58 & 59 & 60. We reach the last year of the XP era: 2006. Thinking about it I wonder wherever a Vista era existed because XP remained the dominant OS up until Windows 7!
The first game is Prey. Built on IDTECH4 engine it is much easier to run than Quake 4. It fully utilizes 2 cores - 70-90% during gameplay - however the killer setting, anti-aliasing, can only be set at 4X maximum. This means GTX 285 has a much easier task this time around so we are not GPU bottle-necked but we are CPU-limited!
Again a good pci-express implementation is mandatory when you want high performance - you may see where this is going!
I used HardwareOC benchmark tool again and tested using the first demo. It does not use sound. Retested just this morning to confirm the first 2 series of results.

Prey - 2006
A. nforce4
B. K8T890

NVIDIA takes no prisoners and kills off VIA in Prey! I would say a pattern emerges: in opengl games based on IDTECH4 nforce4 is far in front while in Directx9 games is equal to K8T890.
This game is very easy to run on any socket 939 system as long as you got a good video card: GTS 250 1gb recommended.

61. I specifically selected Company of heroes to prove a point: games that bring a specific era system to its knees are not first/third person shooters but RTS/simulation games. Just like Age of empires II is unplayable on any CPU released where the game was released such as Pentium 3/Athlon Thunderbird Company of heroes proves to be very tough on the poor Athlon 64 X2 4800. The game utilizes both cores to about 70-80%; as more and more units are produced the processor is overwhelmed and the frame-rate drops dramatically.
I tested using the builtin benchmark which is NOT representative of the gameplay and is more of a graphic card test; everything is already calculated for every single unit. During actual game the CPU will calculate all movements, hit detection, etc for every single unit - that makes the game so difficult for the Athlon.

1024*768
A. nforce4
B. K8T890

62. 1280*1024
A. nforce4
B. K8T890

63. 1600*1200
A. nforce4
B. K8T890

There is 2-3% better performance on nforce4 - nothing to write home about!

64 & 65 & 66. And the last game tested is Microsoft Flight simulator X - original release from 2006 upgraded with SP1 & SP2. Again specifically chosen to demonstrate that 2006 is really the end of the road for socket 939!
The game is simply destroying the Athlon 64 X2 - it does not matter if you got FX-60 you will not be able to play this game at maximum details. Please note that maximum details are FAR beyond the ultra quality found on the main settings screen. I detailed exactly what maximum means in this post: Re: Battle of the platforms: socket 939!
With the plane flying just above ground level both cores stay at 100% so it is dual core aware. I have chosen the free flight with the Air Creation Trike Ultralight on Friday harbor for testing. I let the game load; waited 1 minute before pressing F11 to test with FRAPS; I did not touched the controls, the plane flew in a straight line, no right or left, no up or down.

1024*768
A. nforce4
Frames, Time (ms), Min, Max, Avg
801, 60000, 11, 15, 13.358

B. K8T890
Frames, Time (ms), Min, Max, Avg
901, 60000, 12, 18, 15.014

1280*1024
A. nforce4
Frames, Time (ms), Min, Max, Avg
780, 60000, 11, 14, 13.008

B.K8T890
Frames, Time (ms), Min, Max, Avg
854, 60000, 11, 16, 14.235

1600*1200
A. nforce4
Frames, Time (ms), Min, Max, Avg
742, 60000, 10, 14, 12.378

B. K8T890
Frames, Time (ms), Min, Max, Avg
716, 60000, 8, 17, 11.933

And here is the one and only victory for VIA: it manages to best nforce4 by 12% at 1024*768 and by 9% at 1280*1024; meanwhile NVIDIA comes back on top at 1600*1200 with a small 3% lead.
It feels like a hollow victory, nobody is going to play FSX on a socket 939 system, but, still, a win it's a win.

While some of you might think: but i can play at 15 FPS - be aware that this results are from a level flight. The moment you want to land and come closer to the ground both CPU cores spike at 100% and the frame-rate plummets into the single digits! I con no longer control the plane despite using a genius F31 joystick at 5 FPS!

For one last time VIA and NVIDIA went up against each other to determine the best chipset available for socket 939. Already on socket 754 nforce4 came on top, even if by a tiny margin, thanks to VIA excellent performance in TOM2D. This is not the case here as both nforce4 and K8T890 have serious issues with GUI acceleration.
Back in the day the 2 chipsets were equals in 99.99% of all reviews done by various websites. Today nforce4 won many benchmarks, sometimes by leaps and bound above the competition! Let's see the standings!
VIA gets a 100% base score and NVIDIA is compared to it:

NVIDIA nforce4: 104.24%
VIA K8T890: 100%

So nforce4 did not managed a clear cut victory in the performance department; both chipsets are within the 5% limit of error.
But if we take into consideration the whole package: performance; motherboard layout which is many times better on the board used - Abit AN8 ultra; compatibility with SSD's and SATA optical drives; compatibility with late pci-express video cards - radeon 4890 2gb works fine on nforce4 - large video cards such as the GTX 285 that I used fits just fine while on AX8 requires a pci-express riser we got a clear winner: nforce4 is the platform of choice for both socket 754 & 939!

There is one caveat to all of the above: compatibility with Windows 9X! I did not succeed in installing Windows Me on Abit NV8 and gave up after 2 attempts. After searching the forums it seems that it is quite difficult to get 98SE installed and running correctly on this chipset.
That being said i consider socket 939 far beyond what I would require to play my Windows 98 games. Windows XP and by extension, Windows 2000, fits perfectly and is running like a dream on such a system.

For VIA this is the end of the road. Yes, there are still some leftovers on socket AM2 such as Asus M2V, but for all intents and purposes this is the final chapter for VIA as it leaves the chipset arena with a whimper instead of leaving with a bang!
Also ULI is purchased by NVIDIA - so that's it.
NVIDIA remains the only player standing and practically monopolizes the AMD chipset market with various iterations of nforce4!
That means one thing: nforce4 is the platform of choice for the final part of this comparative!

Chapter 2, part 2: Performance rating evaluation of dual cores Athlon 64 X2 for socket 939

Socket 939 is the first consumer platform for which AMD introduced dual cores CPU's. Paper launched in May of 2005 and available later, the prices were extremely high, starting with 500 USD for the cheapest model: 4200. That made adoption quite difficult - as a result today dual cores X2 for socket 939 are far rarer than any single core model and cost considerable more.
Introduced where most programs and particularly games were considered to be single threaded most people consider them suitable fro productivity use and not 3d graphics. Today we know this is not the case; I already demonstrated there are 2005 games that can take advantage of dual cores: Serious Sam 2,Quake 4, Call of duty 2. The 3 games from 2006 that I tested, are all dual core aware and make excellent use of both cores in some cases: Company of heroes for example.
Unlike the 7 revisions of single core Athlon 64, dual cores are available in only 2 revisions: Manchester and Toledo. Like the single cores before them in practice there is no difference between them performance wise.

During single core testing on both socket 754 and 939 I demonstrated that: a) 512kb of L2 = 200 PR points and b) 200 MHz = 200 PR points.
Lower models respect the rule however the 3500 and 3800 model break it and artificially inflate the PR by 100 points each.
For dual cores AMD decided that 512kb of L2 cache make up for 200 PR points just like in the case of single cores, however 200 MHz are now 400 Pr points: from 3800 to 4200 and from 4400 to 4800 there are only 200 MHz separating them yet there are 400 PR points between them - practically double from the single core ones!
We shall test exactly that! How much performance difference is between dual cores and how the PR scale stands up.
To make things even more equal I got only the Toledo CPU's: all got the E6 stepping; the lower ones got a 89W TDP and the higher ones got a 110W TDP - this is not a problem, I used the 4800 and it was surprisingly easy to cool with an arctic freezer 7 pro!

And here they are, the 5 musketeers (while I as under the impression that i got all of them I am missing the Athlon 64 X2 4600):

Despite that I got both models with 2*1mb of L2 cache and 2 out of 3 models with 2*512kb of L2 cache so I think there is still possible to determine the performance of the missing model by comparing the results obtained with the 4 I got.

Also in order to spice things up and to show off my collection I am going to switch the motherboard from AN8 ultra to: