Games tested but not included in the suite:
- nfs underground 1&2: it seems that is very hard to test any nfs game because I could not get repeatable results. Every single run got different average and minimum fps so I gave up after 2 days
- chronicles of riddick escape from butcher bay: one of the best games ever made, a real gem. I could installed it on every single system however on KU8/ ULI M1689 the game refused to run, it simply did not start without any error message.
- age of empires 3 from 2005: exceptionally tough on the video card, no agp GPU can run that game at max settings so it was dropped. In fact 2004 seems to be the limit for AGP and socket 754. It would be interesting to see what a sttronger pci express video card can do to help socket 754 stay relevant one more year.

I tested Age of Empires 3 and it has very buggy graphics settings. I cannot change resolution from 1024x768 or enable anti-aliasing. Settings have no effect and get reverted on next start. With everything else set to max a random skirmish game initially ran with about 50 fps, but as I constructed more buildings and explored the map it kept dropping, until I got invaded by a large army and at that point it was 25 fps.

I completed the first mission of FEAR in 1600x1200 with maximum settings including anti-aliasing and most of the time I got 72 fps which is the refresh rate (vsync on). Very rarely it dropped to 50-60 fps. No extreme swings in fps.

Pcie can definitely extend the life of s754 by one year until 2005. You need GeForce 9800 GT / 250 GTS or GTX 260 / 275. For AGP I agree it is 2004.

So before putting up the results a short description of how this is going to work.
I divided the comparative into 3 parts:
Part 1 AGP.
Part 2: K* vs K*
Part 3: pci express

Because the majority of socket 754 board you are going to find are going to have an AGP slot part 1 is going to have 2 video cards:
1. Geforce 7800gs
2. Radeon x1950 pro.
An important objective is to find a GPU that is a good match for the Athlon 64 so there is a balance between the processor and the video card.

Even an overpowered GPU, especially if bought cheap can be beneficial in high resolutions or maxxed out anti-aliasing. Most practical resolution for Athlon 64 is 1280x960 / 1280x1024 when using CRT or higher when using LCD (depends on game support).

Support for year 2005 can be interesting because a number of popular games were released: Age of Empires 3 (all series), Silent Hunter 3, Civilization 4, Quake 4, Need for Speed Hot Pursuit 2, FEAR. A GPU that can handle these can also handle NHL 2007, FIFA 2007, Virtua Tennis 3.

Maxxed out Far Cry (2004) in 1600x1200 could challenge AGP.

The Sims 2 (2004) with all updates is a big challenge too and tends to suffer from occasional music cut offs even when using SSD due to CPU overload. I doubt it runs well on AGP + magnetic drive.

Dividing the comparison into 3 groups makes sense as there are multiple questions to answer.

AlexZ wrote on 2025-10-17, 20:17:

Athlon 64 3400+ with GeForce GTX 275 in F.E.A.R. First Encounter Assault Recon (2005) in 1600x1200 it reaches 96 fps average with maximum details except for FSAA which was off. With FSAA set to 2x, average fps drops to 84. With FSAA set to 4x, average fps drops to 76.

Though I've only played just a bit of it, FEAR on my Socket 939 system gets just under 60 FPS with the in game performance test. This is with auto detect settings and the resolution set to 1152x864. The CPU is overclocked to just under 3,000 MHz, a GTX 260, and WinXP with two gigs of RAM.

What is your configuration exactly? No directx9 gpu can pull those numbers.
I prefer dividing the period for which a specific system is best suited for in years because it is much easier. Yes, some games released much later can still be played on a socket 754 system for example but if there are games released the same year that I can not play at max details on the same system than I consider that particular year the cutoff year.

At 3Ghz those numbers in FEAR are possible, but FSAA was probably off and the resolution is also low. My FEAR numbers are average FPS with absolutely maximum graphics quality and in 1200x1600. I didn't use auto-detect settings, but set everything manually. GTX 260 is a good match for s939.

Athlon 64 X2 6000+ (Windsor, 3Ghz) with Gainward GeForce GTX 260 Golden Sample (625 GPU, 1100 memory, 1350 shader), NVidia driver 177.83 got 119 fps average in FEAR in 1600x1200 with everything set to maximum except for FSAA which was off. Re: Any love for AM2?

I figured out how to configure graphics settings in Age of Empires 3 so they apply so I set everything to maximum and resolution to 1280x960. I played one skirmish and the lowest FPS seen was about 30 during big battles. In most of the game it was much higher. It is definitely playable.

A quick recap is im order before going with the results.
Common hardware for the 3 systems:
Athlon 64 3400
2x1gb ddr at 3-3-3-8 CR 1T
LEADTEK geforce 7800gs
Wd 150gb raptor
Enermax eg701ax PSU

System 1: abit NF8 PRO motherboard with nforce3 250gb chipset named from now on NF8.
System 2: abit KU8 with uli m1689 chipset named from now on ... KU8. SSD 120gb as main drive, wd raptor 150gb as secondary.
System 3: abit KV8 PRO board with k8t800 pro named from now on KV8.

nd22 wrote on 2025-10-22, 05:24:

What is your configuration exactly? No directx9 gpu can pull those numbers.
I prefer dividing the period for which a specific system is best suited for in years because it is much easier. Yes, some games released much later can still be played on a socket 754 system for example but if there are games released the same year that I can not play at max details on the same system than I consider that particular year the cutoff year.

Here are screenshots of the settings. It defaults to 1024x768 but I manually set the resolution to 1152x864.

So it isn't maximum details. FSAA is off as I expected, soft shadows are off, just 4x texture filtering, not 16x, volumetric light density is just medium. More relevant options are in "Advanced Computer Options" -> "Physics" and "Effects" . There are too many options and unless you set everything to max it's hard to compare.

In this topic, we should be focusing primarily on s754. Having results for comparison with very similar Socket A or s939 configuration can be interesting, provided we use the same games and settings nd22 is going to use and CPU clock is similar as well.

Does anyone have a motherboard with the NForce 3 150 chipset? I was able to find one, which wasn't easy, but it doesn't work (gigabyte).
In 2003, I heard from experienced overclockers that the best 754 board is... shuttle an50r. Back then, there was no nforce3-250 yet.

overdrive333 wrote on 2025-10-23, 14:31:

Does anyone have a motherboard with the NForce 3 150 chipset? I was able to find one, which wasn't easy, but it doesn't work (gigabyte).
In 2003, I heard from experienced overclockers that the best 754 board is... shuttle an50r. Back then, there was no nforce3-250 yet.

I don`t think so
https://hwbot.org/hardware/motherboards?key=s … rce_3_150-an50r
0 record
For comparison
https://hwbot.org/hardware/motherboards ... x)-ga_k8ne
450 records

Or DFI LANParty UT nF3 250Gb - 1400 record.
https://hwbot.org/hardware/motherboards?key=d … ty_ut_nf3_250gb

overdrive333 wrote on 2025-10-23, 14:31:

Does anyone have a motherboard with the NForce 3 150 chipset? I was able to find one, which wasn't easy, but it doesn't work (gigabyte).
In 2003, I heard from experienced overclockers that the best 754 board is... shuttle an50r. Back then, there was no nforce3-250 yet.

No, I do not have one, but I simulated a nforce3-150 by lowering the hypertransport from 800 MHz to 600 MHz on Abit NF8 PRO. There is a difference that grows as the LOD and resolution grows - that is why I consider that it is useful to test at multiple resolutions and level of details.

PART 1: AGP

All results will appear alphabetically with a different theme selected for easier viewing:
A. NVIDIA nforce3 with the Windows XP embedded theme
B. ULI m1689 with Windows XP zune theme
C. VIA K8T800 PRO with Windows XP luna theme

Just a suggestion, but since we're all the way on page 12 of the thread already, you might want to edit the first post in the thread so that it has links to any posts that have important bits of information, and to whichever posts have benchmark results. I think a lot of people may be curious about this, but most aren't going to make it 12 pages before there are benchmark results.

Archer57 wrote on 2025-10-06, 05:27:

But are they? From what i've seen with s939 single core CPUs they are not all that different from AthlonXP at the same frequency […]
Show full quote

nd22 wrote on 2025-10-06, 04:15:

...Because the processor available on 754 are much more powerful than those on socket 462...

But are they? From what i've seen with s939 single core CPUs they are not all that different from AthlonXP at the same frequency. Yes, there are improvements and they are faster, partly because of integrated memory controller, but not dramatically so. I'd expect <20% difference and that is with dual channel memory on s939.

S754 is also a downgrade from S462 in a sense that it only has single channel memory...

If anything it would be fun to see that 3400+/2.2Ghz Athlon64 compared to 3200+/2.2Ghz AthlonXP directly.

Dual channel memory is a questionable distinction on Athlon XP. Socket A was choked by the capabilities of the CPU bus (8 byte wide, DDR only, 200mhz max).

I find downclocking more interesting on retro hardware as it allows you to approximate more CPU performance regimes of older systems. As far as I'm aware, all S754 CPUs could be freely downclocked whereas only XP-M and perhaps some Sempron CPUs supported PowerNow on Socket A.

I’ve had an MSI K8T Neo board for a while now and it’s definitely a fun platform to use. My Athlon 64 system is made up of late 2003 components,mostly manufactured around late September/early October 2003 (board, CPU, RAM etc.), sometimes earlier (sound card), with the GPU having been made in week 42 2003 (so possibly sold around November ‘03). It’s based around the Athlon 64 3200+ 2.0 GHz, Radeon 9800XT, 1024MBs of Corsair XMS Pro memory (2x512MB, both week 40 2003) and a Sound Blaster Audigy 2 at its core.

It definitely has no real issues executing your average high-profile 2004 game with good performance. I’ve made a few casual captures of game performance on that PC already, with Far Cry being quite smooth in general (albeit not by the craving standards of today): https://www.youtube.com/watch?v=4YRt_y-qkPo

However, not every 2004 game will be a smooth and consistent ride. Vampire The Masquerade: Bloodlines is infamously quite unpolished, out of the box, and some scenarios run better than others but some also tank performance quite radically: https://www.youtube.com/watch?v=hWlJpD_9kzA

Dipshidian wrote on 2025-10-24, 22:05:

1024MBs of Corsair XMS Pro memory (2x512MB, both week 40 2003)

BH-5 ? I only found a Kingston CH-5. Now need to figure out how to supply 3.3V to it.

Dipshidian wrote on 2025-10-24, 22:05:

However, not every 2004 game will be a smooth and consistent ride. Vampire The Masquerade: Bloodlines is infamously quite unpolished, out of the box, and some scenarios run better than others but some also tank performance quite radically: https://www.youtube.com/watch?v=hWlJpD_9kzA

Yeah, I tried Bloodlines on a similar setup yesterday (albeit with a Radeon 9600 XT) and the lack of optimization really shows. The game had a troubled development, and its release was rushed by the publisher, which explains why it shipped in such a poor state.

Another infamously unoptimized game from around that time is Deus Ex: Invisible War. It too ran like crap on period correct PC hardware, despite having its gameworld chopped up into tiny areas, so they could fit inside the OG Xbox's limited memory pool.