As AlexZ correctly anticipated the final part of the first chapter is:

CHAPTER 1 PART 3: performance evaluation of DESKTOP single core Athlon 64 revisions for socket 939!

Athlon 64 has the highest number of revisions out of all AMD classic era (1998 - 2008) CPU's. K6 got 4; K7 got 7: Argon, Pluto and Orion, T-bird, Palomino, Thoroughbred, Thorton, Barton. I tested them all in the first battle of the platforms.
K8 got no less than 9:
1. Clawhammer - the very first Athlon 64 released for the consumers back in 2003; also used for the first 2 FX CPU's: FX-53 and FX-55;
2. Newcastle - AMD downsized the chip to make it cheaper to manufacture;
3. Winchester - trial run for the 90nm manufacturing process;
4. Venice - large scale manufacturing on 90nm;
5. Manchester - dual core with 1 of the cores disabled;
6. San diego - the base for the last 2 FX CPU's: FX-55 and FX-57; highest clocked single core Athlon 64 and the easiest to overclock;
7. Toledo - again a dual core with the defective one disabled;
8. Orleans - single core Athlon 64 for socket AM2 which are not the target of this comparative;
9. Lima - also for socket AM2; energy efficient models.

AlexZ wrote on Yesterday, 06:38:

I didn't have any background apps.

On screenshots of PC Mark I mostly see Irfanview, Total commander, Everest, those should not influence the score much. There are many system processes normally running. It depends on what they do.

On most of the screenshots you can see afterburner running: example Doom 3; Aquamark 3 and so on. I use fraps to double check the results - fraps is set to record the framerate and after each test I compare afterburner with the fraps log.

In order to test as many revisions of Athlon 64 as possible we have to find a good candidate, that is a CPU that is manufactured from the very first revision - Clawhammer - up to the last - Toledo!
That CPU is Athlon 64 3500 that is available in 6 out of the 7 revisions for socket 939. Only Toledo does not have an Athlon 64 3500.
So we have 6 different 3500 to test, starting with Clawhammer and finishing up with San diego!

Perhaps test CPUs with the same frequency rather than the same rating. Interestingly, there is no ClawHammer 2.2Ghz 1MB L2 cache for s939. So AMD gave us 2 IMCs and took 512kB cache from us so the performance benefit was gone. 4000+ would have been very rare as 3700+ in s754. 1MB L2 cache only returned with San Diego. The main advantage was high clocked FX CPUs if you had the money for them.

While many have been bought second hand or I received them for free, I managed to score a NOS Winchester 3500:

AlexZ wrote on Today, 06:33:

Perhaps test CPUs with the same frequency rather than the same rating. Interestingly, there is no ClawHammer 2.2Ghz 1MB L2 cache for s939. So AMD gave us 2 IMCs and took 512kB cache from us so the performance benefit was gone. 4000+ would have been very rare as 3700+ in s754. 1MB L2 cache only returned with San Diego. The main advantage was high clocked FX CPUs if you had the money for them.

All 3500 have the same 2.2 GHz and 512 Kb of L2 cache so they are identical specs wise. 3500 is the most common and present across 6 of the 7 revisions for socket 939. Perhaps you have a better idea?

The 5 musketeers ready to fight ( I am sorry about the photos, I used my phone):
Left to right:
Clawhammer - Newcastle - Winchester - Venice - Manchester - San Diego(not pictured)

And slightly better group photos arranged by manufacturing process:

If all have the same 2.2Ghz and L2 cache then it's fine. I was worried AMD pulled a trick on us and labelled a lower clocked CPU as 3500+. We know how unreliable the rating is. 2.2Ghz was indeed most common and most likely candidate in retro s939 systems.

Yes, I know! 3500 does not warrant its rating actually! I selected it because it is the most common model among all Athlon 64 revisions! We just found out after 2 rounds of tests that dual channel = 200 PR points when compared the same CPU to its socket 754 counterpart and 512 Kb of Level 2 cache = 200 PR points! After part 3 I will try to put up a new PR scale for Athlon 64's based on the tests that you have seen!

The platform selected is another - you guessed it! - Abit board. In order to cycle through the motherboards in storage I selected KN8 SLI as the board of choice for the next tests! When we will get to Athlon 64 X2 you will get "hit"by a fatality 😀.
Also in order to investigate the suitability of single core Athlon 64 for socket 939 and socket 754 for late XP-era gaming I decided to use 4gb of RAM in order to test some games like FEAR, Quake 4, Serious Sam 2 that are known to require large amounts of memory - by the way this will also investigate the frequency, timings and command rate for various Athlon 64 revisions!

The board up close and personal; please note the very good placement of the main 24 pin ATX connector; the same Winbond monitoring chip as the KN8; the Vitesse pci-express based gigabit LAN controller:

The board supports SATA 300 so the SSD should get quite a speed boost! The small card onto which you see written Abit SHADOW CARD is mandatory to be used in a single card configuration otherwise the system will not post! Good luck finding one!
In order to not be limited by some ordinary memory sticks and to get the best timings and frequency possible i selected one of the best kits for socket 939:

I selected arctic cooler freezer 7 pro for cooling duties, some SATA DVD RW and a geforce 7600gt to test install the system:

This particular board does not have the micro GURU chip ( actually a reprogrammed winbond) so no fancy fan and voltages control in the BIOS. Also it is mandatory to connect the molex underneath the last PCI slot to the power supply!