Awesome work! I separated benchmark part with Quake, here it can be noticed that all AMD K6 processors on the Gigabyte motherboard work slower than on DFI and such a difference (~3 FPS avg.) cannot be justified by the cache size alone. You can notice DFI's cache size advantage with the rest of the processors, except for the P5 family, there is something unusual going on here, maybe the BIOS settings are to blame? I think the inferior performance of K6 processors in the Gigabyte motherboard is due to the disabled "AMD K6 write allocate" option in the BIOS settings. This option is usually disabled by default and without it, K6 performance becomes rather abysmal.

I just did a quick bench run on GA-5AA motherboard (AT version of GA-5AX)

Quake 1.06 320x200 under DOS, no sound, AMD K6-2 @ 500MHz, CL2 @125MHz, GF2 MX-400 AGP, TAG integrated, 512K L2:

K6 write alloc. - DISABLED = 80.1 FPS
K6 write alloc. - ENABLED = 92.8 FPS

Grosvince wrote on 2022-10-25, 20:08:

Sometimes K6-2 is better than the K6-III, but why ?

Cache latency I guess. And different latency to DRAM then.
You may compare K6-2 and K6-3 with the cache bench of dosbench suite. Perhaps you can see a difference at 200 MHz.

Thank you guys !

I've notices the K6 wasn't at their best on the Gigabyte. I'll check for the write allocate features in the setup to check.

Dis you notice the best CPU for Doom is the mP6 ? This CPU surprise me on the DFI. But it's handicaped by its tiny L1 cache, and on this test working @66 fsb is below the manufacturer specs (should run at 2*100).

Bye

Hint, compare a cyrix 2.9v 366, a cyrix 200, and a 2.2v cyrix 😉

A test between pm6 and idragon would be cool to see if it was just a die shrink.

Tillamook would be cool, again a die shrink mostly so far as I know.

Sphere478 wrote on 2022-10-26, 06:50:

Hint, compare a cyrix 2.9v 366, a cyrix 200, and a 2.2v cyrix 😉

A test between pm6 and idragon would be cool to see if it was just a die shrink.

Tillamook would be cool, again a die shrink mostly so far as I know.

Hi,

Thank you for your answer.

Can you explain a little more about the 2.2v Cyrix ?

mP6 and iDragon are the same, the part number is the same, it's just the marking on the CPU. All softwares like CPU-Z vintage report it as a mP6 Lynx. mP6 Kirin (250 nm) are incredbly hard to get however.

Tillamook, same, hard to get. I understood Tilamook CPUs don't work well with L2 cache. But it could be interesting to test.

Regards
Vincent

There were 2.0 and 2.8v rise chips from my research. IDragon I’m pretty sure is the 2.0v prob a die shrink.

The 2.2v chips clock much higher but have less performance per clock. With the extra clock it makes them faster though at max clock.

Rise cpu voltage and multiplier settings

Cyrix/ST/IBM Aiming for the stars!

Hello all,

Concerning the Gigabyte, write allocation was enable in the setup. Cache speed graphics made by speedsys clearly show this feature on K6 processors.

Yesterday evening I spend time in my attic to find my MSI 5169. It shares the same chipset with the Gigabyte (ALi Aladin V), hence the same cache size (and same chip of Sram). I've spend one hour more to find enough jumpers to configure it !

Results are pretty differents than with the Gigabyte, and altough the 5169 cannot handle the K6-2+ nor the Rise iDragon (this two CPU are recognized as "unknow" or "Rise unknow" and run extremely slow, speedsys says they score like 486 DX2), the results are on par with the DFI.

So, I wonder what can be wrong with the GA5-AX. The only know issue with this board is the PS2 mouse connector broken (I see a fuse near it, maybe it's the problem ?)

Bye
Vincent

This should help

Re: Diy modding support for k6+And 120gb hard drives into bios roms

http://web.inter.nl.net/hcc/J.Steunebrink/k6plus.htm

If you are actively looking for a non-CXT the K6-2-333 is your best bet. They didn't scale well to 400, hence the fix, so 99% of them are parts in the 300s. There was thought that some of them binned just high enough to be an early small quantity of 400s, but haven't come across any that high myself. I think though they are basically core shrunk K6 with 3DNow bolted on, no different in performance in anything that doesn't utilise 3DNow.

Sphere478 wrote on 2022-10-26, 15:47:

The 2.2v chips clock much higher but have less performance per clock. With the extra clock it makes them faster though at max clock.

Huh? I suppose this is regarding the Cyrix MII 2.2v parts? Are you implying that the 2.2V part regresses performance over the 2.8V part? Why would that be the case?

Garrett W wrote on 2022-10-27, 18:00:

Sphere478 wrote on 2022-10-26, 15:47:

The 2.2v chips clock much higher but have less performance per clock. With the extra clock it makes them faster though at max clock.

Huh? I suppose this is regarding the Cyrix MII 2.2v parts? Are you implying that the 2.2V part regresses performance over the 2.8V part? Why would that be the case?

And at the same mhz the 2.2 V chips are slower yes

But the 2.2 V chips are better you want them because they clock a lot higher

I think they increase the length of the pipeline

I've thought the 2.2V parts are just die shrunk cores. Your claims sounds a bit far fetched to me, do you have any links or sources to back them up?

See link above^ I posted it already. 😀

It’s complicated..

Details on what exactly happened with cores and processes with cyrix/ibm/st chips in socket 7 can be kinda contradictory from sources online, but I am pretty sure you are correct about the 2.2v being a smaller process. But they also made architectural changes it seems.

Edit: here is link: Cyrix/ST/IBM Aiming for the stars!

Some people I have talked to seem to think that the changes may have even started before they went to 2.2v as there can be differences noted between mII chips of 2.9v as well. Which you may be able to infer from my test results in that thread. So the change may not have happened at the same time as the shrink which may make what you say of it being only a shrink, technically correct. Though more data is needed to confirm Or a more reliable source on this than I can find googling.

Like I said, there is some mystery it seems as to what exactly is going on with cyrix’s lineup of that time period. As accounts vary. But best guess is that they started adding pipeline stages to get higher clocks at the expense of performance per clock. That is just a guess though.

Hi all,

I've run another batch of test, focusing on low end Socket5/7 processors, using the DFI K6BV3+.

Entering Cyrix 6x86 (first version, the one you can fry eggs on it), K5 SSA/5 and K5 5K86 (same frequency but not same PR). I've run a Pentium P54C@100 and @133 to compare to the K5s, unfortunately I don't have a 100 MHz Cyrix, and the K6 runs @233 with multiplier @1.5 (like the MMX).

But results are interesting. Especially the K6 "5K86" PR133 compared to the pentium. As always, Pentium and K6 are the most versatile CPUs, altough Pentium are unbeatable when talking about Quake performance.

Bye,
Vincent

wow great post ..
as for the cyrix mII, this disproves the usual rhetoric that they are useless because of weak FPU..
look at the 200 tests:
only 1 fps behind the k6's (in 640x400 quake mode) and 5 behind the mmx.. not a dud at all especially considering they were cheapest chip at the time

Cyrix: https://youtu.be/jx-KE-xlNS0 😁

kennyPENTIUMpowers wrote on 2022-10-29, 15:06:

wow great post .. as for the cyrix mII, this disproves the usual rhetoric that they are useless because of weak FPU.. look at t […]
Show full quote

wow great post ..
as for the cyrix mII, this disproves the usual rhetoric that they are useless because of weak FPU..
look at the 200 tests:
only 1 fps behind the k6's (in 640x400 quake mode) and 5 behind the mmx.. not a dud at all especially considering they were cheapest chip at the time

Thank you for your appreciation.

5 FPS out of 27 is a HUGE margin. On the first board (GA5-AX), all CPUs seems to be slowed down but especially the K6 as someone pointed it in this thread, but I cannot find any reason. If you look at the results of the K6BV3+ or the MSI, the results are more coherent.

MII is not the worst CPU at Quake, just because I included the C6 ! Even the obscures Rise mP6 or Winchip 2 are better. Considering the Winchip is a beefed up 486, it's a pretty weak performance for the Cyrix. The only benchmark where the MII shines is the 3D Bench 1.0c ; this benchmark tends to favorise bigger L1 caches (Intels and mP6 are the worse here).

The MII is the most uneven CPUs of this benchmark, the K6 family is the most consistent in every test, followed by Intel. Bu I have very different results in the 3 mainboard !

I should try on a Intel chipset equipped board (I have an ASUS TXP4 board with a TX chipset).