VOGONS


Socket A: Nvidia vs Via - battle of the platforms!

Topic actions

First post, by nd22

User metadata
Rank Member
Rank
Member

Today I am going to start a new project to see which is the best socket socket A platform! Nvidia and Via will go head to head to determine one more time which is the best for your socket 462 processor! I am fully aware that back in the day nforce was considered the best chipset and a nforce based motherboard the best choice for socket A however using state of the art hardware and up to date software I will try to see if the new results will result in a new winner or a tie!

Without further ado let's begin:

Stage 1:system specs
Common hardware:
1.CPU: Athlon XP 3200 2.2ghz with 400 FSB (actually it's 200)
2.RAM: Corsair 2*1gb CAS 3-3-3-8 DDR400 with silver heat-spreaders!
3.GPU: Leadtek geforce 7800gs 256mb. To all the people that will remind me that are stronger video cards on the AGP interface: I tried installing both the Radeon 4670 and the Radeon 3850! The system enters a continuous restart loop with either of the cards installed but runs just fine with geforce 7800gs/7600gt/X1600XT!
4. Storage: Maxtor 80gb HDD; Intel 180gb SSD (yes ,I know it's not period correct but I used it to speed up the boot times which are really slow on the Maxtor) various IDE optical drives; the system will not run with a SATA optical drive installed!
5. PSU: 350w FSP with strong 5V rail; old but good!
The motherboards:
In one corner we have the Abit AN7 with nforce2 ultra 400 chipset; this motherboard does not have SATA integrated in the chipset but provided via a Silicon Image 3112 controller; 3 memory dims; after serious problems and numerous testing I have found a kit capable of running in dual channel configuration on this board and this is what I am going to use! Also the board has Sound Storm that provides one the best audio experience on systems of this era.
In another corner we have the Abit (you see a pattern here don't you?) KW7 with via kt880 chipset; SATA is integrated in the chipset; 4 memory dims and is far less picky about the memory running with just about every memory module I have; can run with 4 1gb dims in dual channel mode however I choose to use only 2*1gb because any 32 bit Windows (XP or 2000 - tried both) will recognize only 3gb and you must set 2T command rate in Bios for the system to be stable wherever with 2 dims the system runs at 1T!
Both systems have been configured to run only with minimum required hardware in order to complete the tests; wireless cards, extra USB cards have not been installed; unnecessary Bios options have been disabled!

Stage 2: software
1. operating system: Windows XP SP3 fully updated with POSReady2009 patch and all extra updates installed; I pondered about installing windows 2000 or 98 but I decided that XP is period correct and will also provide the best drivers and all the enhancements for the 2 systems. Latest version of DirectX9.0c is also installed.
2. benchmarks: 3dmark 2000/2001/2003; Aquamark3; driver version 175.19 has been used!
3. any other suggestions are welcomed!

Reply 1 of 60, by phosgene

User metadata
Rank Newbie
Rank
Newbie

The nForce2 always had the elite gamer cool factor for me when I was younger, so my vote is for that. In fact I almost want to see if I can find a DFI LAN Party NFII Ultra motherboard, and do a black and green themed case mod with a Cheiftec Dragon, to relive the glory days.

Reply 2 of 60, by appiah4

User metadata
Rank l33t++
Rank
l33t++

I was always a VIA guy and KT400A/KT600 were my favorite Socket A chipsets..

Retronautics: A digital gallery of my retro computers, hardware and projects.

Reply 3 of 60, by dionb

User metadata
Rank l33t
Rank
l33t

nVidia vs Via is all very well, but what about SiS?

SiS' 735 chipset was very popular in mid-SoA times, both because it outperformed the contemporary AMD760 and Via KT266A chipsets, and because of the excellent and ludicrously cheap ECS K7S5A motherboard (pretty much the only SiS735 based board you'd see).

They also had a 400MT/s FSB chipset, the SiS748. Its specs were basically the same as the KT600. Reviews back in the day were a very mixed bag, indicating immature BIOS. In terms of memory performance they are competitive with the KT600 but no match for the dual-channel chipsets, however their I/O performance was generally regarded as being best-in-class. I for one would be interested to see how they would perform in this comparison, particularly in more I/O heavy loads in combination with that SSD. Do you by any chance have one? 😉

Reply 4 of 60, by nd22

User metadata
Rank Member
Rank
Member

I can buy an Abit (you see the pattern, don't you? 😊 )KV7 motherboard for a very cheap price (around 10 euro) and redone the test with the Via KT600 chipset but i think that KT880 will generally score better despite the difference being only the dual channel memory (as far As I remember but I have to make a search for that!) SIS is regarded as being inferior in performance and I did not searched for a SIS based motherboard! However I can search for one if you think it will make a difference and perform on a similar level with nforce2 and kt880!

Reply 5 of 60, by shamino

User metadata
Rank Oldbie
Rank
Oldbie

Although it's probably not the focus of this comparison, the KT880 should have an appealing I/O advantage at least, since it has SATA and GbE integrated in the southbridge, which connects to the northbridge on a 533MB/sec link.
NForce2 doesn't have SATA or GbE, so those both have to use PCI, quite likely at the same time.
I remember testing an ABit AN7 in the Intel NAS benchmark some years ago. It's performance wasn't very good in that role due to the PCI issue.
In modern desktop usage, I find myself using SATA drives and transferring large files over the LAN much more than I used to, so at least for me this becomes somewhat relevant even for a desktop.

However, for desktop gaming purposes, the nForce2 is a great chipset. My AN7 was one of my favorite desktop systems I've ever had.
I never had a KT880, but I have to imagine VIA must have caught up by then.

I recall there was a funny quirk with the optimal RAM timings on the nForce2, where setting one of the latency figures to a higher (slower) value would result in faster performance. I had 2-3-2-6 rated RAM, but I think that last figure needed to be set to 11 for best throughput.

Reply 6 of 60, by dionb

User metadata
Rank l33t
Rank
l33t
nd22 wrote:

I can buy an Abit (you see the pattern, don't you? 😊 )KV7 motherboard for a very cheap price (around 10 euro) and redone the test with the Via KT600 chipset but i think that KT880 will generally score better despite the difference being only the dual channel memory (as far As I remember but I have to make a search for that!) SIS is regarded as being inferior in performance and I did not searched for a SIS based motherboard! However I can search for one if you think it will make a difference and perform on a similar level with nforce2 and kt880!

SiS had very bad rep due to their UMA integrated graphics chipsets, which were invariably used in the lowest of the low end So7 and So478 systems (and to a lesser degree with So370 and SoA). UMA crippled performance, and cheap crap boards led to a reputation for instability which had nothing to do with the chipsets but stuck to them nonetheless.

In fact their non-integrated chipsets were frequently significantly better performing than Intel and AMD alternatives. In 486 days, SiS chipsets were considered high-end over Intel and Via. The old SiS 501 was the first non-Intel chipset for the original P5 Pentium and unlike the i430LX it did support things like EDO RAM and PLB cache. The 5582 was the best performing and most flexible of the sub-100MHz So7 chipsets. And the SiS735 was when it was released the undisputed SoA champion, beating the Ali Magick 1 (not that that was too hard), Via KT266 and AMD760. The later KT266A more or less drew equal (KT266A had better AGP performance, the 735 better RAM and I/O), and only once the KT333 and nForce1 appeared did it lose its crown.

As for the 748 - it depends on the benchmarks. No way is it in the same league as the nForce2 Ultra in terms of RAM bandwidth (2 channels beat 1), but I would expect it to probably beat the nForce2 and KT880 on I/O - and with an SSD involved (something obviously not benchmarked 14 years ago) that might well give it a significant edge. In Winstone benchmarks and various video-editing benchmarks the 748 managed to beat all other SoA chipsets, although it was never really competitive in gaming. The DFI 748-AL would be the board to go for, although I suspect it would be easier to find an Asrock A7S8XE+ (if you enjoy recapping that is).

Reply 7 of 60, by nd22

User metadata
Rank Member
Rank
Member

I have read a review about that DFI board: http://www.ocworkbench.com/2003/dfi/748al/748al-1.htm and it seems that in games which will be one the main points of this comparison there is simply no match against nforce2 and KT880 which are both dual channel platforms! Only KT600 seems to be within reach!

Stage 2: code creatures pro has been added to the benchmarks! I can also use some games and I am thinking wherever I should use directx9 games or directx8 or both! If i go with directx9.0 my picks would be: Far cry 1, Half life 2, Doom 3 but there is a strong chance that these games will be GPU limited even with a geforce 7800gs. I am thinking of adding Pcmark 2002/2004 to the benchmark suite; what do you think?

Stage 3 - Settings: in order to best represent the era in which socket A systems were common I have a 19 inch LCD monitor with 5:4 aspect ratio with a maximum resolution of 1280*1024; I think that 1024*768 and 1280*1024 are representative for the retro comparison that I will try to do! If there are suggestions to go higher i have newer monitors that can go up to 1920*1200 but anything over 1280*1024 will probably be GPU limited.

Reply 8 of 60, by shamino

User metadata
Rank Oldbie
Rank
Oldbie

Screen resolution is a variable that will depend on the needs of the reader. 1280x1024 does represent the mainstream of the Socket-A days, but 1600x1200 was also a persistent standard for higher end 4:3 monitors for many years. People building systems today are likely to have those monitors since they aren't expensive to get anymore.
If the person reading the results today wants to run on a 1920x1200 monitor (or a 1080p) then they may be interested in those results also.
I guess it comes down to how many variables you want to have to deal with in your benchmarks. As you said I think the changes in resolution will show up more in video card performance than anything else.

I'd be interested in seeing some benchmark for how these systems can handle rendering of modern script-heavy web sites, but I don't know of any controlled, specific test for that. This is a weakness in the practical usability of old systems nowadays, so it could be interesting to see if there's a setup that sucks less in this area.
I don't know how many people would care about this, but I like to get everyday usability out of older systems to whatever extent is possible.
I also have some interest in video playback tests for the same reason. But I don't know how much you want to get into non-gaming tests. Too many tests and variables can get burdensome.

Reply 9 of 60, by squiggly

User metadata
Rank Member
Rank
Member

- A 7800gs is ridiculously powerful to match with an Athlon XP. I run a Ti4400 and it handles games up to around 2003 easily.
- SoundStorm is "just OK". It is mainly a Realtek codec combined with Nvidias "dolby digital out" (Nvidia APU). It only supports EAX2.0, making it equivalent to SB Live! released in 1998. I run an Audigy2ZS on my "Athlon XP generation" PC and it handles games from this generation well (I have another PC from this generation, a P4 with a Vortex2 to play A3D2.0 games). You might want to consider getting a Vortex2 yourself as a bunch of games from this generation need A3D2.0 to get the best surround sound experience.
- make sure you set RAM to be synchronous to the FSB on a nForce mobo, i.e. 200mhz FSB to get 400mhz DDR. Virtually any Athlon XP will work with a 200mhz FSB if you set the clock multiplier to keep the internal clock at a reasonable level.

Reply 10 of 60, by swaaye

User metadata
Rank Moderator
Rank
Moderator

Let me suggest you run the Windows GUI benchmark Tom2D. Typically graphics cards have much lower GUI performance on VIA motherboards compared to nForce and Intel chipsets. I find this is best noticed in practice with web browser drawing performance and with minimizing/maximizing windows.

I have it up on Mediafire because it's hard to find otherwise. It originated from Tom's Hardware. They made it to demonstrate how 2010 GPUs had rather poor GUI performance.
http://www.mediafire.com/file/op7h65rvubj3atn … ms-2D-Bench.zip

Reply 11 of 60, by nd22

User metadata
Rank Member
Rank
Member
squiggly wrote:

- A 7800gs is ridiculously powerful to match with an Athlon XP. I run a Ti4400 and it handles games up to around 2003 easily.
- SoundStorm is "just OK". It is mainly a Realtek codec combined with Nvidias "dolby digital out" (Nvidia APU). It only supports EAX2.0, making it equivalent to SB Live! released in 1998. I run an Audigy2ZS on my "Athlon XP generation" PC and it handles games from this generation well (I have another PC from this generation, a P4 with a Vortex2 to play A3D2.0 games). You might want to consider getting a Vortex2 yourself as a bunch of games from this generation need A3D2.0 to get the best surround sound experience.
- make sure you set RAM to be synchronous to the FSB on a nForce mobo, i.e. 200mhz FSB to get 400mhz DDR. Virtually any Athlon XP will work with a 200mhz FSB if you set the clock multiplier to keep the internal clock at a reasonable level.

1. 7800gs is absolutely required for games like Doom 3, Far cry 1, Half life 2, NFS underground! Anything older or any directx8.0 game will probably run with medium settings OK with a 4400. I really mean it because testing with maximum in game settings including AA & AF with a 6600gt the results were very poor!
2. RAM:FSB is set to 1:1 on both boards! I will post screenshots and pictures with both systems!

Reply 12 of 60, by Radical Vision

User metadata
Rank Member
Rank
Member

I prefer every day Radeon AGP cards bcuz they are better then the nVIDIA ones.
The DFI LanParty is better then ABIT NF7 V2 series only out of the box. But when you flash modded BIOS on the NF7 that will be the better board, someone will say why, well many many reasons.. Yeah the LanParty Ultra B looks way better and UV slots, but still the stability and the quality of components is way lower compared to the ABIT...
You can see on HWbot the best AthlonXP results, you will see mostly ABIT NF7 V2 boards, and not much LanParty, not to mention for KT880 not much of them as well..
Also i don`t see how the KT880 is the better chipset, as KT600 did lack AGP/PCI lock, (so i think the KT880 lack the AGP lock as well not 100% sure). Also the stability of the nForce 2 Ultra 400 is superior compared to KT880...
There is other thing i did test 2GB DDR500 memory from socket 939 era, and other guy did test other good 939 kit, the thing is if you want 2GB on faster speed then 400, both KT880 and NF2 U400 are useless as they cant maintain that speed...

SiS well there was only crap boards mostly low end with that chipset brand, so not worth it at all...

zj3y5r-6.png
ptdfb9-3.png
s3yhar-2.png
9s6qlr-3.png
rd0y5f.png

Reply 13 of 60, by squiggly

User metadata
Rank Member
Rank
Member
nd22 wrote:

1. 7800gs is absolutely required for games like Doom 3, Far cry 1, Half life 2, NFS underground! Anything older or any directx8.0 game will probably run with medium settings OK with a 4400. I really mean it because testing with maximum in game settings including AA & AF with a 6600gt the results were very poor!

OK fair nuff. I play those games on my 4th generation PC (Core2Duo/Radeon HD5870) and it destroys them. I consider the Athlon XP to be generation 3, and I tend to play games from around 1999-2003 on it (Half Life 1, NOLF, Deus Ex, BF1942, MoH, etc).

The reason I said it was the XP3200 was released in 2003, and the 7600 was released in 2006.

Personally I also run Win98 on my gen3 box as it was the dominant gaming platform until around 2003. By 2004 most people had switched to XP and games were starting to require it.

Reply 14 of 60, by Radical Vision

User metadata
Rank Member
Rank
Member

And so what if one CPU is released 3-4 years before a graphics card, is way better then it will be...
As the cards from the time of the processor will struggle with every single game to run it above mid quality...
For example in 2003 was the Radeon 9800 or FX5950 Ultra, well the cards are solid, BUT they will never be able to run most games that Radeon 3850 or Radeon 4670 will be able to run on higher settings..

zj3y5r-6.png
ptdfb9-3.png
s3yhar-2.png
9s6qlr-3.png
rd0y5f.png

Reply 15 of 60, by nd22

User metadata
Rank Member
Rank
Member

I did not post for a couple of days because i hit a very curious problem with both systems: it seems that they can only be powered on using a PS/2 keyboard!
I was using a normal USB keyboard and after the first 2 days of testing both systems refused to start when i pressed the power button. After fully dissembling both systems I did some testing on the open bench and after the first power on using a power switch both refused to start! I was puzzled and was thinking maybe the power supplies went both bad! Changed to new power supply on nforce2 system and it turned on using the power switch first time but refused afterwards. Tried numerous combinations of power supplies, memory kits and video cards disconnecting from the main power every single time but the problem persisted – first time after switching on the PSU the system started with the press of the power button afterwards it refused to start – until I changed the keyboard to an old PS/2 one and set the system to start using the ‘press any key” option in BIOS and setting the jumper to the corresponding position on the motherboard. This is happening with 2 different cases.
Now both systems – nforce2 based one and the kt880 based one – will start only with the help of a PS/2 keyboard. I must mention that the power button is good because in windows I set the systems to shutdown when I press the power button and both do and I also tested with another motherboard/processor combo and it starts normally!
Any ideas about solving the above problem; has anyone encountered such a problem before?

Reply 16 of 60, by nd22

User metadata
Rank Member
Rank
Member

I have been busy at work but also at home bench-marking and testing games and after 2 weeks I can say that you are in for a big surprise!
I will start with VIA KT880 platform; the system is exceptionally stable and bug-free; the results are not influenced by the memory modules used - but during the course of the testing I used the Corsair; they are the same with Corsair, Kingston and AENEON modules so because the Corsair modules are used in the nforce2 system the KT880 is equipped with whatever I had at hand after the benchmarks have been completed!
1. screenshots with the system:

Attachments

Last edited by nd22 on 2018-03-04, 14:58. Edited 3 times in total.

Reply 17 of 60, by nd22

User metadata
Rank Member
Rank
Member

2. Picture with the system up and running (I can not post more photos because they are over 2mb in size!!):

Attachments

  • 20180304_160150.jpg
    Filename
    20180304_160150.jpg
    File size
    1.66 MiB
    Views
    1188 views
    File license
    Fair use/fair dealing exception

Reply 18 of 60, by nd22

User metadata
Rank Member
Rank
Member

4. Benchmarks results for Code creatures pro; 3d mark 2000/2001/2003/2005; all tests have been done numerous times and every single time the benchmark have been set to repeat 3 times in order to get an accurate result; first screenshot is with the default settings except for the "repeat 3 times" option; the next are done at 1280*1024/1680*1050 resolutions with all the settings turned all the way up:
A. Code creatures Pro:

Attachments

Last edited by nd22 on 2018-03-04, 14:44. Edited 1 time in total.

Reply 19 of 60, by nd22

User metadata
Rank Member
Rank
Member

B. 3d mark 2000 - you can see that the results are clearly CPU limited even at 1680*1050:

Attachments