. Was getting it near stable but warm at 333, now I have to drop back to 300. Don't know what will happen if I put the M2 back on here now. Oddly though, the BF pins started behaving better after the flash. Somehow the new BIOS seems to affecting current or voltage handling somewhere.

I'm trying to figure what to try for stability at 3.5x83 on the Abit board now. CPU is not heating much, probably a voltage thing, or is not as stable at 83 with a faster CPU as it was with the P-133oc166. Trying to figure out if I should try and find 4x on this board for 4x75 for 300, wondering if the extra 50Mhz over the now rock stable 250 is compensation enough for the bus speed loss. Possibly serious experimentation will have to wait until the board is out of this machine, because although it is all soft setup for the CPU there are still some jumpers onboard to play with.

Hmmmm possibly the voltage regs overheated and caused the errors, because there is this huge long soundcard, (Boca sound expression se14rs, never buy one, I will cheerfully break this one over my knee when it is replaced) full slot length between the CPU and the voltage regs that completely blocks airflow from the CPU fan washing over the regs heatsinks.

Thanks for the support Peter.

Road Warrior

[RoadWarrior]

Oh sorry, it crashed explorer once at 2.7v before I went to 2.8 or 2.9 AFTER that the crash was in GDI.EXE when I tried to access anything in the start menu. Now, I noticed that this CPU hyperaccellerated the graphics card in benchmarks on the Asus board. However that board had an S3 card with 50ns EDO RAM in, this has a trident card with 60ns EDO. I am wondering if it is possible to determine whether it is a problem with CPU speed+overclocked PCI+graphics, or just a CPU speed issue.

Road Warrior

[RoadWarrior]

Overclocked again at 3.5x83, stable this time, I hope, went back and fiddled the memory timings some more, and now they are really on the ragged edge. The SIMMs I have in this however won't remain in it, the SIMMs I had in the ASUS board are going to be used, which offers hope that it will push it a little further onto the stability platform.

Memory access timing is AWFUL now, wondering if it might be better off at 4x75. Proud of the CPU so far, seems to be doing great, and the problems are motherboard and memory related.

[araaraara]

[Winchip2A]: Gggrrrrrrrrrrr!