The Celeron 466 would definitely be more powerful, but I've heard K6 systems tend to be more flexible in terms of underclocking, which may be useful for some games.

I have built a K6-2 500 super-socket platform at beginning of this year. Looking back this was the best option for me, cause i use it as a "time-maschine" with 386, 486 and Pentium speed.

(End of 90's, i had not one Celeron 466's, but two with the famous Abit BP6 board. I've used W98 and W2000, but more games were running on W98 platform. DOS was not a topic in the past...)

With a Celeron Mendocino, the core is PII based, this is not possible, due to disabling the caches. The speed is not scaling as good as a AMD K6- series CPU, or a Pentium, or Pentium MMX.
See also this thread: Let's benchmark our systems with caches disabled

When you want to play also DOS games, which are speed relevant
http://www.vogonswiki.com/index.php/List_of_C … sensitive_games
, then you should definitely go with AMD K6... or an old Pentium (MMX).

With the + versions of K6-2/-3 , your're right. Nowadays these CPU's are more popular than the others, due to the little speed+ of L2 cache in case of the K6-2,
and also for the "comfortable" switching multipliers and caches.
For me, the K6-2 is enough, cause i run it mainly at lowest 2.5 multiplier setting, and have a simple control via BIOS of 386SX-40 / 486SX-25 / Pentium 166 range.

For little more speed i would better go with PIII or Athlon due to late DOS/Win98 games...

So, honestly I think the answer to this really completely depends on you. The K6-2 system will ultimately allow you to play a wider range of games from diffrent eras. I think Phils video on this has caused a lot of people to develop a similar build. If you do a K6 based build I would do an ISA sound card like AWE64 or SB16 and you won't have to worry about the SB-Link. Also, the Voodoo3 is complemented by the 3D=Now in the K6-2 quite nicely so if you going for Glide games this might actually be better.

The Celeron build will however perform better at the top end, and the 440ZX(which is kinda a strange stripped down version of the 440BX) chipset will allow for more upwards mobility in terms of processors. Looks like it can go up to a 100Mhz Bus Pentium 3 but RAM is limited to 512MB. But once again going to to that hardware level isn't really nessecary for dos gaming and in the process of doing that you are cutting off the ability to play some of the older dos games.

Ultimately both systems would serve you well you just have to decide what you want.

From my point of view, the 440ZX is as good as the 440BX. There are only these main differences:

SMP: 440BX- Yes / 440ZX - No
Max.Mem: 440BX- 1024MB / 440ZX - 512MB
RAM Parity/ECC: 440BX- Yes / 440ZX - No
PCI support: 440BX- 2.1 (64-bit optional) / 440ZX - 2.1

So, at least 512MB are really enough, also when you see Solano (i815) platform has also max. 512MB.

Bancho wrote:

I want to use this mainly for DOS games and earlier Windows games etc.

I see no reason to consider the Celeron, unless your games will exclusively be high end and demanding. For DOS, the K6 will be fast enough for even the most demanding games and will be much more flexible (see link melbar provided) if you do have games that need slowdown.

Like the others said, K6 CPUs allow you to pull of some great tricks as regards to Time Machine (tm 🤣 ) systems, however you really want K6-2+/III+ to make the most out of them which is a no go in your case, so that kinda sucks. On top of that, K6-2 500 is much slower than the Celeron 466. The K6-2 at this speed is more aking to a Celeron 300A for 3D games and even that is debatable.

The Celeron can be useful for some SVGA DOS games too, though in cases where more oomph is required, you're much better off with something faster overall, like a high clocked Pentium III.

I have not tried any + versions of eighter K6-II or K6-III.
My Asus P5A is running an K6-II-500 right now, and my Gigabyte GA-5AX are running an K6-III-400.
Both are stable, and are running great. Personally, I prefer Ali Aladdin-V over MVP-3.
For a time machine, the AMD platform are cool.
Though I prefer running MVP3 with P-133/166 instead, when dealing with the Time Machine concept.

I never build a time machine. If Bancho intends to use any games which will require slowdown, I'd guess he would be better off with K6-2, if only for the purpose of slowing the system down. s7 is also one of the most versatile sockets with chips from many different manufacturers and makes available, so if Bancho ever wants to try out some different CPUs and see how these run, K6-2 will be muhc more fun.

The ZX will be almost the opposite with very little ability to change, but it will probably be more "plug & play", easier to build.
I really liked my s370 Mendocino when I used it. Was also an effective way to dump excess PC-100 cl3 memory, as the FSB runs at only 66MHz anyway, so I could keep the lower latency modules for other builds.

I think the ZX chipset will be fine and as someone else pointed out, its limits compared to BX aren't of any significant concern in this case.

I'd keep both systems.

Regarding slow-down: If your S370 board can run VIA C3 CPUs, especially an EZRA-T, then that CPU should scale more smoothly than the K6-2, something like a 286 to a P2. Also, a game like Ultima 7 renables L1 cache to run too fast on K6s, but not VIA C3.

Use Setmul to change multi and ICD/BPD to get this flexibility on the VIA C3 (avoid Nehemiahs for slow-down flexibility).

j^aws, if you have a VIA C3 that you can benchmark, it would be great if you could contribute results. I maintain a database of cpu benchmarks with cache enabled and disabled with setmul. The link is in my signature. 😀

j^aws wrote:

I'd keep both systems.

Regarding slow-down: If your S370 board can run VIA C3 CPUs, especially an EZRA-T, then that CPU should scale more smoothly than the K6-2, something like a 286 to a P2. Also, a game like Ultima 7 renables L1 cache to run too fast on K6s, but not VIA C3.

Use Setmul to change multi and ICD/BPD to get this flexibility on the VIA C3 (avoid Nehemiahs for slow-down flexibility).

Personally I'd do the same, just don't choose, build all of them! Try all of them and see which ones I like! And at some point I had about 20 working rigs (some only had Windows installed and tweaked and only used for running superpi) along with a few more systems which were in various stages of cannibalization 🤣!

I have systems based on Celeron "mendocino" 466MHz & also one based on a AMD K6-2 450MHz. I have to say I love them both, they are very individual machines.
Build both! 😉

Best,
Chris

@clueless1: I'll probably start a thread or something soon. Regarding cache enabling and disabling, the speed flexibility comes from various parameters - especially significant is ICD for Ezra/ Ezra-T.

K6 is better for DOS with it's slowdown possibilities, if it will not ruin compatibility with the games you need. As Intel CPUs are the only about which game developpers cared and Intel had more resources to do compatibility testing/designing. For most Win9x games up to 1997-98 that K6 should be enough.
Don't forget about DOSBox option for DOS games, as it may fit your needs for most DOS games in not worse way.

j^aws wrote:

@clueless1: I'll probably start a thread or something soon. Regarding cache enabling and disabling, the speed flexibility comes from various parameters - especially significant is ICD for Ezra/ Ezra-T.

As there is no uniformity in CPU benchmarking with slowdown technics (between 2 CPUs after slowing, 1st may to be faster in one test and to be slower in other), it's better to do as many tests as you find acceptable.

K6-2 but id run a K6-3+ and just disable cache as needed

j^aws wrote:

@clueless1: I'll probably start a thread or something soon. Regarding cache enabling and disabling, the speed flexibility comes from various parameters - especially significant is ICD for Ezra/ Ezra-T.

Looking forward to seeing your thread! Would still appreciate if you could document in my database how the cpu does with various caches disabled, that way others who are viewing the database can compare to other cpus in the database. It would be a nice gift to our retro community. 😀 It's very simple--just 3dbench, pcpbench, speedsys and doom, all run from Phil's prepackaged benchmark kit. Do runs with all combinations of caches (L1, L2, L1+L2) disabled plus one run with all caches enabled as reference.

Cheers!

I can do some VIA CPU tests for the cacheless database chap.
866, 1.0 & 1.2 MHz C3's.
Will get onto that later on.

@ BSA Starfire 😎

clueless1 wrote:

j^aws wrote:

@clueless1: I'll probably start a thread or something soon. Regarding cache enabling and disabling, the speed flexibility comes from various parameters - especially significant is ICD for Ezra/ Ezra-T.

Looking forward to seeing your thread! Would still appreciate if you could document in my database how the cpu does with various caches disabled, that way others who are viewing the database can compare to other cpus in the database. It would be a nice gift to our retro community. 😀 It's very simple--just 3dbench, pcpbench, speedsys and doom, all run from Phil's prepackaged benchmark kit. Do runs with all combinations of caches (L1, L2, L1+L2) disabled plus one run with all caches enabled as reference.

Cheers!

Alright, I will add them at somepoint. And BTW, started the thread here:
WIP 1: Ezra-T slow-down machine, S370 industrial build...