34 & 35 & 36 & 37. The next version of PC mark is 2004. Here, finally, Palomino wins decisively: 5% better CPU score, 10% better memory score and again 5% better graphics score! It desperately needs to win more otherwise it will not break even the 5% limit of error between it and Thunderbird!

38 & 39 & 40 & 41. The final version of PC mark is 2005. It requires SSE instruction set so here are the scores for the Palomino 1600. Please note that they will not be taking into account in the final comparison!

42 & 43 & 44. We reached the games section of the tests and the first one is Farcry 1 from 2004 that I recently finished on realistic difficulty. This is mostly a CPU bound game therefore this is a good test to see if either CPU is limiting the performance - which of course they are! I used HardwareOC bench tool to provide accurate and repeatable testing conditions.
1400C

Far Cry Benchmark

The benchmark started at 17.08.2024 17:19:56

System Information
Operating system: Microsoft Windows XP
System memory: 2,0 GB
CPU: AMD Athlon(tm) processor
CPU speed: 1400 MHz
Sound system: : NVIDIA(R) nForce(TM) Audio
VGA Information
Graphics card: NVIDIA GeForce 7600 GT
Memory: 256.0 MB
Current GPU speed: 560 MHz
Current memory speed: 700 MHz
Driver version: 6.14.0010.9371 (English)

Resolution: 1024×768
Ultra quality option, Direct3D renderer
Level: Volcano, demo: hocvolcano.tmd
Pixel shader: default model
Antialising: 8×
Anisotropic filtering: 16×
HDR: disabled
Geometry Instancing: disabled
Normal-maps compression: disabled

Score = 43,23 FPS (Run 1)
Score = 42,05 FPS (Run 2)
Score = 42,71 FPS (Run 3)
Average score = 42,66 FPS

Resolution: 1280×1024
Ultra quality option, Direct3D renderer
Level: Volcano, demo: hocvolcano.tmd
Pixel shader: default model
Antialising: 8×
Anisotropic filtering: 16×
HDR: disabled
Geometry Instancing: disabled
Normal-maps compression: disabled

Score = 43,24 FPS (Run 1)
Score = 42,36 FPS (Run 2)
Score = 42,90 FPS (Run 3)
Average score = 42,83 FPS

Resolution: 1600×1200
Ultra quality option, Direct3D renderer
Level: Volcano, demo: hocvolcano.tmd
Pixel shader: default model
Antialising: 8×
Anisotropic filtering: 16×
HDR: disabled
Geometry Instancing: disabled
Normal-maps compression: disabled

Score = 41,59 FPS (Run 1)
Score = 40,20 FPS (Run 2)
Score = 41,32 FPS (Run 3)
Average score = 41,03 FPS

Copyright 2002 - 2006 Zoltan Nemeth - Roadside

Athlon XP 1600

First game and the first victory for Palomino with 5% lead against Thunderbird across al resolutions!
Up until now the 10% advantage seen in period correct reviews is nowhere to be seen!

45 & 46 & 47. The second game is the legendary Half life 2. This is the 2004 version and not the updated Steam version!
I tested on the first level right in front of the train station where Dr Breen is making his speech using max details as the geforce 7600gt is more than capable of handling this game.
I used Fraps 1.9D because HardwareOC bench tool refused to work and because later versions of Fraps require SSE instruction set. I am sorry for the lack of screenshots - Fraps 1.9D outputs the results to a .TXT file.

1400C
1024*768
2024-08-17 19:44:46 - hl2
Frames: 2779 - Time: 82593ms - Avg: 33.646 - Min: 13 - Max: 47

1280*1024
2024-08-17 19:37:30 - hl2
Frames: 1995 - Time: 61766ms - Avg: 32.299 - Min: 21 - Max: 45

1600*1200
2024-08-17 19:43:11 - hl2
Frames: 2246 - Time: 71360ms - Avg: 31.474 - Min: 24 - Max: 46

Now is the turn of the Palomino 1600 in Half life 2:

1024*768
Min: 27
Max: 58
Avg: 41.55

1280*1024
Min: 27
Max: 53
Avg: 37.95

1600*1200
Min: 26
Max: 43
Avg: 33.417

Finally Athlon XP 1600 manages to win decisively! It crushes the Thunderbird 1400C with a 23% lead at 1024*768 and 17% at 1280*1024!
The game is actually playable on the palomino 1600 - as you can see the minimums are just below 30 FPS - whereas on the Thunderbird you can not play Half life 2!

48 & 49 & 50. The third and final game is Doom 3 which does not run on Thunderbird once you install Resurrection of evil and the final 1.3 patch! However on Athlon XP 1600 it runs even if you can not actually play the game because of the very low minimums!
These are the results on Palomino at 1024*768, 1280*1024, 1600*1200 ultra quality, AA = off!

51. We return with a synthetic test: Super Pi where a shorter the time to calculate 1 million decimals constitutes a better result!
Again Athlon XP scores another decisive win: 10% better than its predecessor!
1400C
Palomino 1600

52. The final test is Cinebench 2003. It also requires SSE so the only screenshot is with the Athlon XP 1600:

As I forgot to take screenshots with the obligatory CPU-Z and system info from Everest here they are after quickly dropping in the XP 1600:
You can see the memory timings are as tight as possible: 2-2-2-5 1T!

As I recall, you need to set timing to 2-2-2-8 on Nforce 2 chipsets to achieve better results (quirks of the memory controller).

You are right! I remember forum posts from back in the day mentioning some odd timings on the nforce2 in order to achieve better results!
I did practical tests with different timings. With the exception of the CAS latency the rest of the timings have practically no influence on the scores! Not even 0.5%!! Because I tested at multiple resolutions and level of details not even the command rate seems to influence the results. As soon as I tested anything but default/lowest settings - example 3dmark 2000 at 1024*768 with no AA and all settings at standard including 16 color depth - the influence of the memory timings is 0.

And there you have it! Palomino and Thunderbird went head to head one last time with the best hardware and the best drivers to determine if the performance advantage of the second generation K7 is worth it!
Based on the tests completed by both processors Palomino is better by 5.34% at the same clock speed! Barely above the margin of error and far less than all period correct reviews!
The question is: how is that possible? Surely the reviewers from Tom's or Hexus did everything right.
Their results reflect the hardware used: 256mb of RAM was the norm - https://hexus.net/tech/reviews/cpu/253-amd-xp2000-cpu/ - making XP heavily dependent on the page file which I was very careful to avoid in all tests by disabling it; geforce3 was the best video card available when Palomino launched making very likely a video card bottleneck; I used a geforce 7600gt that eliminates such a possibility.
I used tests that range from DirectX 7 to 9, with some of them very tough on the CPU - all 3 games tested are practically "eating" every single socket 462 CPU for breakfast!
The final question is: You have a Thunderbird in your retro system. Should you upgrade from a Thunderbird to a Palomino? The answer is NO. The TDP is very high leading to high temperatures and the need for a PSU with a strong 5V rail.
All boards that can take a Palomino can also take a Thoroughbred with the same frequency. If you motherboard supports only 266 MHz FSB and you have a 1 GHz Thunderbird than don't go for an Athlon XP Palomino 1800! Go for a T-bred 2000 - an easy and very cheap upgrade that will vastly improve the performance of your system while consuming less so you don't need a turbine cooler or a 5V heavy PSU.

After testing all socket 462 chipsets - with the exception of SIS chipsets for which Abit did not manufactured a single motherboard , all generations of the K7 architecture for socket A - Thunderbird, Palomino, Thoroughbred A+B, Barton (with which i started testing back in 2018); all possible FSB available: 200, 266, 333 but not 400 MHz because there are no 2 socket 462 processors with the same clock speed only a single topic remains to be covered on socket A!

CACHE SIZE INFLUENCE ON PERFORMANCE ON SOCKET A/462.
There are 3 possible variants for the Level 2 cache size on socket A: 64kb, 256kb, 512kb. LEVEL 1 cache on all processors for the socket is actually very generous: 128kb for data and instructions; by comparison Pentium 4 has only 16kb!
A very big advantage of the K7 is the use of an exclusive cache design where the L1 cache's contents are not duplicated in the L2, thereby increasing total cache size, in stark contrast with the Pentium 4 who utilizes an inclusive cache design where data from L1 are duplicated in the L2.

1. 64kb: The only processors with this tiny amount of L2 cache are the Durons. Released as a budget processor maybe AMD cut the cache a bit too much! There are 3 generations of Durons: first one is Spitfire, second one is Morgan - ads SSE and data prefetch - and the third is Applebred - switches to a 266 MHz FSB. I got the 1300 MHz Morgan and the 1800 MHz Applebred. I chose to test the final model with the highest clock speed: 1800 MHz.
There is an exact match - same clock speed, FSB and multiplier - in the Athlon XP lineup: Athlon XP Thoroughbred 2200 part number AXDA2200DUV3C with 256kb of L2 cache.
The first series of test will practically test 64kb versus 256kb of cache. Also it seems that the final Duron, the Applebred version, was not reviewed by any major site, as I could not find a single review!

2. 256kb: the vast majority of socket A processors have this amount of L2 cache. Thunderbird, Palomino, Thoroughbred A+B, Thorton, all have 256kb. The performance difference between Palomino and Thunderbird is a little over 5%, Thoroughbred has the same performance as palomino (< 3% difference), Thorton is a barton with defective cache disabled which makes him a Thoroughbred in all but name so it has identical performance so i did not tested it.
The Athlon XP 2200 will go up against Duron 1800 to see the advantages that the 256kb of cache brings to the table!

3. 512kb: the only Athlon generation with this amount of L2 cache is the Barton. Representing the pinnacle of socket 462 these CPU's are sought after by anyone who wishes to build an ultimate socket A system. In reviews of the era Barton was not considered worthy of its performance rating: https://web.archive.org/web/20031210060054/ht … lonxp-3000.html or https://www.guru3d.com/review/amd-athlon-xp-3000-review/. Many reviewers said that the PR was unjustifiably inflated.
We shall see if that still stand true today using better hardware and software than what was available in 2003.
There are 2 processors in the Sempron lineup with the same clock speed, FSB and multiplier that match perfectly: 2000 MHz, 333 FSB and 12X multi. Sempron 2800 part number SDA2800DUT3D with 256kb of L2 cache is based on the Thoroughbred core and Sempron 3000 part number SDA3000DUT4D with 512kb of L2 cache is based on the Barton core.
These final tests will put head to head Barton versus Thoroughbred, 512 vs 256 KB of cache.