I'm currently doing a lot of benchmarking for a Voodoo 2 project. Currently using a slot 1 system, but once I'm done with that will move to the Gigabyte board with the VIA chipset. I will include 1 CPU (PIII 750) that I also used on the BX440 board and will see if there are any differences.

But I don't have any long-term experiences.

You can see (and hear) that motherboard in action here: https://www.youtube.com/watch?v=lozF7Nx97j8

With the Intel 815 chipset. Is it true that not all of them are Tualatin compatible?

I read something about a B version or stepping?

Memory performance on VIA Apollo Pro is less then one would expect. I suppose it needs 133MHz FSB just to be near a BX on 100MHz FSB.

VIA C3 Ezra-T CPU at 133x9.0=1200MHz on i440BX mainboard, write combining enabled, RAM at CAS 2:
PC Player benchmark 640x400: 50,5 FPS
Quake I Timedemo 2 640x480: 38,1 FPS

VIA C3 Ezra-T CPU at 133x9.0=1200MHz on VIA Apollo Pro 133A mainboard, write combining enabled, RAM at CAS 2:
PC Player benchmark 640x400: 36,0 FPS
Quake I Timedemo 2 640x480: 24,8 FPS

I don't think Tualatin socket compatibility is derived from the chipset itself, but from the mainboard design.
I like the Upgradeware Slot-T Tualatin adapter for Slot 1. There is also the Lin-Lin adapter for Socket 370, I bought a second one of those for $1 recently. Or the asian ebay guy that sells modded Tualatins. Powerleap Slot 1 adapters are usually expensive/more rare.

NJRoadfan wrote:

blakespot wrote:

Mau1wurf1977 wrote:

VIA chipset boards would give you all these options. 133 FSB, Tualatin support and AGP.

Work w/ P3 866 right? Any negs?

Yeah, its a VIA chip set.

(now I'm going to have nightmares about 4-in-1 drivers)

Hmm. So better the Intel 440BX than the VIA, despite AGP being over clocked when 440BX @ 133 sys bus? Seems I've heard a few no-VIA rec's lately.

bp

gerwin wrote:

Memory performance on VIA Apollo Pro is less then one would expect. I suppose it needs 133MHz FSB just to be near a BX on 100MHz FSB.

Many older VIA boards had memory interleaving disabled by default. On some of these boards, the only way to enable it is with a software patch.

Most newer (Tualatin-era) boards automatically enable memory interleaving, which greatly increases memory performance. I get similar memory performance to the i815 on my TUV4X, clock-for-clock. My personal PIII 3DMark01 record was set on this board (12,210 with an overclocked 9800 Pro), and that benchmark is quite sensitive to FSB/memory performance.

Unfortunately AGP is still a problem even on Tualatin-ready 694/Apollo 133 boards. This board refuses to run most cards at AGP 4X (although on a PIII, AGP 2X usually isn't a bottleneck), and NVIDIA cards newer than the GF4 TI series are also troublesome. However, any ATI card at AGP 2X is rock stable.

Might be easier going for an early Pentium 4 platform instead. They all have intel chipsets, AGP and should be rock solid.

I had a look and even for S478 you can get slow CPUs starting at 1.4 GHz.

Not sure why this option never gets mentioned. It's always the Tualatin that gets the praise 😀

I have seen quite a few 815 chipset boards, but I believe most aren't Tualatin compatible.

ISA sound is something I wouldn't worry about with such fast CPUs. PCI DOS support is quite good with late games and I haven't had any trouble with Vortex 2 sound card so far (Turtle Beach).

Standard Def Steve wrote:

Many older VIA boards had memory interleaving disabled by default. On some of these boards, the only way to enable it is with a software patch.

VIA C3 Ezra-T CPU at 133x9.0=1200MHz on VIA Apollo Pro 133A mainboard, write combining enabled, RAM at CAS 2,
Now with Voodoo 3 AGP like on the BX, instead of the Geforce 6200 it had before:
PC Player benchmark 640x400: 32,8 FPS
Quake I Timedemo 2 640x480: 22,6 FPS

This is now with 2 sticks of 256MB, considering 4-way interleaving requires two identical sticks.
It seems the BIOS already sets 4-way interleaving for the occupied banks (value D5 becomes D6), but performance does not really decrease when I turn it off again? Setting other memory options through registers 50 and 51 to value FF gives a tiny FPS gain. I give up...

I also remember that DOS high resolution performance was very poor on this VIA chipset. I made a thread about this a while ago, but can't find it.

http://www.zen26266.zen.co.uk/Chipset.htm Some old reviews about the via chipsets.

gerwin wrote:

VIA C3 Ezra-T CPU at 133x9.0=1200MHz on VIA Apollo Pro 133A mainboard, write combining enabled, RAM at CAS 2, Now with Voodoo 3 […]
Show full quote

VIA C3 Ezra-T CPU at 133x9.0=1200MHz on VIA Apollo Pro 133A mainboard, write combining enabled, RAM at CAS 2,
Now with Voodoo 3 AGP like on the BX, instead of the Geforce 6200 it had before:
PC Player benchmark 640x400: 32,8 FPS
Quake I Timedemo 2 640x480: 22,6 FPS
This is now with 2 sticks of 256MB, considering 4-way interleaving requires two identical sticks.
It seems the BIOS already sets 4-way interleaving for the occupied banks (value D5 becomes D6), but performance does not really decrease when I turn it off again? Setting other memory options through registers 50 and 51 to value FF gives a tiny FPS gain. I give up...

This is what i got with mine,
Pentium 3 at 100x8.5=850MHz on VIA Apollo PLE133 mainboard, LFB write combining enabled, RAM at CAS 3, with GT520 PCI:
PC Player benchmark 640x400x8 : 90.2 FPS
PC Player benchmark 640x400x16: 63.8 FPS
RAM=512+256MB PC133 in asynchronous mode, 4-way interleaving enabled.
Regs 50 & 51=FF, 64 & 65=E6, 68=45 and onboard VGA disabled

Yeah, with a PIII the VIA Apollo scores are better.
I put the Geforce 6200 back in and replaced the VIA Ezra-T CPU with a Pentium III-S Tualatin. And set it at 8.5x100=850MHz:
PC Player benchmark 640x400: 92,8 FPS
Quake I Timedemo 2 640x480: 70,8 FPS

The VIA Ezra-T CPU was used before, because I happened to have it benched at the same speeds on both boards. For normal use I would not recommend an Ezra-T or older. A VIA Nehemiah is much better already.

Unfortunately we don't have a proper comparison with a i440BX score now. I do have these scores:

Soyo 440BX; Pentium III-S 9.5x105=1000MHz; Abit Siluro Geforce 4 MX440, write combining enabled, CAS 3:
PC Player benchmark 640x400: 135,0 FPS

Gigabyte 440BX; Pentium III Tualeron 4.0x100=400MHz; Voodoo 3 AGP, write combining enabled, CAS 2:
Quake I Timedemo 2 640x480: 50,9 FPS

gerwin wrote:

Unfortunately we don't have a proper comparison with a i440BX score now. I do have these scores: […]
Show full quote

Unfortunately we don't have a proper comparison with a i440BX score now. I do have these scores:

Soyo 440BX; Pentium-III-S 9.5x105=1000MHz; Abit Siluro Geforce 4 MX440, write combining enabled, CAS 3:
PC Player benchmark 640x400: 135,0 FPS

Gigabyte 440BX; Pentium III Tualeron 4.0x100=400MHz; Voodoo 3 AGP, write combining enabled, CAS 2:
Quake I Timedemo 2 640x480: 50,9 FPS

Sorry, the maximum i could bring it closer to your Soyo system is by overclocing mine to 8.5x112=950 MHz, but the remaining parameters will not match anyway. So no need bothering with that. I guess, we have to wait it out for Mau1wurf1977's bench results then. However i would be surprised if the 440BX doesn't beat the VIA Apollo Pro 133A chipset soundly, judging by your scores and other reviews. Interested only in knowing how much?
Presently i'm booting to WinME's DOS boot disk and from there loading the Phil's benchmark suite from a USB thumb drive as I don't have space in my IDE HDD. I'm able to run the 3dBench2 and PCPBench, but not Doom or Quake1.

I put something together, the scores are in:
Gigabyte 440BX; Pentium III-S 8.5x100=850MHz; Abit Siluro Geforce 4 MX440, write combining enabled, CAS 2:
PC Player benchmark 640x400: 120,1 FPS
Quake I Timedemo 2 640x480: 93,3 FPS

gerwin wrote:

I put something together, the scores are in: Gigabyte 440BX; Pentium III-S 8.5x100=850MHz; Abit Siluro Geforce 4 MX440, write co […]
Show full quote

I put something together, the scores are in:
Gigabyte 440BX; Pentium III-S 8.5x100=850MHz; Abit Siluro Geforce 4 MX440, write combining enabled, CAS 2:
PC Player benchmark 640x400: 120,1 FPS
Quake I Timedemo 2 640x480: 93,3 FPS

MSI 6368 PLE 133; Pentium 3 at 100x8.5=850MHz, Zotac GT520 PCI, LFB write combining enabled, RAM at CAS 2:
PC Player benchmark 640x400x8 : 90.3 FPS(no change between CAS2 and CAS3 as well as between PC100 & PC133)
Quake I Timedemo 2 640x480: 25.1 FPS(This low score might be due to GT520)

That's a whopping 25% difference 😢 in PCPBench. Unless the performance improvement due to microcode updates (Tualatin vs CuMine) counts and presuming the Tualatin is not a 512KB L2 Cache version. Initially I thought that maybe the GT520 PCI is limiting it to 90(however ridiculous it might sound). But then I retrieved its PCPBench scores on a Core2Duo(157), so that shouldn't be it.

gerwin wrote:

This is now with 2 sticks of 256MB, considering 4-way interleaving requires two identical sticks.

Correct me if I'm wrong, I thought that interleaving pertains to multiple banks(2 or 4) inside a memory module. That's why the option is there to enable 2 or 4 way interleaving individually in registers 64, 65, 66 & 67, each one for a particular DIMM slot. So interleaving can be enabled even if there is only one stick of RAM. In fact I got slightly higher memory scores with only the 512MB Kingston stick in comparison to 02 sticks (not identical however).

When I wrote Pentium III-S, with the S, it concerns a 512kB L2 cache model. But I benched the S model on the VIA motherboard too, so that is fair.

I calculate 23,5% FPS difference on average for both DOS benchmarks, PCPbench and Quake1. Maybe VIA drivers do some more tuning in windows. Don't know.

The highest multiplier I can set any of my coppermine core P IIIs is 8.0x. So Tualatins were used instead. Earlier I found that intel L2 speed/size does not matter that much for VESA benchmarking, you can see that here. The Tualatin has some edge over coppermine though. Maybe I should add a Tualeron 256kB to that post for comparison.

It said here that 1 DIMM supports 2-way, and two DIMMs support 4-way interleaving. I just assumed that to be right, even though one would imagine otherwise. My BIOS seems to set the bank to 4-way even with just one DIMM installed?

gerwin wrote:

I calculate 23,5% FPS difference on average for both DOS benchmarks, PCPbench and Quake1. Maybe VIA drivers do some more tuning in windows. Don't know.

I wonder what is the reason for the low DOS performance(not sure about Windows) as all the register tweaks have been exhausted. Poor memory controller, CPU interface, Buffer implementations or overall chipset design...? Another thing i noticed is, enabling/disabling memory interleaving in the primary DIMM slot results only in a few FPS difference in Quake1 timedemo 2 in 320x200 (84 vs 82 FPS), no change when doing the same in secondary DIMM slot. Ditto for PCPBench(90.3 vs 89.9)

gerwin wrote:

The Tualatin has some edge over coppermine though. Maybe I should add a Tualeron 256kB to that post for comparison.

As you have used the same Tualatin in both the chipset motherboards, it should not matter like you said.

gerwin wrote:

It said here that 1 DIMM supports 2-way, and two DIMMs support 4-way interleaving. I just assumed that to be right, even though one would imagine otherwise. My BIOS seems to set the bank to 4-way even with just one DIMM installed?

What i read about interleaving was from here. Wonder which one is correct!!

I'd like to give my opinion on the 440BX and 133MHz FSB.
I liked to use a P3 1GHz (133MHz) with a Gigabyte BX2000 I think. I must have been using a GeForce 2 Ti at the time and I always got some random blue screens or system hangs. I didn't give it much thought until later, when I used a 100MHZ FSB CPU and saw all of those go away.

Granted, AGP Voodoo cards can have a high tolerance to bus overclock, I wouldn't like to get them any hotter than they already get 😉. It's like swaaye said, it really depends on the motherboard quality and graphics card. Otherwise, it's kind of a gamble, in which case I'd rather go with an Athlon machine 😀. I generally avoid Athlons and Via chipsets because they're not nearly as trouble free as 440BX or i815 😀

I recently got a few NOS Gigabyte GA-6BX7+ boards. These boards have the Intel BX chipset, AGP, 6x PCI and 1 ISA slot. I already tested 4 of them just with these settings:
1000MHz coppermine CPU running at 133MHz FSB, PCI divider set to 33MHz, AGP is overclocked of course. 400 watts levicom PSU. 2x256MB SDRAM (MDT brand, PC133 CL2). I put a simple GeForce FX5200 in the AGP slot and all of these 4 tested boards ran stable for hours. I remote tested them with 3DMark2001 (loop) and also had a nice Q3A session (~4hrs), no problem at all. I also put in a GeForce 2 Ti card (from Aopen I think) and again - no stability issues whatsoever.

But I agree that it depends on many variables. Maybe also the power consumption of the card has to be put into consideration. The same applies to the SDRAM you use (I used MDT brand RAM, PC133, CL2 set to AUTO in the BIOS) and other parameters you set in the BIOS.
I keep two of the boards in my collection and already built a nice rig with one of them to max out 2 Voodoo2's . In this setup I also overclocked the 1000Mhz Coppermine a little to 1050MHz (140MHz FSB) and got zero trouble with the old and trusty FX5200 card.

Regards
Fabian

[quote="Mau1wurf1977"]Might be easier going for an early Pentium 4 platform instead. They all have intel chipsets, AGP and should be rock solid.

I had a look and even for S478 you can get slow CPUs starting at 1.4 GHz.

Not sure why this option never gets mentioned. It's always the Tualatin that gets the praise 😀

Thank you! I have mentioned this before! You can get a really nice socket 478 board and have the option to swap between 3.4ghzish 9x ROCKET to a p3 equivalent. especially since the 1.4ghz Willamette p4 is considered to be slower then a 1.4s Tualatin.

Tualatin can be used on boards with ISA slots. That is extremely uncommon for 478. With a Slot 1 setup one can also swap a Tualatin Celeron 1400 on a Slotket for a PII 233 in about a minute. Some PII CPUs can run down to 133 MHz. Disable caches and you have a 386. A very flexible setup.