51. At 1600*1200, against all odds, Athlon XP scores its only victory! I repeated the test many times and every time XP 3200 manages a very good average at this particular resolution every single time. I do not have an explanation for this, Barton + nforce2 simply performs better than Venice + VIA at this extreme resolution! The score is consistent with the one obtained 5 years ago when I tested socket A.

A. Athlon XP 3200
B. Athlon 64 3200

Off the record results for Quake 4 and FEAR.
I tested both games however the final results will not include them because these 2005 games are too much for these systems.

Quake 4

ATHLON XP 3200
ATHLON 64 3200

Now it would seem based on these automatic tests that Athlon 64 has a better chance of playing Quake 4! This is not true; when actually playing the game on socket A system I put in an extra 1gb of ram which is not possible on the socket 754 system. Playing the game with the Athlon XP feels much smoother than on Athlon 64 because you do not need to use the page file - which results in freezes every 10 seconds or so that makes the game unplayable. granted the framerate is very low but you do not get that horrible hiccups that you get on socket 754!
But there are socket 754 boards with 3 memory slots - many of you will think! yes, there are - I personally tested Asus K8N-E DELUXE that my best friend has with 3 1GB dimms and it passes Memtest. However the memory frequency is DDR-333 which results in a penalty in every single game/synthetic test! So you really want 2 1GB dimms in your socket 754 system even of your board can take 3!

FEAR - 1024*768 max details except SS which are disabled:
ATHLON XP 3200
ATHLON 64 3200

The poor geforce 7800gs was screaming for mercy the entire time! You really need GTX260 at a minimum if you want to play FEAR - remember SS are disabled otherwise you will get a slideshow!

FEAR - 1280*1024
ATHLON XP 3200
ATHLON 64 3200

FEAR - 1600*1200
ATHLON XP 3200
ATHLON 64 3200

There is such a strong GPU bottleneck that both systems are practically equal! These tests are simply for your viewing pleasure; they will not appear in the final result!

All games testing is done, now there are only 3 tests left!
46. SUPER PI - only CPU matters in this one!
ATHLON XP 3200
ATHLON 64 3200

A shorter time constitutes a better result! Athlon 64 wins again!

47. Cinebench 2003 is the second to last benchmark:
ATHLON XP 3200
ATHLON 64 3200

14% lead by the newer K8 CPU! In line with period correct reviews!

48. Final test is the infamous Tom2D! VIA won overwhelmingly against nforce3 and ULI in the first 2 rounds of tests! Because I misplaced the screenshot for the Athlon XP 3200 system and I am away, I put up again the result from 2021 when i tested it for the first time as you can see in this link: Re: Socket A: Nvidia vs Via - battle of the platforms!

ATHLON XP 3200
ATHLON 64 3200

So no loop sided victory for either platfrom in this final test!

The final screenshot is from Everest 5.50 cache/memory benchmark - this one was requested by AlexZ. Will take one from Athlon 64 when I return home!
Athlon XP 3200

Time for the rankings: It is obvious that Athlon 64 won, the only question is by how much?
Athlon XP 3200 is awarded a 100% score and represents the baseline!

ATHLON 64 3200 VENICE: 114.64%
ATHLON XP 3200 BARTON: 100%

Modern retro reviews put the Athlon 64 at a far greater advantage. Because of heavy reliance on 3d performance the new micro architecture does not distance itself so much from the old one - retro gaming puts a very heavy emphasis on video card performance. I believe that - like myself - most people building a socket A or 754 system will most likely use it for games so i consider that the accent on 3d performance by choosing many 3d tests and very few CPU-only benchmarks is correct and in the real world the performance difference by choosing the Athlon 64 is about 14-15% when you will game on it when compared to a socket A system.
There is another advantage not seen in the screenshots: all socket 754 boards will use the 12V rail so no more chasing PSU with strong 5V rail - any modern PSU will suffice!

So K8 starts with a very good but not huge advantage over the K7! I wonder how dual channel would help the Athlon 64 at the same clock speed and L2 cache amount....

I written all the results in the original Excel file:

nd22 wrote on 2025-12-12, 09:38:

I written all the results in the original Excel file:

Cool, but in proprietary Excel...
Maybe online in Google Sheets?
Well, or in PDF format, all browsers know how to open it.

Well, that is because I payed an arm and a leg for Office 2003 retail back in 2005 and I learned to use it pretty well! In fact even today at work we are not allowed to use docx, xlsx formats only the old ones so everything stays compatible! I will try google when i get back home.

Before continuing to part 3 there are 2 things to mention:
1. how much does MSI afterburner and FRAPS impact the results?
Everything that you have seen is with both AB and FRAPS running. this will lead to a performance penalty!
Example: 3dmark 2001
A. no MSI AB + FRAPS
B. MSI AB + FRAPS

There is a 5-6% performance hit with AB and FRAPS running present in all the screenshots. Because I manually checked the frame-rate of both Ab and fraps after each run I let them ON at all times.

2. your scores are too low! in a recent video Nathan from PIXEL PIPES - https://www.youtube.com/watch?v=DJ3ms_4J3TQ - compared 2 vastly inferior graphic cards, radeon 9800XT and geforce 5950 ultra and they had extraordinary performance. For example in Farcry radeon 9800Xt got 89.5 FPS on the volcano map; on the very same map with a far stronger geforce 7800gs I got 86.13 FSP, lower than 9800XT.
I also got the radeon 9800xt and I tested it on KV8 pro before putting in the X1950 pro and the score I got is less than half what Nathan in his video got! In fact the moment you set maxed out settings as you can clearly see in the video you kill the radeon 9800XT. The moment I select ultra quality with max AA&AF it is all over for the radeon! How could a far stronger video card get lower FPS it's a mystery.

Excellent review.

It shows that despite Athlon XP being generally slower, it does have an advantage of using 3 (or perhaps 4) memory modules and thus having 3GB RAM at DDR400 speed. I used to have an nForce 2 board with 4 DIMMs.

Athlon 64 with AGP is about 15% faster at the same clock speed and cache, but there are 2.4Ghz and 1MB cache models that would push it about 8-9% further without being too expensive. 23-24% faster is noticeable in games. 2.4Ghz 512kb L2, 2.2Ghz 1MB L2 are the CPU models worth getting or one has to OC slower model. Besides having the 12V P4 CPU power connector, another advantage is quiet CPU cooler availability. You can replace the fan on stock s462 heatsink with Noctua but they have low airflow.

When it comes to the issue of Quake 4 memory swapping, I have not observed it with 2GB RAM as long as other apps are closed and GPU has sufficient memory. Probably at least 512MB is needed, GTX 260 has 896MB and it is definitely sufficient. Quake 4 and FEAR are not something one would play on AGP, even though details could be reduced to make it work.

Thank you sir.
Regarding the difference between different models the difference is really small. There is no more than 5% difference between the 3200 any revision and 3400 any revision! performance between all socket 754 Athlon 64 models is really close! That means that Athlon 64 3400 is around 20% better than XP3200 - a significant but not extraordinary lead.

Now after part 1 - VIA's revenge and part 2 - family fight it is time for the final part:

PART 3: the way it's meant to be...!

The vast majority of socket 754 boards you are going to find out there are AGP. There are a few PCI-express based boards mainly based on nforce4 chipset and some exotic ones. Abit manufactured a single board with PCI-express: NV8. Based off nforce4-4x it is the only model found in the Abit lineup for socket 754. It has only what the chipset offers and nothing else but it is more than sufficient! I already described the system with NV8 and all the settings earlier in the topic; now only a short summary.