Past that they perform basically the same at equal clocks. But the + can overclock higher.

11,1% slower clock, but up to 18% more performance ... (Dosbench test a )
Was tested, not by me, but no multi 2x/6X was possible.

The 2+ can sometimes be modded to a 3+ by the way.

Way out of my league, sorry.

The CPU only cost me 20€, so everything is fine.
I think, i`ll sell it. Or keep it, spare parts or something. 😉

Demolition-Man wrote on 2023-09-30, 17:53:

11,1% slower clock, but up to 18% more performance ... (Dosbench test a ) Was tested, not by me, but no multi 2x/6X was possibl […]
Show full quote

Past that they perform basically the same at equal clocks. But the + can overclock higher.

11,1% slower clock, but up to 18% more performance ... (Dosbench test a )
Was tested, not by me, but no multi 2x/6X was possible.

The 2+ can sometimes be modded to a 3+ by the way.

Way out of my league, sorry.

The CPU only cost me 20€, so everything is fine.
I think, i`ll sell it. Or keep it, spare parts or something. 😉

You are saying that you found a performance difference between III and 3+? All else equal.

Comparing 2+ to a III isn’t a fair test. That will be different. The 2+ has 128k of cache disabled.

I recall being surprised at how a K6-3+ at the same clock speed as a K6-3 scored noticeably higher on 3D Mark. I didn't document it so I can't say more, but I do recall there being a difference.

Repo Man11 wrote on 2023-09-30, 22:04:

I recall being surprised at how a K6-3+ at the same clock speed as a K6-3 scored noticeably higher on 3D Mark. I didn't document it so I can't say more, but I do recall there being a difference.

Do you have an explanation?

Sphere478 wrote on 2023-10-01, 00:03:

Repo Man11 wrote on 2023-09-30, 22:04:

I recall being surprised at how a K6-3+ at the same clock speed as a K6-3 scored noticeably higher on 3D Mark. I didn't document it so I can't say more, but I do recall there being a difference.

Do you have an explanation?

Same motherboard, same MHz, same cache - the only thing that's left is the additional instructions that the plus CPUs have. I don't recall it being a huge difference, but I remember being surprised that there was any difference.

Any sources out there that make note of the + having a change like that that you are aware of? Like reviews circa the time it was released?

I'm working here with a defective onboard L2 cache and with a single 256 MB RAM module. The only explanation I have, is the difference in cache on the CPU.
(Tyan S1590S early version, 256 MB PC100 RAM, 20 GB HDD, Creative S3 Savage 4 PCI, Voodoo 2 8MB, and a CT2230)

DOSBENCH
K6-III 450
1. 33,3
2. 326,3
3. 448,3 / 269
4. 99,6 / 59,7
5. 132
6. 33,2
a) 2134 /182
b) 2134 / 762
c) 969 / 12,4 / 78,1
d) 969 / 31,9 / 30,4
e) 969 / 17,1 / 56,8

K6-2+ 500
1. 00,0
2. 337,7
3. 459,5 / 275,7
4. 98,6 / 59,2
5. 128,5
6. 32,8
a) 2134 / 215
b) 2134 / 798
c) 969 / 12,5 / 77,2
d) 969 / 32,4 / 29,9
e) 969 / 17,1 / 56,5

Win98
SiSoft Sandra is not practical, so only the results of the 3D Mark.
K6-III 450
3DMark 99 (Voodoo 2)
Score: 1930
CPU 6612

Savage 4
Score: 2730
CPU: 6572

K6-2+ 500
Voodoo 2
Score: 1851
CPU: 6328

Savage 4
Score: 2631
CPU 6319

What do you think? With a few exceptions such as the theoretical CPU benchmarks in SiSoft Sandra and a few DOS tests, the K6-III 450 is better despite being clocked 50 MHz less.

2+ vs 3 is no fair comparison for the question at hand unless you mod the. 2+ to 3+

But aside from that, looks like you are ready to start playing some games! :p

Demolition-Man wrote on 2023-10-01, 06:31:

What do you think? With a few exceptions such as the theoretical CPU benchmarks in SiSoft Sandra and a few DOS tests, the K6-III 450 is better despite being clocked 50 MHz less.

I haven't tested a K6-2+ myself, I just remember some other people's benchmarks (maybe philscomputerlab) that seemed to show them on par with a K6-3 if clocked 50MHz higher.

In your tests, 3DMark 99 does favor the K6-3 450MHz - but they're so close it would take a lot more tests in real games to make a strong conclusion. Based on what you've posted I would just say they're close.
I think most K6-2+ chips can run 550MHz without difficulty, certainly no more difficult than running an original K6-III at 450MHz. So at the same level of ease/stability the "+" chips are good for +100MHz IMO, which I think would put them more consistently in the lead. There might be some games that want the bigger cache though.

I have an S1590 that I tried some benchmarks with a few years ago. It has a 3+ in it and to be honest, I try to avoid swapping CPUs on that board anymore because I'm always afraid I'll break the socket.
When I upgraded from the K6-III 450MHz to the K6-3+, I tried clocking it the same at 450MHz. I did notice the 3+ seemed to score slightly slower in L2 cache speed in SpeedSys. The difference was very small, but repeatable. It made me wonder if they could have tweaked something with cache latency in the "+" version to allow higher clocks. The difference was so tiny it's in the realm of a minor curiosity though and doesn't really matter.
I looked through my old spreadsheet trying to find actual game tests that could compare the K6-III and K6-3+, but it's hard to tell if any of them are exactly comparable. I didn't test much with the K6-III before I replaced it with the 3+.

But aside from that, looks like you are ready to start playing some games! :p

I play with one of my retro PCs almost every day. 😉
Gives me more than playing modern games on today's overpriced high-end PCs with their incredible power consumption.

I'm always afraid I'll break the socket.

I did. I need to use a special bracket to hold my cooler.

That's right, enough testing and modifications for now. The CPU works fine as it is. I have a large retro PC collection and can upgrade to a more powerful PC at any time. I'm still looking for alternative PCI graphics cards and maybe another sound card. We have clarified the most important thing: the K6-2+ CPU works. It was my first purchase of hardware from China via eBay USA in US dollars delivered to Germany. And it all worked so well that I'm ordering again. They still have nice Cyrix CPUs for my older machines.

And Thank You!

I was correct that there was a slight difference, but I was incorrect about which CPU had the advantage! The K6-3+ used here is a K6-2+ 570 modified to enable the full L2 cache.

The k6-3 had a slight advantage across the board!

How odd.. 🤔

Try again with l3 disabled? Curious

Would also be curious which was better with central tweaking unit maxxed out

The K6-2+ 570 used in the above test might be the one where a tiny chunk of the core stuck to the heat shield when I did the mod. With that in mind, I decided to use a different test system and a different K6-3+ (but the same K6-3 as it is the only one I have). This time it's a low voltage K6-3+ 400. The motherboard is a PCChips M520 with an Evergreen Spectra upgrade adapter. The results were similar, and again the K6-3 came out on top.

And here's the K6-3:

Thank you for the benchmark screenshots. The comparisons are very interesting. Unfortunately, I'm no expert, but what's the deal with the L3 cache? Is the onboard cache still used, or does it perhaps depend on the respective board or the software? That's how I came up with the K6-III processor in the first place. This made it possible to replace the defective onboard cache and the PC was fast enough again for Windows 98 games, for example. The M5ALA board still has its cache but unfortunately I don't have anything to compare it to.

Demolition-Man wrote on 2023-10-02, 17:18:

Thank you for the benchmark screenshots. The comparisons are very interesting. Unfortunately, I'm no expert, but what's the deal with the L3 cache? Is the onboard cache still used, or does it perhaps depend on the respective board or the software? That's how I came up with the K6-III processor in the first place. This made it possible to replace the defective onboard cache and the PC was fast enough again for Windows 98 games, for example. The M5ALA board still has its cache but unfortunately I don't have anything to compare it to.

The 1024k cache on the MVP3 motherboard, when combined with a CPU that has on die L2, has a beneficial effect on overall performance. On the VX chipset motherboard, I've run benchmarks with it enabled and without, and enabling the motherboard's cache has a slightly detrimental effect on performance. I think the only way to be sure if it is helping or hurting is to run benchmarks both ways.

...enabling the motherboard's cache has a slightly detrimental effect on performance.

Measurable or noticeable?
SS7 boards are rare and expensive.
It is also one of my few original pieces from the period. 😉

Measurable

It seems that most ss7 cache enabled is better.

While about half of s5/7 goes one way or the other

It's not easy with a DOS/Win 98SE PC. With every MHz more, or every little bit more speed (cache), the chance that some DOS game will no longer work properly increases.