Yeah, I think it is an unfair comparison, the whole chipset is from a different generation.
Thank you for posting your results! I like benchmarks! 😁

What I would try to do is to get a Pentium Pro system with the i440FX chipset.
Then you have 2 Pentiums with 512kb L2 cache. I would really like to see the difference 😀

But then on the other hand - no MMX or AGP, 32Bit... not really a "gamer CPU".

Yes, the Pentium II was that much faster. You can have a bit more fun by running some games, you could maybe try stuff like Quake II and even some early Lithtech games such as Shogo. The difference in average framerates and especially minimum framerates (which will lead to much much less stuttering on the PII) should be quite significant.

Pentium II 233 was a cool overclocker too. Unlocked multiplier meant it would almost always be able to run at 266. Not bad at all.

TandySensation wrote:

Couldn't find any old reviews comparing both chips at the same speed(except for the 686 comparison on Vogons normalized to 133)

Did you find the data contained within the Ultimate 686 Benchmark Comparison insufficient to fulfill your curiosity? I have included the raw data as well, not just the P233 normalised data. The PII-233, for example, scored 2394 3DMark99Max CPUMarks, while the P55c-233 scored 1619. I do not understand how you got 8718 for your PII-233 3DMark99Max CPUMark score. I do not imagine the 440BX chipset is 4x faster than the Apollo Pro 133A. I also do not envision the Deschutes being 4x faster than the Klamath per clock.

I think all those tests are probably using 32-bit code, which definitely runs better on 686. That was when they began optimizing the CPU for 32-bit and compromising 16-bit performance.
Testing some 16-bit apps/games could be interesting, but I don't know anything specific to suggest. Anything that uses 16-bit code is probably already so fast on either of these CPUs that the difference wouldn't matter.

Whether the 440BX is fair depends on the goal of the test. The 440BX is definitely a newer chipset than that CPU, but it does represent what you'd likely utilize if building such a system today. However, at the time of introduction, those early P2s were matched up with a 440FX. 440FX was typical for Pentium Pros but on slot-1 boards it's pretty uncommon, I think there was very little time before the 440LX was released. Anybody who bought an early P2 + 440FX board was spending real money compared to a P55C.

I'm not surprised the PII runs that much faster.

-Much faster FPU in the PII
-Out of order execution on the PII.
-L2 cache runs faster and is on its own backside bus, greatly reducing latency.
-Better memory controller in the 440BX

Sticking that PII on a 440FX board will definitely slow it down. I've benched my own Klamath-300 on 440FX as well as 440BX. On the FX board, it's not much faster than a K6-2 300. On the BX board, it's more like a K6-III 450.

I upgraded from a not very old Pentium MMX @ 292 MHz to a P2 233 @ 292 MHz back when the Pentium 2 was the new hot thing. I was impressed but the big improvement wasnt seen until I replaced the motherboard with a BX board and pushed the PII 233 to 350 MHz. It ran hot as hell even with improved cooling but I diddnt care much about my PCs health back then as long as I had the fastest system at the LAN-party 😁

Keep in mind that PII has P6 architecture, in other words, it's predecessor was Pentium Pro and not Pentium. P6 is not a newer/better version of P55c, but rather a very different CPU.

I wonder if my 333MHz Deschutes has a locked multiplier. I have a 440FX board to compare it to a 233MHz Pentium, that would be fun.

Most 333s are locked.

From my point of view and from what I can remember if you're using a 440LX or a VIA133 chipset it doesn't even matter if your using PII 233, PII 333 up to PIII 500
paired with a voodoo 3 or a G400 (I don't even remember if I was able to boot with a TNT2 Pro on the LX) it will not be enough to smoothly play some of the 98/99 era fps with full or reduced details without some stuttering and fluctuating framerate (due to the asthmatic memory speed trasnfert and the low agp implemantation from both the cards tested and the LX MB ?).
Heck starting PIII 733 Mhz on a VIA133 you can really feel the difference but there is always some visible stuttering in Shogo for exemple... WTH?
That should be a complete different matter with a BX chipset BTW... With a good socket 370 MB it sure is.
I Do not have tried myself yet (and I will not be able to... because my LX board died itself after a long storage period) but a 233 PMMX on a SS7 should not be that far from a PII 233 on a LX...

I had an Via133, I got really pissed when I discovered that any newer gpu after 2001 would run only at agp2x and not ag4x that the mb support ¬¬

RacoonRider wrote:

Keep in mind that PII has P6 architecture, in other words, it's predecessor was Pentium Pro and not Pentium. P6 is not a newer/better version of P55c, but rather a very different CPU.

I wonder if my 333MHz Deschutes has a locked multiplier. I have a 440FX board to compare it to a 233MHz Pentium, that would be fun.

if you read the old PII reviews from launch, it states intel took the Pentium Pro, added MMX and fixed 16bit preformance, the question is how do they do with 8bit preformance compared to a Pentium Classic 🤣.

Standard Def Steve wrote:

On the FX board, it's not much faster than a K6-2 300. On the BX board, it's more like a K6-III 450.

I highly doubt it. A 450 K6-2 on a shitty motherboard, maybe, but not the K6-III.

I tested my 450MHz PII on both platforms - 440FX (Compaq Deskpro EB) and 440BX (HP vectra VL? - can't remember exactly) and a K6-III 450@500MHz is faster in most benches. It's not surprising, since the K6-3 has on-die L2 cache running at CPU frequency - and in my case it's also clocked 50MHz faster then the P2. I used a Lucky Tech P5MVP3 board + 2x128MB 100MHz Compaq SDRAM (Hyundai chips) for all three machines.

At times the K6-III will even go toe to toe with the early P3 CPUs (except for FPU benchmarks where it lags behind a Pentium III 450 when not using 3DNOW! patches - about 10-15% difference if I recall).

Synthetics aside, when benching in Quake 2 w/o the 3DNow! patch I remember the PII + 440BX had an advantage when using an AGP card (Creative Annihilator GF256 SDR), but with a Voodoo2 or even Voodoo3 card the K6 is just slightly faster. Using the 3DNow patch widens the gap.

those 3 days before Pentium III came out, the K6-III was the fastest CPU on the market 🤣.

I wrote the wrong numbers down for 3dmark99, even at 350MHz the P2 doesn't get those scores. I was writing notes down on paper over the course of a few days working on it during my lunch hour. The other numbers look more legit but maybe I should redo everything.

1839 3dmarks and 2330 CPU 3dmarks at 233
2669 and 3450 at 350Mhz.

I didn't realize the 686 comparison had all that information in the PDF, I only downloaded some of the graphs until now. Thanks for pointing it out.

kanecvr wrote:

I highly doubt it. A 450 K6-2 on a shitty motherboard, maybe, but not the K6-III. […]
Show full quote

Standard Def Steve wrote:

On the FX board, it's not much faster than a K6-2 300. On the BX board, it's more like a K6-III 450.

I highly doubt it. A 450 K6-2 on a shitty motherboard, maybe, but not the K6-III.

I tested my 450MHz PII on both platforms - 440FX (Compaq Deskpro EB) and 440BX (HP vectra VL? - can't remember exactly) and a K6-III 450@500MHz is faster in most benches. It's not surprising, since the K6-3 has on-die L2 cache running at CPU frequency - and in my case it's also clocked 50MHz faster then the P2. I used a Lucky Tech P5MVP3 board + 2x128MB 100MHz Compaq SDRAM (Hyundai chips) for all three machines.

At times the K6-III will even go toe to toe with the early P3 CPUs (except for FPU benchmarks where it lags behind a Pentium III 450 when not using 3DNOW! patches - about 10-15% difference if I recall).

Synthetics aside, when benching in Quake 2 w/o the 3DNow! patch I remember the PII + 440BX had an advantage when using an AGP card (Creative Annihilator GF256 SDR), but with a Voodoo2 or even Voodoo3 card the K6 is just slightly faster. Using the 3DNow patch widens the gap.

The K6 will be quite slower in 3D Games, which is what really matters in all honesty. The Quake 2 3Dnow! patch is pretty eye opening as to how much special care for a certain CPU can help performance.

TandySensation wrote:

I wrote the wrong numbers down for 3dmark99, even at 350MHz the P2 doesn't get those scores. I was writing notes down on paper o […]
Show full quote

I wrote the wrong numbers down for 3dmark99, even at 350MHz the P2 doesn't get those scores. I was writing notes down on paper over the course of a few days working on it during my lunch hour. The other numbers look more legit but maybe I should redo everything.

1839 3dmarks and 2330 CPU 3dmarks at 233
2669 and 3450 at 350Mhz.

I didn't realize the 686 comparison had all that information in the PDF, I only downloaded some of the graphs until now. Thanks for pointing it out.

I also included the Excel file with locked row and colum headers, which makes finding the data you desire quite a bit easier.

for this discussion, i would introduce a few more models for clock to clock performance comparision:
p55c vs k6-2: p55c has better pure-fpu performance, but due to lack of 3dnow instructions, k6-2 is quite a bit faster in most 3d games and benchmarks.
k6-2 vs k6-3: with the addition of integrated l2 cache, k6-3 is usually 20-30% faster.
k6-3 vs celeronA: despite lack of 3dnow instructions, celeronA's monstrous fpu and p6 bus throughput allows it to easily smoke the k6-3.
celeronA vs p2: p2 has 512kb slow cache, celeronA has 128kb integrated l2 cache, in general p2 is just marginally faster.
conclusion: p2>=celeronA>>k6-3>>k6-2>p55c
typical benchmark scheme to my standard is 3d games with d3d or opengl. glide seems more optimized for k6-2/k6-3 and they can catch up with p2/celeronA, although p55c still gets smoked.

noshutdown wrote:

for this discussion, i would introduce a few more models for clock to clock performance comparision: p55c vs k6-2: p55c has bett […]
Show full quote

for this discussion, i would introduce a few more models for clock to clock performance comparision:
p55c vs k6-2: p55c has better pure-fpu performance, but due to lack of 3dnow instructions, k6-2 is quite a bit faster in most 3d games and benchmarks.
k6-2 vs k6-3: with the addition of integrated l2 cache, k6-3 is usually 20-30% faster.
k6-3 vs celeronA: despite lack of 3dnow instructions, celeronA's monstrous fpu and p6 bus throughput allows it to easily smoke the k6-3.
celeronA vs p2: p2 has 512kb slow cache, celeronA has 128kb integrated l2 cache, in general p2 is just marginally faster.
conclusion: p2>=celeronA>>k6-3>>k6-2>p55c
typical benchmark scheme to my standard is 3d games with d3d or opengl. glide seems more optimized for k6-2/k6-3 and they can catch up with p2/celeronA, although p55c still gets smoked.

actually the K6-III 450 smoked the Pentium II 450, a K6-III 400 as seen nearly caught up and beat a PII 450 🤣

http://www.xbitlabs.com/articles/cpu/display/k6-iii.html

As long as the drivers and game supported 3dnow you where golden

That's the problem: besides quake and the 3dmark99 cpu test there are very few titles that make good use of 3dnow!

In actual games a p2-400 wins hands down to even a k6-3+@600mhz in my testing. I love the k6 systems i have but a p2 is easier to find and faster (and less picky about graphics cards as you can use the all mighty 440BX).
The k6-3+ scores even more cpu 3dmarks than my p3-700, but in games in gets slaughtered by a p3-450...

Also in games, a k6-2 isn't faster compared to a p55c. When you have a 233mhz p55c it will take a k6-2-300 to match it in fpu, so only at higher clockspeed it is faster. For old dos games the k6-2 can be faster and in integer it is king.