Celerons? New Celerons, like the Conroe-L? If so, they go above Sempron and Athlon64, pretty much a clock parity with 1MB L2 cache Althon64s (such as the Athlon64 FX).
Also, I gotta defend P4. Sure, the Williamette P4, with a 400MHz fsb, 256kb L2 cache, and no hyperthreading may not be faster than a Pentium III clock for clock, but if we took an HT enabled Northwood or Prescott or Cedar Mill and pitting it against a PIII, the P4 would win, hands down. More L2 cache, and a 10 to 20% performance boost thanks to HT, not to mention SSE2 and SSE3 (Prescott and Cedar Mill). After all, we are using modern benchmark applications, aren't we?
Here's the list:
K5
Pentium
K6
Pentium MMX
K6-2
Pentium II
K6-III
Celeron (Mendicino<Coppermine)
Celeron (Pentium 4 Williamette/Northwood)
Atom
Pentium 4 (Williamette<Northwood)
Celeron (Tualatin)
Pentium III (Coppermine)
Celeron D (Prescott<Cedar Mill)
Athon (Orion/Pluto)
Duron
Pentium III (Tualatin)
Athlon (Thunderbird)
Pentium 4, HT enabled (Prescott<Northwood<Cedar Mill<Gallatin)
Athlon XP (Palomino/Thoroughbred A/B<Barton)
Atom Dual Core
Sempron
Pentium M/Celeron M (Banias/Dothan)
Athlon 64
Pentium D
Celeron (Conroe-L)
Athlon64 FX (or 1MB L2 cache Athlon64)
Core-Solo
Celeron M (Merom)
Athlon X2
Core Duo
Core 2 Duo
Phenom
Core 2 Quad
Phenom II
Core i5 (guess)
Core i7
On this list, it's pretty tight from Pentium II to Celeron, and then again from Celeron Tualaton to Athlon T-Bird. They all perform very much equal to each other on a clock for clock basis. Once again, when we come to Pentium M to Athlon FX the performance would be very close, so it's hard to say in those groupings which one will be better than the other for absolute certainty. However, I've grouped these based on the assumption that all benchmarks are multi-threaded. If single threaded benchmarks are used, the performance of anything SMT or multi core would drop significantly. Now this list is once again based on the assumption that all CPUs were running at 1.0GHz. Everything else, such as the FSB, ram type available to the platform the CPU would have to run on, would remain unchanged, so this is most definitely a list of IPC of a given platform, as I'm sure some of these CPUs are held back by slow FSB, off die cache, and slow ram. Since some of these CPUs are capable of operating at much higher frequencies (e.g. the P4), an absolute performance chart for the CPUs given factory clock speeds would end up looking quite different.