Didn't realize I'd start WW3 by asking this question

"Which is better, X or Y?" always starts WW3 😁

candle_86 wrote:

Didn't realize I'd start WW3 by asking this question

we did not make any human sacrifices to the gods of overclock yet

candle_86 wrote:

Didn't realize I'd start WW3 by asking this question

🤣

But it was a very good question. Btw, I never knew FPU performance of K6-3 was actually that good clock for clock compared to P1MMX.

Yeah, I always though FP was the strong suit of the PMMX.

yeah the k6-3 is overall faster, however the pmmx manages to pull ahead in some benchmarks.
the pmmx has stronger pure-fpu performance, while the k6-3 's advantage lies in 3dnow instructions and l2 cache. the l2 cache doesn't give it a large lead over the k6-2 at low clock though.

The testing feipoa did contradicts this, though.

noshutdown wrote:

yeah the k6-3 is overall faster, however the pmmx manages to pull ahead in some benchmarks.
the pmmx has stronger pure-fpu performance, while the k6-3 's advantage lies in 3dnow instructions and l2 cache. the l2 cache doesn't give it a large lead over the k6-2 at low clock though.

The pentium MMX "seems" faster because of P54 optimized code in lots of apps and games from back in the day. Raw performance-wise it's not that impressive.

meljor wrote:

No they do not. In any ''real'' game there is not a single k6 cpu that can even come close to a p2-350mhz, let alone a faster Intel. Only in 3dnow! optimised software they perform amazing but unfortunately there are not many of such progs outside 3dmark and quake.

That's not entirely true, but I can see where you're coming from. I made a few benchmarks and found out that the K6-III+ can be all over the place. You can have a look here:
Pentium II/III VS K6-III+&

Well then K6-3 @ 233 it is then, I'm guessing even though my board doesn't support below 1.8V it will sip power at 233 vs stock 233 and not get all that hot?

Likely not that much of a difference in terms of power load. Switching losses will be less, but intrinsic leakage current will be roughly the same. They weren't a hot running chip anyway. If you want to save power, moving to solid state or a laptop hard drive will probably be a good direction to go.

F2bnp wrote:

meljor wrote:

No they do not. In any ''real'' game there is not a single k6 cpu that can even come close to a p2-350mhz, let alone a faster Intel. Only in 3dnow! optimised software they perform amazing but unfortunately there are not many of such progs outside 3dmark and quake.

That's not entirely true, but I can see where you're coming from. I made a few benchmarks and found out that the K6-III+ can be all over the place. You can have a look here:
Pentium II/III VS K6-III+&

F2bnp wrote:

meljor wrote:

No they do not. In any ''real'' game there is not a single k6 cpu that can even come close to a p2-350mhz, let alone a faster Intel. Only in 3dnow! optimised software they perform amazing but unfortunately there are not many of such progs outside 3dmark and quake.

That's not entirely true, but I can see where you're coming from. I made a few benchmarks and found out that the K6-III+ can be all over the place. You can have a look here:
Pentium II/III VS K6-III+&

Ok, fair enough. From your own benchmarks this conclusion may be a little better? : k6-3 550 gets sometimes beaten by p2-333, gets beaten a lot by p2-350 and gets beaten most of the times by a p2-400.

I do love the platform a lot though, my k6-3+ on a p5a and v2 sli is one of my favorite pc's. I compared it once to my p2b + p2-450 setup and that simply was a LOT faster in every single test i did.
But if you factor in the price of the Intel platform in the 90's, the k6-3 was a very nice alternative!

I remember having to choose between a k6-2 350 setup and a celeron 333 (300a was sold out everywhere as they were great overclockers!). k6-3 was much more expensive here and comparing to the k6-2 the celeron was a better choice, even if it meant that i had to buy a very cheap board. the good slot1 boards were expensive 🤣

meljor wrote:

F2bnp wrote:

meljor wrote:

No they do not. In any ''real'' game there is not a single k6 cpu that can even come close to a p2-350mhz, let alone a faster Intel. Only in 3dnow! optimised software they perform amazing but unfortunately there are not many of such progs outside 3dmark and quake.

That's not entirely true, but I can see where you're coming from. I made a few benchmarks and found out that the K6-III+ can be all over the place. You can have a look here:
Pentium II/III VS K6-III+&

There is significant difference between my Quake 2 results and the Quake 2 results in that thread. In my tests the 550 K6-3 beats the PII at 500MHz w/o using a 3dnow patch. Frankly in most openGL games the K6 has an edge, while in most d3d games the PII has an edge. The performance difference is so small tough it's not worth considering.

PIII vs K6-III is a different story tough. SSE makes a huge difference, and since most apps of the time support it while ignoring 3dnow!, it gives the PIII a practical advantage.

kanecvr wrote:

There is significant difference between my Quake 2 results and the Quake 2 results in that thread. In my tests the 550 K6-3 beats the PII at 500MHz w/o using a 3dnow patch. Frankly in most openGL games the K6 has an edge, while in most d3d games the PII has an edge. The performance difference is so small tough it's not worth considering.

PIII vs K6-III is a different story tough. SSE makes a huge difference, and since most apps of the time support it while ignoring 3dnow!, it gives the PIII a practical advantage.

Your systems were severely bottlenecked by the Voodoo Banshee. I ran quite a few games using a Voodoo5. I had also done a very similar test back in 2013 using a Voodoo 3, but I wasn't very happy with how it turned out, so I wanted to eliminate GPU bottlenecks and also include a few more games for good measure. I wouldn't really say SSE made a huge difference in any actual games. Synthetics and 3DMark is another story of course.

Pentium MMX is competitive with Pentium II as well.

Intel Pentium MMX vs Pentium II

Ignoring 3dnow optimized progs, it is like i said when tested on my own machines with different cpu's , voodoo's, nvidia's etc. etc.

p2 300-333mhz can sometimes be a match, 350mhz is in a lot of cases faster and the p2-400 is simply a better cpu compared to the k6-3+ @ 550mhz.

For me, the super socket 7 is the ''even more retro'' system so i like it better. But speaking about pure performance it doesn't hold a candle compared to the p2, let alone p3.

The biggest problem indeed is it ''beeing all over the place''. When it is fast it can be as fast as a p2-400 or maybe even a 450mhz. But when it is slow it is as slow or slower as a p2-300!

But it is indeed a super socket 7 a.k.a super pentium1, not so much a p2 or p3.

Stretch wrote:

Pentium MMX is competitive with Pentium II as well.

Intel Pentium MMX vs Pentium II

^This.

F2bnp wrote:

Your systems were severely bottlenecked by the Voodoo Banshee. I ran quite a few games using a Voodoo5. I had also done a very similar test back in 2013 using a Voodoo 3, but I wasn't very happy with how it turned out, so I wanted to eliminate GPU bottlenecks and also include a few more games for good measure. I wouldn't really say SSE made a huge difference in any actual games. Synthetics and 3DMark is another story of course.

The banshee is period accurate, the V5 is borderline. Besides, very few people would use a V5 in a K6 or P2 machine - they will bottleneck the card. I find a V5 works best with a fast athlon XP. At one point I benched my P2 overclocked to 500MHz (333Mhz model) and my k6-3 with Geforce 2 MX400 cards. They performed similarly. The P2 had a slight atvantage in HL and a large advantage in Expandable. The K6 does better in HW and Quake 2 as well as Quake 3 (but Q3 runs like crap on both).

kanecvr wrote:

The banshee is period accurate, the V5 is borderline. Besides, very few people would use a V5 in a K6 or P2 machine - they will bottleneck the card. I find a V5 works best with a fast athlon XP. At one point I benched my P2 overclocked to 500MHz (333Mhz model) and my k6-3 with Geforce 2 MX400 cards. They performed similarly. The P2 had a slight atvantage in HL and a large advantage in Expandable. The K6 does better in HW and Quake 2 as well as Quake 3 (but Q3 runs like crap on both).

I don't agree with you. We are comparing different CPUs and we want to see how fast they can go. There's no reason to introduce GPU bottlenecks.

I think most people interested in a K6-III system or Pentium II today will probably want to pair it with a Voodoo 3 or a Voodoo 5 (for the best performance on SS7) for an enjoyable experience, unless they are aiming for very specific builds that they had once and want to replicate.

F2bnp wrote:

kanecvr wrote:

The banshee is period accurate, the V5 is borderline. Besides, very few people would use a V5 in a K6 or P2 machine - they will bottleneck the card. I find a V5 works best with a fast athlon XP. At one point I benched my P2 overclocked to 500MHz (333Mhz model) and my k6-3 with Geforce 2 MX400 cards. They performed similarly. The P2 had a slight atvantage in HL and a large advantage in Expandable. The K6 does better in HW and Quake 2 as well as Quake 3 (but Q3 runs like crap on both).

I don't agree with you. We are comparing different CPUs and we want to see how fast they can go. There's no reason to introduce GPU bottlenecks.

I think most people interested in a K6-III system or Pentium II today will probably want to pair it with a Voodoo 3 or a Voodoo 5 (for the best performance on SS7) for an enjoyable experience, unless they are aiming for very specific builds that they had once and want to replicate.

Yeah you kind of have a point. I use a GF2 GTS in my K6-3 rig so I can play at 1600x1200. The machine can't do that with a slower card - but I'd NEVER stick a V5 in a K6 or a P2 - that's kind of silly. I'd rather put it in a socket A rig to take full advantage of the card's speed and avoid CPU bottlenecks with the voodoo 5.

Other cards also get bottlenecked by period correct setups. For example a GF4 Ti4600 will score a measly 8000-9000 pts in 3dmark01 on a fast P3, but it will score over 14k points on a 939 Athlon 64. Same story with the 9800XT and X800XT.

I guess it depends on if you want to get max performance out of the system (and use a way faster video card for silly resolutions + AA) or max performance out of the video card (and use a way newer CPU). Depending on the system, I'm guilty of both.