Almost a year later, after ”upgrading ” from a Sandy bridge to a Core 2 quad and not missing anything, I come back to this topic to continue with the testing. A lot of things to do both at home and at work prevented me from posting further updates. Now finally I got free time to continue!

It is time for the conclusion for all the chipsets supporting 400 MHz FSB. Nvidia nForce2 ultra, VIA KT880 and KT600 fought each other and after taking into account all the scores here is the final standing:
Nforce2 ultra represents the baseline and is awarded a 100% score and the other 2 platforms are a percentage higher or lower.
First place: Nforce2 ultra 400: 100%
Second place: VIA KT600: 95.97%
Third place: VIA KT880: 94.80%

It may seem counterintuitive that KT600 is faster than KT880; that is due to the fact that VIA VT1616 audio codec present on Abit KV7 and is vastly superior to the Realtek ALC650 present on Abit KW7. This is most easy to see if we take into account only the 10 tests with sound:
Nforce2 ultra: 100%
VIA KT880: 93.35%
VIA KT600: 104.3%
What do you know? KT600 actually is faster than nforce2 – albeit within the 5% margin of error!
Dual channel does not help KT880 in any way; the scores are identical in both dual and single channel configurations. If we look only at the tests without sound so KT600 does not has the advantage of the superior codec the situation is as follows:
Nforce2 ultra: 100%
VIA KT880: 97.78%
VIA KT600: 93.63%
The difference between the 2 VIA chipsets has a lot to do with the disastrous result of KT600 in Tom 2D otherwise the difference between them would be around 3%.
Overall nforce2 ultra is the best socket 462 chipset! The only possible pretender to the throne would be a motherboard with VIA KT880 equipped with VIA VT1616 codec, however I don’t know of any such board!

The final chapter in the socket 462 saga will be the 266 MHz chipsets! They represent the first generation for the socket 462 and consist of 3 players: NVidia, AMD and VIA.
Representing NVidia is nforce 415-D, the very first chipset manufactured by NVidia. The motherboard is the Abit NV7 – 133Raid with the MCP – D Southbridge that has the Sound storm APU. When first appeared it was considered a truly breakthrough platform for the Athlon processor but reviewers were strongly disappointed when they actually got to test it. We shall see how it performs today where we got al the software and hardware quirks ironed out.

Representing AMD is AMD760, the second chipset for the Athlon processor from AMD. The motherboard is the Abit KG7 – raid with the VIA 686B south bridge and not the AMD766 – this is actually better than the one from AMD because it has more integrated functions such as hardware monitoring.

Representing VIA is the KT266A, the second chipset from VIA for socket 462 with DDR memory support. The original KT266 is not included here; it performs very poor and Abit did not manufactured a single board with it. The motherboard is Abit KR7A – 133Raid with the 8233A south bridge – more advanced than the 686B but still not as complete as the MCP-D from NVidia.
Also VIA has a second chipset present with a 266 MHz system bus: KT133A – this one is different form the rest because is supports SDRAM rather than DDRAM; while it is true that KT266A can support SDRAM the board used is for DDR only and I see no reason to use one with SDRAM today, maybe back in the day to ease the upgrade path you could buy a board with SDRAM/SDRAM+DDRAM support but today we are using only the best option for each chipset family. The motherboard used is Abit KT7A – raid which means that we have 2 variables now: one is the chipset and the other one is the memory.
Up until now I used the same CPU, GPU, RAM in order to isolate the one thing I wanted to test: the chipset represented each time by a different board. But with KT7A – raid not only we got a different type of memory, we also got a different amount: 1.5 GB. So this time a second variable is introduced, the PC – 133 SDRAM that already is a bottleneck: the Athlon XP uses a double data bus of 266 MHz that greatly benefits from using DDR memory so this platform should represent the lowest performing out of all 4. We will see if this is true but before we even start testing I shall talk about some obvious questions regarding the testing procedure.

Clarifications about testing procedure:
Question 1: first thing first: the”elephant” in the room – does 1.5 GB of RAM which is the max amount on KT133A is sufficient or not? Using the hard drive as virtual memory does have an impact in tests – average frame rate and particularly minimum one will drop. In all tests since I started this project virtual memory was disabled so everything was loaded in the system memory and I monitored RAM usage to see how high it would go. The highest amount is always during game testing; synthetic benchmarks do not require large amounts of RAM. Far cry 1 and Doom 3 are the biggest offenders: at settings that I used total system RAM utilization is between 1.3 – 1.4 GB so even KT133A with its 1.5 GB limit was sufficient to run all tests and Windows XP never complained about it – the message out of virtual memory never appeared, not even once. While there are later games that use much more RAM because I used games up to 2004 this is not the case in this comparative.
Now about the limited bandwidth provided by SDRAM: we will see the results first, but how much does the KT133A chipset and PC – 133 impact performance will be a matter for debate after I put up the results.

Question 2: how about ali magik 1? This chipset was implemented in very few boards and furthermore there are 3 revisions of it: A0, B0 and C1. There is no way to differentiate them unless you are removing the chipset cooler to read the markings. Abit made no boards with it and it is not included in the tests.

Question 3: where is SIS 735? There is only one manufacturer with only one mass produced model: ECS K7S5A. There may be more models such as Chaintech 7SID, MSI K735 Pro but I am not sure how many of them are out there as it seems SIS was not loved by the Taiwanese boards partners at all. In any case Abit did not manufactured a single model with SIS 735 so it is not included in the tests however I should point out that period correct reviews were highly positive. Sis 735 was considered the best 266mhz chipset for the Athlon XP! I should also point out that most if not all implementations of the SIS 735 are of low quality and cheap from questionable manufacturers. Phil did make a video about ECS K7S5A Pro and it was always behind boards with KT266A from Asus so maybe the reviews back in the day were somehow enthusiastic.

Question 4
Sound: up until now I used integrated sound. So if the board had a Realtek codec this was in all cases a weak point; if it had sound storm APU as was the case with Abit AN7 this greatly helped the scores in all tests with sound. VIA VT1616 audio codec was a huge surprise: it was right up there with sound storm and it provided a huge boost to Abit KV7 scores.
However of the 4 boards to be tested only Abit NV7 – 133raid has integrated audio: it is the famous sound storm so already nforce1 starts with one advantage; the rest of the boards – KT7A – raid; KR7A – 133raid and KG7 – raid do not have any kind of integrated audio.
In order to provide audio, level the playing field and compare the original nvidia APU with the competition I have decided to use Creative live 5.1 revision 0100 in all systems, including the one with NV7 – 133R. This means there will be 2 sets of result for nforce1, one with sound storm, one with Creative live. Let’s see if sound storm can defeat a 3 years old Live.

Question 5
Video: up until now I used geforce 7800gs in all the platforms I have tested. It worked flawlessly and I was sure there was no GPU bottleneck.
However in NV7 – 133R I can no longer use this card because there are several capacitors including a big one right next to second IDE port that prevent the card from being inserted in the AGP slot; this is also true in the case of KR7A – 133R – lots of capacitors prevent the card from being inserted.
In the case of KG7 – raid there is another problem where the backplate of 7800gs hits the RAM retention clips and can not be inserted in the AGP slot; this problem is also present with the KT7A – raid!
So because I could not use the geforce 7800gs I went for the next best thing: Geforce 7600 GT! Performance wise at lower resolution like 1024*768 this card is a bit faster than 7800gs however at higher resolution such as 1600*1200 performance is a bit lower than 7800gs. Overall 7600 GT makes sure there is no bottleneck from the video card in the system.

Before presenting the systems used I will mention the common parts and software used in all the tests:
CPU: Athlon XP 2000 Thoroughbred revision B part number AXDA2000DUT3C
RAM: 2*1gb DDRAM various brands all set at the same timings in the BIOS; 3*512mb Tanysis SDRAM.
GPU: Leadtek geforce 7600 GT that works just fine in all the systems!
PSU: Enermax various models that work just fine 20 years later!
CPU cooler: some copper ones that kept the temperatures in check
Storage: hard drives were used all the way, no problems encountered; some random dvd – rw that I got was used for optical media.
Software: Windows XP SP3 with no updates; nforce version 5.10, VIA platform driver 4.56v, nvidia forceware 93.71, default creative live driver installed by XP; fraps 3.4.7, MSI afterburner, irfan view; Everest version 5.50.

Let’s go now over the systems specs one by one. I will start with the star of the show: NV7 – 133R based on nvidia nforce 415 – D, named nforce1 from now on. This is the fully fledged version of the chipset with all the bells and whistles: sound storm, dual channel DDR, but lacks the integrated geforce2 mx which is actually a blessing, I am not interested in testing that anyway.

At this moment most of you will jump right away: but nforce1 can only support a maximum of 1.5gb of RAM! Wikipedia says so!
Well, this is not actually true. As I said, 75% of info on the web is right the rest – I found new limits for all the boards. With the BIOS version C3 the board will take 1gb dimms and you can safely use 2gb of DDRAM in dual channel. If the latest version DW is flashed you can no longer use 1gb dimms! You must use that particular version in order to be able to use 1gb modules.
Also despite official support for Athlon XP 266 with 266Mhz FSB the board will NOT take any CPU that has a multiplier over 12.5X – all you get is a black screen! And NO, the final bios version will NOT let you use Athlon XP 2600mhz with a 16X multi! So Athlon XP 2000 is the common CPU for all systems.

Pictures of the motherboard and components used – you can see the large capacitor that is blocking any long AGP cards from being used:

Pictures with the system info after Windows XP and all drivers have been installed – 2gb of RAM are recognized just fine:

The one of the very few nforce1 boards ever made runs just fine with 2gb of RAM; the scores with the C3 BIOS version that supports 2gb and the DW BIOS version that supports only 1gb in dual channel are the same! There is no performance loss by going with an older BIOS revision only the advantage of 2gb of RAM.
It seems that picture compression is awful on Vogons 😀!

The second system is the one with the AMD 760 chipset: Abit KG7 – raid, considered on the best boards with this chipset and the only one I know of with 4 memory slots!

Unlike the NV7 - 133R this board does NOT have the auxiliary power connector so a PSU with a strong 5V rail is a must!

All the components before final assembly: