Probably not, your RAM and cache will continue to run at 100mhz.

The processing speed of the 550 will obviously be faster. There were some core revisions along the way but I can’t recall if anything significant happened between 400 and 550. Obviously, k6 -3 and the plus models brought more speed to the table.

Games that are unplayable on 400 will most likely still be unplayable on 550. Its not the same as overclocking Celeron 300A to 450MHz.

From what I've read, 400-500MHz K6-2's should be based on the same Chomper Extended (CXT) core.
If K6-III(+)'s/K6-2+'s were available here at a reasonable price, this of course would be a no brainer.

It seems most people only think of the L2 Cache equipped Super Socket 7 CPU's, completly disregarding everything else for this platform, which makes it hard to look for advice when you are given such a motherboard for free (S7AX), without the CPU, and don't want to spend a fortune.

Seems to be the second time when I receive a free AT motherboard (EDIT: first is PCChips M747) and by looking at Vogons posts coming to a conclusion that it's junk 😒

EDIT: Well, maybe not junk, but nothing really exciting.

Im sure it will be exciting to someone who grew up on Socket7 (aka with Stockholm syndrome 😜)

After about 400MHz the scaling on the K6-2 really drops off. Although I have noticed in certain tasks, e.g LAME wav to mp3 encoding you do still see a big jump in performance.

More results:
https://www.vogonswiki.com/index.php/Super_So … et_7_Benchmarks

The best cheap K6-2 I found and bought is a 450MHz one. Will be interesting to see how it compares to the Celeron 333 Slot 1 that I also have on that PCChips motherboard, the Celeron should be better, right?

EDIT: How about overclocking on both of those?

"bxpro" scam pcchips, you will be lucky to get it not crashing

Eh, it's not that bad, it works fine. Still haven't figured out why the CMOS settings on it are not saved which I've described in another topic of mine, can't say for certain if it's the diode, capacitors or something else.

Seems I am like 15 years too late for being excited about more period-correct Win98SE components, been mostly spending these past years running Win98SE on overkill and latest compatible components, mostly in unofficial ways. It did get somewhat boring once the current limitations of Win98SE and KernelEx (and its updates) have been reached. There is still some untapped potential however. 😀

Maybe I am doing myself a dissservice by reading about stuff I got on VOGONS first, rather than actually testing it. I guess the times of people not being informed/educated about retro stuff is truely over and there's barely anything that excites people anymore.

Back on topic, is K6-2 really that bad?

MrMateczko wrote on 2022-11-13, 16:44:

Back on topic, is K6-2 really that bad?

Well, kind of. It's not as bad as some people make it, but it's also not as good as AMD advertised it. The issue with the K6-2 is the limitations of the Super Socket 7 platform. Intel's fastest Socket 7 Desktop processor was the Pentium MMX 233. That processor ran at 66MHz FSB. The socket 7 processor interface scales with FSB, so increasing the FSB clock to 100MHz gives a 50% boost to the platform. So the Super Socket 7 platform at FSB100 is equally able to power a hypothetical FSB-100 Pentium MMX 350 as it is able to run the real FSB-66 Pentium MMX 233. Intel's Socket 8 / Slot 1 / Socket 370 platform, the successor to the Socket 7 platform, moved the L2 cache from the mainboard to the processor (except for the low-end "Covington" celeron, that didn't have any L2 cache).

At higher processing power, the performance of the socket becomes a limiting factor in many use cases. So going to higher processor clock speeds results in diminishing returns once the performance is bottlenecked by the processor waiting for the front-side bus and not the executing engine's performance. At "low" clock frequencies, like 266 to 350 MHz, the K6-2 is a real good performer on the Super Socket 7 platform. A FSB100 K6-2 at 300MHz provides a significant performance boost compared to the fastest Intel processor at the same FSB, the Pentium MMX 233 (unless we have code specifically optimized for the Pentium Processors. So no Quake 1 software rendering benchmarks as example, please...).

So a K6-2 at 550MHz will definitely outperform a K6-2 at 400MHz if the L1 cache hit rate is high enough to not make the frontside bus an issue. The reason K6-2 are touted to be "that bad" is that the K6-2 is limited to the "L2 at 100MHz" concept, whereas the direct competitor wasn't the Pentium MMX anymore, but the Pentium II. At that time, the Pentium II ran the L2 cache at half the core clock. So we are comparing an K6-2 at 400MHz with L2 at 100MHz against a Pentium II at 400MHz with L2 at 200MHz. Guess which processor has the higher L2 cache bandwidth... In use cases where the working set significantly exceeds the L2 cache size of the Pentium II (which is 256KB), and in use cases that run mostly within the L1 cache, the L2 performance doesn't matter that much. Both Super Socket 7 boards and Pentium II boards usually run SDRAM at 100MHz and make good use of CL2 modules. Both the Super Socket 7 frontside bus and the Slot 1 frontside bus are fast enough to provide the full memory performance to the processor, so in situations where L2 performance doesn't matter, the K6-2 and the Pentium II had comparable performance. The "issue" is that the L2 size of the Pentium II has been chosen in a way that it does matter, i.e. it has been chosen so big that the presence of that cache provides a significant win to many relevant use-cases at that day.

That should explain why K6-2+, K6-3+ and K6-III processors are so hyped in the retro community: They bring Pentium II capabilities (core-coupled L2 cache) to the Super Socket 7 platform, which is a unique selling point of these processors. As the L2 cache size of the K6-III and K6-2+ processors equals the cache size of the Pentium II, these processors eliminated the FSB bottleneck just as the Pentium II did. The higher the clock, obviously, the higher the advantage of on-die L2 cache compared to mainboard L2 cache. That's why threads like this one exist, asking about the clock frequency where increasing core clock stops making sense unless you have on-die L2 cache.

Great summary mkarcher, thanks.
So it is a good thing I haven't paid more for a 500MHz one, 450MHz should be enough.

It is a learning experience for me to actually experience some proper retro hardware, rather than just relying on emulation or using more modern, non-retro stuff (USB booting, SATA, and PS/2 BIOS emulation in DOS for USB HID devices? How boring!)

I guess the hardware I got now could be a nice retro stopgap to have some fun with, before I find some good prices/save up money for some proper good retro hardware, it might take a while but I'm not in a hurry. My overkill 98SE build also took some time, have some plans regarding it. Looking for retro hardware might be also a good excuse to visit some other countries. 😁

100mhz fsb pentiums are fun. I have a correction, technically the fastest socket 7 pentium was the 300mhz tillamook.

But it gets even crazier from there. I actually own (soon) a chinese golden fake 366

(Which I expect to be a 266 or 300 with a lying label.)

Anyway, you can run many ppga tillamooks at 100x4 pretty easily. Some have even gotten them into the 500s

Little off topic, but interesting.

K6-2 wasn’t that bad. Just like cyrix wasn’t either. Winchip though. That’s the celeron of socket 7 🤣

CPU Galaxy made some nice videos about those fake CPUs, you probably watched them already 😁
https://www.youtube.com/watch?v=wsKjX6UGYUQ
https://www.youtube.com/watch?v=l-GbiiuiCBY
https://www.youtube.com/watch?v=736iXRdXuTE

MrMateczko wrote on 2022-11-13, 18:10:

I guess the hardware I got now could be a nice retro stopgap to have some fun with, before I find some good prices/save up money for some proper good retro hardware,

Be careful to not escalate on what "proper retro hardware" is. Back in the end of the 90s, K6-2 computers were widespread. K6-III computers were rare, as the price advantage of the Super Socket 7 platfrom compared to the Slot 1 platform diminished when buying a K6-III processor. K6-2+ and K6-3+ were late to the K6 game and mostly used for laptops. They are basically an optimized and improved K6-III core that was shipped when the Super Socket 7 platform was no longer relevant in the Desktop PC market.

Some retro enthusiasts today really like to pump up their systems to the maximum that was technically possible, even (or especially) if that maximum wasn't affordable at the time when that hardware was common, just to show "what could have been". I don't think there is anything wrong with it (heck, I have one of those K6-2+ 570 that swamped ebay around two years ago myself), but don't let owners of these hardware tell you what "proper good retro hardware" is. As the name "retro" implies, it is about the experience we had back in the day, when the "retro games" we remember and want to re-experience, were current games. A 450 MHz K6-2 was a common entry-level PC in 1999. No one used a K6-2+ 570 in a desktop computer in those days. So maybe the 450MHz processor is even "more proper retro" than the K6-2+...

There is one valuable point about the K6-2+ processors, though: As they were intended for laptops, power saving by reducing the clock (and the voltage, if the laptop mainboard supports it) was a design goal. They have their multiplier switchable at real time. The consequence is that a wide range of performance dropping methods is available on a K6-2+: You can lower the multiplier from x6 to x3, you can disable L2 cache on the chipset, you can disable L1 cache on the processor, so you can tune the performance of your one Retro PC in a wider way than you can with an everyday K6-2 processor. This is helpful for games that would run too fast (becasue they were written when 286 computers were common, and fail to work properly on everything faster than a 386). This makes your "retro rig" in fact more versatile. We didn't have this back in the day, though, although we might have fetched the old 286 computer from the attic to re-live 1980s games. It should be your choice whether you want to spend a lot of money on hype hardware that was rare even back in the day (Gravis Ultrasound, K6-3+, Adaptec 1542 SCSI controllers for ISA systems, the one perfect SoundBlaster card (which doesn't even exist), a Voodoo 5 5000 graphics card), or you just go with what you have yet and enjoy the experience you had back then without the bells and whistles. Enjoy the retro computers! Enjoy the retro software! Enjoy talking about retro stuff! Enjoy tinkering with retro stuff! Enjoy having hype hardware if you get it a price that you think it is worth! But don't let the retro community members give you the impression that not having the hype hardware is "improper" or "bad" retro hardware.

I believe through central tweaking unit that 2x and 2.5x are supported by 2+\3+

For me it’s about pushing the limits and making cool combinations and frankensteins. Combos that weren’t possible like irams and ssds and sata are totally on the table for me. Make something that would make people of 1996 go waaaaaa?!?

MrMateczko wrote on 2022-11-13, 18:23:

CPU Galaxy made some nice videos about those fake CPUs, you probably watched them already :D https://www.youtube.com/watch?v=wsK […]
Show full quote

CPU Galaxy made some nice videos about those fake CPUs, you probably watched them already 😁
https://www.youtube.com/watch?v=wsKjX6UGYUQ
https://www.youtube.com/watch?v=l-GbiiuiCBY
https://www.youtube.com/watch?v=736iXRdXuTE

Indeed, someone send him a note that there exists a 366 model

Later run 400 come off the exact same wafers as 550s, which were cherrypicked out. Earlier, they were same design but maybe process wasn't quite tuned enough for them to go faster than 450... so you can get some cheapy 400 and 450s that will happily do 500+, they may have been downbinned due to demand fluctuations or because thermals or power consumption didn't quite meet spec at higher speed.

Now apples to apples, if you have a 400 running at 6x66, and stick a 550 on a very fast super 7 that can do 124Mhz FSB and not flake out it's cache AGP or PCI, then at 4.5 x 124 that would be as quite an exciting jump as the celeron from 366 to 550. However, since you could also run a 400 on the same bus speed, if not quite as high on the multi, the difference then would be less noticable.

You'll often hear something like "K6-2 suuuucks becuz muh Quake slow" whereas in 10,000 games that aren't quake related, ID's nifty pentium specific hack of interleaving the FPU and Integer units, doing like sneaky hyperthreading in a very application specific way, doesn't apply. Then it's down to level playing field performance, and K6-2 can often win. Also there's more games playable in the 300-600Mhz range where 3DNow helps, than there are SSE supporting ones... i.e. if a game supports SSE you better have PIII-700 up or it will get nasty. (Or a Duron 600 😜 )

BitWrangler wrote on 2022-11-13, 18:43:

i.e. if a game supports SSE you better have PIII-700 up or it will get nasty. (Or a Duron 600 😜 )

Just to make sure this is not misunderstood: The Duron 600 might be fast enough to run a game with SSE support at a reasonable speed, but it does not support SSE. If the game only runs at acceptable speed on a PIII-700 because it makes heavy use of SSE, then it will not be playable on a Duron 600. On the other hand, if "supports SSE" just references an era in which SSE support was common, but games were still designed to work quite well even without SSE (possibly at lower graphics detail level), then a Duron 600 could be a adequate non-SSE platform to run that game.

Thank you all for your feedback, I'll have fun with whatever hardware I receive. I wasn't even thinking about a possibility of receiving two Baby AT PCs this year for free so I am somewhat happy. 😀
I actually met a person locally who is into retro stuff too, and is happy to grab some of my stuff that I won't be using, I really do not like to have extra hardware laying around 😀
He was living in UK some time ago and it makes me somewhat jealous about hearing his retro hardware adventures he had there, including retro consoles and alike, good stuff.

I already took many years of my life doing overkill Win98SE stuff, and I am ready to take it to the next level, I have all the hardware already, now to actually make something unique with it, I have plans 😁

As for the Celeron 333MHz/AMD K6-2 450MHz - how would they compare with same RAM and S3 Virge GX (or similar/better GPU)? I'll be doing tests myself as well.

Also bought at a reasonable price an ESS1688 ISA sound card, I think it's a good choice from what I've read. Would be interesting to test the OPL emulation between it and the CMI8330 that it integrated in the PCChips motherboard.

My family's first PC was in circa 1999-2000 with a Celeron 533MHz Socket 370 with 192MB RAM that was upgraded from unknown stock amount, and a S3 Savage4 GPU (upgraded later to I think GeForce 2 MX or something alike). It had a yellow motherboard with a VIA chipset...and that's all the information I have on it, I still do not know what was the model of the motherboard and what was the sound card. I remember DOS games running well so maybe it was ISA, dunno. Of course Win98SE, only a few childhood pics remain 😀

mkarcher wrote on 2022-11-13, 19:13:

BitWrangler wrote on 2022-11-13, 18:43:

i.e. if a game supports SSE you better have PIII-700 up or it will get nasty. (Or a Duron 600 😜 )

Just to make sure this is not misunderstood: The Duron 600 might be fast enough to run a game with SSE support at a reasonable speed, but it does not support SSE. If the game only runs at acceptable speed on a PIII-700 because it makes heavy use of SSE, then it will not be playable on a Duron 600. On the other hand, if "supports SSE" just references an era in which SSE support was common, but games were still designed to work quite well even without SSE (possibly at lower graphics detail level), then a Duron 600 could be a adequate non-SSE platform to run that game.

Derp yes, Duron didn't get SSE until morgan cores. Doesn't help my memory circuits that journos used to refer to 3DNow and 3DNow+ as "an SSE implementation" since they are streaming extensions just not the "SSE" set Intel uses.