The non plus CPUs don't support software multiplier manipulation.

As far as I know is the difference between the plus version that it has 3D-Now. And it went from 0.25 µm on 0.18 µm.

K6-2 has 128KB cache and the K6-III has 256KB.

They all have open Multiplier. If it is possible to adjust those via OS is our of my knowledge. I mostly set it via jumpers.

Here, it's easy to inform about all the cpu properties:

K6-II
process: 250 nm
L1-Cache: 32 + 32 KiB (Data + Instructions)
MMX, 3DNow!
https://en.wikipedia.org/wiki/AMD_K6-2

K6-III
process: 250 nm
L1-Cache: 32 + 32 KiB (Data + Instructions), L2-Cache: 256 KiB, fullspeed
MMX, 3DNow!

K6-III+
process: 180 nm
L1-Cache: 32 + 32 KiB (Data + Instructions), L2-Cache: 256 KiB, fullspeed
MMX, Extended 3DNow!, PowerNow!
https://en.wikipedia.org/wiki/AMD_K6-III

Edit: The K6-2 has no 128kB L2 cache. The K6-2+ has it !!!

A very helpful response I received about memory on a + and non + k6, from when I built a k6 super socket 7 system.

Tetrium wrote:

I'm not a technical guy, but I'll try to explain a bit how this matters, what it affects and how you can prevent it causing prob […]
Show full quote

Whiskey wrote:

When the system posts it counts all the RAM but I know that not got anything to do with this cache limit, amirite? I'd love to k […]
Show full quote

When the system posts it counts all the RAM but I know that not got anything to do with this cache limit, amirite?
I'd love to know more about this board cacheable memory limit, I can't really find an explanation as I don't know the proper terms. Is it to do with L caches/External Cache? Or just an inherent limit in the amount of memory the CPU or mainboard can access at once (because of controllers or the northbridge) ?

After a little searching here on vongons I found a thread about the GA-5AA, which said it had a 128mb cache limit but no other info as to why. I have the checked the mobo manual but its rather lacking in technical specs too so any insight would be a huge help.

Cheers, W.

I'm not a technical guy, but I'll try to explain a bit how this matters, what it affects and how you can prevent it causing problems.

The main problem is decreased speed when using more memory than the system will cache. This issue is separate from how much any given motherboard can accept and run with.
Intel 430TX is a good example (this is a common s7 chipset), it can run with I think 256MB, but the chipset will only allow the system to cache 64MB and basically if you put anything more in it, your system will run slower.

How much it can cache, depends on several things, but mostly it's to do with chipset, the motherboard cache and the CPU cache.

What the CPU is concerned, it comes down to this:
Does the CPU have any L2 cache? --> No problem!
Does the CPU not have any L2 cache? --> check chipset/motherboard

A general rule of thumb is this:
Anything Slot 1/Slot A and newer --> forget about cacheable area (except for the very first Celeron, but that CPU is in a class of its own 🤣)
Socket 8 --> forget about cacheable area, you're good to go 😁
Anything Socket 7 which uses a CPU which has L2 cache --> No problem!
Anything Socket 7 and earlier with a CPU without L2 cache --> check motherboard/chipset/cache/etc.

CPUs for s7 which do have L2 cache include: AMD K6-3, AMD K6-2+ and AMD K6-3+ (the AMD K6-2 does not have any L2 cache, it's different from the K6-2+).

The K6-2+ has 128KB L2 cache while both K6-3 variants have 256KB and this is one of the reasons these K6-3's are so popular for (Super) Socket 7.

Install one of these and you're basically good to go.

When not using any of above mentioned CPUs, it's up to the motherboard, chipset and the available L2 cache on the motherboard itself to cache the RAM.
There are many exceptions here though, but generally speaking Socket 7 usually caches 64MB and Super Socket 7 usually caches 128MB (but some can cache more and a few ones perhaps less).
Intel 430HX is a noticeable exception to this rule, but like I said, theres lots of "ifs" and "buts" here 😀

Regarding the difference between the two K6-III's.
The plus version is a laptop CPU, and runs colder and can do software multiplier.
The non plus does not have it and runs more hot.
Running them both on full speed, makes no difference, as they are equally fast.
Running the non-plus just requires a beefier heatsink/cooler, and you have to set
the multiplier manually. Both CPU's do the cache-trick equally well.

melbar wrote:

Edit: The K6-2 has no 128kB L2 cache. The K6-2+ has it !!!

Yes.. I noticed. My fault 😁

And they al have 3D-Now but the + has extended 3D-Now. And the Plus has PowerNow! And that is what is it about: the software scaling... https://en.wikipedia.org/wiki/PowerNow!

needs to be +
for retrogaming by changing multiplier by software

I'll pop in with a related question here.

In his video Phil sets up the CPU as 66MHz x6 rather than 100MHz x4 and I'm not sure I understood why.

What are the benefits / drawbacks of this?

He can clock it slower that way, so he's able to dig a little deeper into 386 territory.

c0keb0ttle wrote:

I'll pop in with a related question here.

In his video Phil sets up the CPU as 66MHz x6 rather than 100MHz x4 and I'm not sure I understood why.

What are the benefits drawbacks of this?

It probably has to do with the fact that it is made to slowdown. With software you can't turn down the frontside bus, only the multiplier. So starting with 66mhz you can go lower and by using the 6x multi you can still achieve 400mhz with that cpu.

It is a little slower compared to 4x100 but in general not a really big deal.

Makes sense. Thanks guys.

meljor wrote:

c0keb0ttle wrote:

I'll pop in with a related question here.

In his video Phil sets up the CPU as 66MHz x6 rather than 100MHz x4 and I'm not sure I understood why.

What are the benefits drawbacks of this?

It probably has to do with the fact that it is made to slowdown. With software you can't turn down the frontside bus, only the multiplier. So starting with 66mhz you can go lower and by using the 6x multi you can still achieve 400mhz with that cpu.

It is a little slower compared to 4x100 but in general not a really big deal.

I know it only counts for windows, but can't setfsb do that ? Or does it only work for slot 1 machines ? I wonder if that software couldn't be ported to DOS

what
the
fu...??

i was going to build a pc like this and got a motherboard gigabyte ga-5smm
Turned it on...and i see a COMPAQ bios boot screen, wtf?

apparently the motherboard was also used by COMPAQ on their machines, with a lame bios that had almost nothing in it.
Searching the web i saw comments from people that flashed the compaq motherboard with original Gigabyte bios, so it will be better when i flash it.

and when i took a closer look at the board....
PCCHIPS syndrome? they took out the SDRAM for the compaq model.

those chips do show on the gigabyte website