feipoa wrote on 2022-03-12, 14:04:

Nemo1985 wrote on 2022-03-12, 02:25:

Probably, still the higher v2.9 frequency is 250 mhz (little to none overclocking chance), while the v2.2 is 285 (or 300 for the […]
Show full quote

Probably, still the higher v2.9 frequency is 250 mhz (little to none overclocking chance), while the v2.2 is 285 (or 300 for the almost impossible to find 433gp).

That's the performance difference at 250mhz clock:
Immagine.png

The configuration was slightly different compared to the 200 mhz, but somewhat comparable.

At 200 mhz the difference (quake 320 resolution) was 1,7 fps, while at 250 mhz it was 2,2!
So apparently going up with clock increase the performance gap.

I was surprised to see the difference in performance between the 2.9 V and the 2.2 V at the same operating frequency. I don't believe I tested for this condition. I wonder what sacrifice Cyrix made in the 2.2 V design. If both were tested on the same motherboard and under the same software conditions, I don't think the error is within the margins because all tests demonstrated a performance advantage to the 2.9 V piece.

From my experience, the MII-400GP's should all do 333 MHz relatively well. At 350 MHz, some tests will fail. While I have an MII-433GP, I haven't stress tested it at 350 MHz and probably don't want to due to rarity of product. I tested a few MII-333GP 2.2V/4x chips at 300 MHz and they all did well. I don't think I tested them at 333 MHz. Note that at 333 MHz, you've only got about the gaming performance of a Pentium MMX 250 - PII266, depending on the game.

The main concern is how far can the higher voltage go. Probably not much more than 250mhz. Glad to know that the v2.2 is able to achieve at least 333mhz.

I was thinking that it may also be a revision issue, the Ibm I tested was revision F, while I don't know my v2.2 parts what revision they are since all of them are cyrix and not IBM, if this prove to be true (maybe some of the optimizations were disabled in the older revisions) the utility I used attached before gives complete output of the enabled and disabled registers.

Since you have the greatest collection of cpus of the forum and you are experienced in benchmarks, are you willing to do some testing to prove if the v2.2 version are effectively faster, please? That could clarify this matter for good.

0xCats wrote on 2022-03-12, 11:32:

Whatever performance difference you are finding here is margin of error. […]
Show full quote

Whatever performance difference you are finding here is margin of error.

I know this because the Cyrix M2 2.9V and the M2 2.2V chip are the same electrical design.
The 2.2V chip is simply manufactured on a new node (die shrink).
The fact they are not scoring the same in your results, and that you have no error-bars or margins for the test results provided means you have got a testing methodology flaw.

Not true. The 2.2V MII has worse memory latency than the earlier 2.9V version. It's pretty clear when you run cache/memory benchmarks on both a -300 and a -400 in the same board with the same RAM, same bus frequency, same multiplier, same BIOS settings, and get different results.

So just to be clear since I received the accusation that my methodology of testing is flawed, without any proof, nonetheless.

I tested some of the Cyrix\ibm cpus in my collection, I did the msdos test suite only since despite what someone says their result are quite steady (the margin of error is contained in 0,1 fps or 1 tick in doom), I don't think it's a coincidence that the number are pretty much the same with different cpus tested.
I also used 6x86ctl with -q switch to have a briefly check of the revision and enabled features (zipped as attachment), first letter identify the cpu brand (cyrix or ibm), the numbers is the pr-rating printed on the cpu, for pr333 the last letter indicates the top color s for silver, g for gold.

Test Machine:

Cyrix\ibm cpus at 200 mhz (100x2)
Aopen AX59PRO 1mb cache
128 mb memory 7ns with tightest timings
Geforce 2mx400 video card
Compact flash card

Enough with empty words and let's just show numbers.

Tested cpus:

Benchmarks:

So apparently the problem is not the voltage, there has been a performance cut between the pr300 and pr333, since the frequency and bus are the same on my testing, it's clear enough that they changed something in the cpu, and even revision doesn't seem to matter.

Feel free to prove me wrong or right with your tests and not empty accusations.

Anonymous Coward wrote on 2022-03-12, 14:15:

BitWrangler wrote on 2022-03-12, 04:28:

IBM weren't still "in" by the time they went to 0.25 micron were they?

Those silver top Cyrix MIIs were all manufactured by IBM. They're 2.2V and .18 micron. Officially they shouldn't exist, but for some reason they do.
I had never really noticed them in the past, but recently a whole bunch started showing up online.

Yah, very little seems to line up with the official what, why, where, when. ... and CPU-Z or Everest CPUID ain't sorting the sheep from the goats, they're both saying 0.35 micron for CPUs that really can't be.

feipoa wrote on 2022-03-12, 14:04:

From my experience, the MII-400GP's should all do 333 MHz relatively well. At 350 MHz, some tests will fail. While I have an MII-433GP, I haven't stress tested it at 350 MHz and probably don't want to due to rarity of product. I tested a few MII-333GP 2.2V/4x chips at 300 MHz and they all did well. I don't think I tested them at 333 MHz. Note that at 333 MHz, you've only got about the gaming performance of a Pentium MMX 250 - PII266, depending on the game.

There's a 400Mhz result in the 6x86 roundup, did you get one clocked that fast or was that extrapolated?

Nemo1985 wrote on 2022-03-12, 16:21:

So just to be clear since I received the accusation that my methodology of testing is flawed, without any proof, nonetheless. […]
Show full quote

So just to be clear since I received the accusation that my methodology of testing is flawed, without any proof, nonetheless.

I tested some of the Cyrix\ibm cpus in my collection, I did the msdos test suite only since despite what someone says their result are quite steady (the margin of error is contained in 0,1 fps or 1 tick in doom), I don't think it's a coincidence that the number are pretty much the same with different cpus tested.
I also used 6x86ctl with -q switch to have a briefly check of the revision and enabled features (zipped as attachment), first letter identify the cpu brand (cyrix or ibm), the numbers is the pr-rating printed on the cpu, for pr333 the last letter indicates the top color s for silver, g for gold.

Test Machine:

Cyrix\ibm cpus at 200 mhz (100x2)
Aopen AX59PRO 1mb cache
128 mb memory 7ns with tightest timings
Geforce 2mx400 video card
Compact flash card

Enough with empty words and let's just show numbers.

Tested cpus:
01.jpg
02.jpg

Benchmarks:
photo_2022-03-12_17-17-24.jpg

So apparently the problem is not the voltage, there has been a performance cut between the pr300 and pr333, since the frequency and bus are the same on my testing, it's clear enough that they changed something in the cpu, and even revision doesn't seem to matter.

Feel free to prove me wrong or right with your tests and not empty accusations.

That's good enough for me. The previous post showing the difference across a range of benchmarks seemed to say it wasn't run to run variation anyway.

Now as to the why is one faster. There's a question in my head about that linear burst mode some boards are meant to enable and some boards don't. I know this is the same board, but I'm wondering if there's something between the chips maybe steppings later than the BIOS code that makes it think the earlier ones are burst capable, and because it doesn't "know" the later ones, doesn't enable it for them.

If there is a open ended question here about this I can do a test between my 400gp 2.2v and 366gp 2.9v

BitWrangler wrote on 2022-03-12, 20:04:

Now as to the why is one faster. There's a question in my head about that linear burst mode some boards are meant to enable and some boards don't. I know this is the same board, but I'm wondering if there's something between the chips maybe steppings later than the BIOS code that makes it think the earlier ones are burst capable, and because it doesn't "know" the later ones, doesn't enable it for them.

Unlucky this doesn't seem the reason. With 6x86ctl those things can be enabled or disabled.
I posted the reports of all the cpus to show that the settings applied (from the motherboard) are the very same (the list of 0s and 1s) from ccr0 to ccr5.

This is the verbose reports with the bits enabled and disabled (took with pcem, just as example, do not compare with my reports):

1CCR0: 02  00000010
2          ......1.   NC1
3CCR1: 82  10000010
4          ......1.   USE_SMI
5          .....0..   SMAC
6          ...0....  *NO_LOCK
7          1.......   SM3
8CCR2: 82  10000010
9          .....0..   LOCK_NW
10          ....0...  *SUSP_HLT
11          ...0....   WPR1
12          1.......  *USE_SUSP
13CCR3: 00  00000000
14          .......0   SMI_LOCK
15          ......0.   NMI_EN
16          .....0..   LINBRST
17CCR4: c7  11000111
18          .......1  _IORT1
19          ......1.  _IORT2
20          .....1..  _IORT4
21          ...0....  *DTE_EN
22          1.......  *CPUID
23CCR5: 21  00100001
24          .......1  *WT_ALLOC
25          ...0....   LBR1
26          ..1.....   ARREN
27Far COF's in BTB are enabled.

The bits can be enabled or disabled with that utility.
Feipoa is the most expert in such matter due to the extensive tests he did in the past.

I now can't do further testing since I moved away from this motherboard and messing with something else.

Sphere478 wrote on 2022-03-12, 20:15:

If there is a open ended question here about this I can do a test between my 400gp 2.2v and 366gp 2.9v

Apparently the difference is between the pr300 and pr333 and the the voltage doesn't matter as I supposed initially, but if you could do some tests, the more the better.

Nemo1985 wrote on 2022-03-12, 20:24:

Unlucky this doesn't seem the reason. With 6x86ctl those things can be enabled or disabled. I posted the reports of all the cpus […]
Show full quote

BitWrangler wrote on 2022-03-12, 20:04:

Now as to the why is one faster. There's a question in my head about that linear burst mode some boards are meant to enable and some boards don't. I know this is the same board, but I'm wondering if there's something between the chips maybe steppings later than the BIOS code that makes it think the earlier ones are burst capable, and because it doesn't "know" the later ones, doesn't enable it for them.

Unlucky this doesn't seem the reason. With 6x86ctl those things can be enabled or disabled.
I posted the reports of all the cpus to show that the settings applied (from the motherboard) are the very same (the list of 0s and 1s) from ccr0 to ccr5.

This is the verbose reports with the bits enabled and disabled (took with pcem, just as example, do not compare with my reports):

Copy code to clipboard

1CCR0: 02  00000010
2          ......1.   NC1
3CCR1: 82  10000010
4          ......1.   USE_SMI
5          .....0..   SMAC
6          ...0....  *NO_LOCK
7          1.......   SM3
8CCR2: 82  10000010
9          .....0..   LOCK_NW
10          ....0...  *SUSP_HLT
11          ...0....   WPR1
12          1.......  *USE_SUSP
13CCR3: 00  00000000
14          .......0   SMI_LOCK
15          ......0.   NMI_EN
16          .....0..   LINBRST
17CCR4: c7  11000111
18          .......1  _IORT1
19          ......1.  _IORT2
20          .....1..  _IORT4
21          ...0....  *DTE_EN
22          1.......  *CPUID
23CCR5: 21  00100001
24          .......1  *WT_ALLOC
25          ...0....   LBR1
26          ..1.....   ARREN
27Far COF's in BTB are enabled.

The bits can be enabled or disabled with that utility.
Feipoa is the most expert in such matter due to the extensive tests he did in the past.

I now can't do further testing since I moved away from this motherboard and messing with something else.

Sphere478 wrote on 2022-03-12, 20:15:

If there is a open ended question here about this I can do a test between my 400gp 2.2v and 366gp 2.9v

Apparently the difference is between the pr300 and pr333 and the the voltage doesn't matter as I supposed initially, but if you could do some tests, the more the better.

The 300 and 333 came in 2.2 and 2.9v flavors

Which specific ones are in question?

I assume the tests are being done at the same mhz and settings on mobo yes? (Except voltage if relivant)

In any case, I have 2.9v 300 and 366 and 2.2v 400 I can test.

I also have some lower speed chips, 150, 166, 200 (see my signature)

Let me know what you want.

Will super pi runs suffice? Or what benches do you need?

Sphere478 wrote on 2022-03-12, 20:30:

The 300 and 333 came in 2.2 and 2.9v flavors […]
Show full quote

The 300 and 333 came in 2.2 and 2.9v flavors

Which specific ones are in question?

I assume the tests are being done at the same mhz and settings on mobo yes? (Except voltage if relivant)

In any case, I have 2.9v 300 and 366 and 2.2v 400 I can test.

I also have some lower speed chips, 150, 166, 200 (see my signature)

Let me know what you want.

Will super pi runs suffice? Or what benches do you need?

Yes, very same setting, the only thing I did was switch the cpu and voltage when needed.
The performance difference was between pr300 and pr333, doesn't matter the voltage. If you read my table the 300gp (v2.9) has better performance compared to the 333gp (both v2.9 and v2.2) and pr400.

I would advice to do dos benchmarks since it is "single task" there can't be any background stuff that may give different results.
I use a modified version of phil's benchmark pack.
You could test doom and quake, or use any other benchmark I did if you can.

Every Cyrix chip was a unicorn

Cyrix and IBM had to modify a lot of little things to break the 240mhz barrier some chips were many many layers thick to make up for old process limitations.

There were “tests” here historically that showed the faster Cyrix cpus had lower per clock cycle performance the pr433’s have the lowest performance per clock but considering the higher mhz they still end up faster overall anyway.

Funny thing is despite IBM designing and selling the highest clock speed Cyrix chips before VIA, IBM never sold a chip rated past pr333

After IBM was out of the picture Cyrix started selling lower clocked 250mhz super 7 chips as pr350/366 while IBMs 266mhz chips were marketed as pr333 go figure

So somewhat literally, I've been digging for gold all evening. Going through junk drawers trying to locate Cyrixes. Found A64 X2 6400, Mini PCI Wifi card, 6 floppy disk labels, 30+ writing implements, 5lbs of keys that who knows what they fit any more, but thus far, no 6x86 🙁 .

Innnterestingly however, I came upon some scribbles from over a decade ago, when I must have been feeling the wolf at the door financially, I assessed the gold value in my CPU holdings. So what is interesting is that the gold recovery source I found at the time said there was 0.21g in the Cyrix CPU and 0.25g in the IBM CPU... so I don't know if that bodes any better for overclocking them or is a thicker coat on the pins for less corrosion comebacks or something. On that page of scribblings though, I have 4 IBM 6x86 listed, which are accounted for, only the 266 is very interesting for purposes of this thread, and 5 Cyrix CPU... now I am not sure which CPU were counted, I currently know where 2 are, but one of them might not have been counted at that time... there may be a stray in a system I counted then... so somewhere between 2 and 4 Cyrixes are hiding in the house somewhere. Remember kiddies, if you get crap out to show someone or something, put it back afterwards, otherwise you'll drive yourself nuts 🤣

Oh man, it’s like a treasure hunt!

Okay, I’m ready to do some testing here..

I don’t have dos set up on this and it would take a while to get there.

I could do it on my tyan but then would be limited to pci and 66fsb.

I thought it would be better to run the test on the fastest system I have

And to account for background task interference I will do more than one run to confirm consistency.

2.9v 300gp at 2.5x100
Spi 512k first run 4m30.800s
Spi 512k second run 4m30.149s
Crash on 3dm 2000
No second attempt, moving on to 366gp

2.9v 366gp 2.5x100
Spi 512k first run 4m38.861s
Spi 512k second run 4m37.009s
Reboot
Spi 512k third run 4m42.556s (ignoring)
Spi 512k fourth run 4m38.721s

2.2v 400gp 2.5x100
Spi 512k first run 4m55.575s (ignoring)
Spi512k second run 4m42.376s (ignoring)
Spi512k third run 4m37.299
Spi 512k fourth run 4m36.277s
Spi 512k fifth run 4m35.186s

If I get rid of some outlier results it seems were probably caused by background processes, there does seem to be a pattern here, one that matches what was claimed earlier in the thread.

Sphere478 wrote on 2022-03-13, 07:08:

I don’t have dos set up on this and it would take a while to get there.

No need to have a proper dos machine, press f8 just after the post and choose the option to boot on dos instead of windows.

I have a multi boot menu to be able to use dos or windows 98se whatever I need.

Nemo1985 wrote on 2022-03-13, 12:10:

Sphere478 wrote on 2022-03-13, 07:08:

I don’t have dos set up on this and it would take a while to get there.

No need to have a proper dos machine, press f8 just after the post and choose the option to boot on dos instead of windows.

I have a multi boot menu to be able to use dos or windows 98se whatever I need.

Winxp

Win9x uses ctrl key to get to boot menu if I recall.

I don't think I ever managed to push my 400GP 2.2V above 300MHz (3x100), unfortunately 🙁.

Sphere478 wrote on 2022-03-13, 12:50:

Winxp […]
Show full quote

Nemo1985 wrote on 2022-03-13, 12:10:

Sphere478 wrote on 2022-03-13, 07:08:

I don’t have dos set up on this and it would take a while to get there.

No need to have a proper dos machine, press f8 just after the post and choose the option to boot on dos instead of windows.

I have a multi boot menu to be able to use dos or windows 98se whatever I need.

Winxp

Win9x uses ctrl key to get to boot menu if I recall.

nothing to do then, windows xp purged any dos trace :\

Ok I've tested my fastest Cyrix so far.

Cyrix PR366 (2.5x100MHz). BIOS reports 333 but should be 366 (maybe bios needs updating).
DFI MVP3 Motherboard w/1mb l2 cache.
Voodoo 3 3000 AGP 16mb.
128mb SDRAM 100MHz
Windows 98se
No sound card (important for q3 result)

3dmark2000 - 505 3dmarks
quake3 demo - timedemo 1 - default 640x480 - 19.8fps

mwdmeyer wrote on 2022-03-14, 11:58:

Ok I've tested my fastest Cyrix so far. […]
Show full quote

Ok I've tested my fastest Cyrix so far.

Cyrix PR366 (2.5x100MHz). BIOS reports 333 but should be 366 (maybe bios needs updating).
DFI MVP3 Motherboard w/1mb l2 cache.
Voodoo 3 3000 AGP 16mb.
128mb SDRAM 100MHz
Windows 98se
No sound card (important for q3 result)

3dmark2000 - 505 3dmarks
quake3 demo - timedemo 1 - default 640x480 - 19.8fps

There was some talk in the k6 aiming for the stars thread about timings for the chipset. Someone managed to copy the timings from the p5a to a gigabyte mobo and my freeway design seems to boast similar performance on a via mobo. Basically a lot of the mobos out there are lacking behind these mobos in performance. It also may be related to wb/wt pin on cpu.

More looking into this may be in order,

Perhaps your result though has more to do with the voodoo 3 though. Try a radeon 9800 pro or a geforce 2/3

Oh yeah lack of T&L holds it back a lot, but that is ok 😀