GeForce 4 Ti.

Parhelia has no Win98 driver. G400 is slow but has nostalgia value. Quadro is probably going to have game quirks of some sort. GF6 is less ideal than GF4 for a few old games. ATI has compatibility issues for old D3D and OpenGL games.

Perhaps this configuration then:

Via C3-1200 - Matrox G400 Max
Dual PIII-850 - GF4 Ti4400
Dual PIII-1400 - GF 6600GT

I wasn't sure if it was safe to run the GF4 Ti4400 in a 440BX's AGP 2X slot since this card does not have a external power connector.

Yes, that's exactly what i would do.

G400max is good with weak cpu's
ti4400 bit overkill but leaves room for 4xAA
6600GT bit overkill but leaves room for 4xAA

To be honest I think a Voodoo Banshee / Voodoo3 might be best for that C3. G400 is actually fairly demanding of the CPU. C3 1200 performs something like a P3 600, I believe. You'd certainly benefit by having Glide available as an option, with its low CPU overhead, and as we saw in the other thread Voodoo cards tend to do well in general with slow CPUs. Though also I would be skeptical of 3DMark results because it was an obvious optimization target.

I do like the G400Max historical factor. I just checked current eBay pricing, AGP Voodoo3's are selling for $50+ shipped to Canada, so that option is out.

I do have some other AGP graphic cards that I didn't list because I wasn't originally considering them.

FX 5200 128 MB (NV34, 2.1 GB/s)
GF4 MX4000 64 MB (NV18, 64-bit, 3.2 GB/s)
GF4 MX440 64 MB (NV17, 128-bit, 5.3 GB/s) PNY Tech.
GF4 MX440 128 MB (NV18, 128-bit, 4.6 GB/s) XFX

Are any of these more suitable for the C3?

The MX440's would be good with the C3 IMHO. Assuming you are playing non DX8 games of course. I would think it would be faster than a Matrox G400 but I really do not know. Worst case scenario try them both out and use whichever one you like more in that system.

Would you rather have a MX440 with 64 MB RAM and 5.3 GB/s or a MX440 with 128 MB of RAM and 4.6 GB/s throughput? I beleive I grabbed these numbers from GPU-Z. Both cards are fanless.

I'd go with the 64MB version myself, I always found memory bandwidth more important than memory capacity on graphics cards. At least as far as real-world performance went.

Yep 64MB should be a more useful card. 128MB of ram is a bit of a waste on the MX440.

Anything with HW T&L will be good for computers with slower CPUs, so it should work well on the C3 (depending on drivers).

According to wiki, the G550 has T&L while the G400 does not. So will the G550 outshine the G400 Max?

I'm pretty sure the HW T&L in the G550 was either disabled or used for some specific features, that may have changed in later drivers but I wouldn't be surprised if it doesn't work the same as the Geforce cards. Matrox had pretty much given up on 3D gaming by then.

So I doubt it would make much difference. Also many games don't use T&L, but Quake 3 does and I have noticed improvements when using a Pentium or Pentium 2 and then adding in even a Geforce 2 MX200 PCI, gives a nice little speed increase.

For some reason the Tualatin PIII-S 1400 MHz will not function well with the GF 6600GT - there is some problem addressing all the memory when this card is installed. The FX 1100 will pass dozens of rounds of memtest but only if 2 GB is used (I can use 3 GB, but need to use CL3 timings).

Alternately, the GF4 MX400 can use 3 GB with CL2 timings and pass 40 rounds of memtest. But is the GF4 MX400 holding the dual Tualatin PIII-S system back?

The one other AGP card I have which works well in this system (3 GB CL2) is an Radeon HD 4350, which I originally sourced a year ago for GUI/flash/HD acceleration tests (which turned out being unremarkable). Would there be any benefit to using the HD 4350 card and making the system XP-only? Or should I stick with the MX400 and use 98SE/W2K/XP? Or use the FX1100 with GeForce drivers and accept only 2 GB RAM?

I have decided to leave the G400Max in the VIA C3 for now. The C3 runs well at 1400 MHz and stock voltage.

Running the C3 at 156x9 (1.4 GHz) or 133x10.5 (1.4 GHz) resulted in 26.1 fps in Quake III. 1280x1024x32

Running the C3 at 133x4, or 533 MHz, resulted in 25.2 fps in Quake III. Is a VIA C3 Nehemiah at 533 MHz too fast for a G400Max? Using a Tualatin 1.4 GHz also results in 26.1 fps. The G550 on the C3 only scored 21.6 fps.

Q2 gets 45.5 fps at 1280x1024x32
Q1 gets 28.6 fps at 1280x1024x32, but sometimes exhibits choppy movement.

NT4 cannot run any of the 3 Quakes. I am not sure if this is a Matrox G400/550 driver issue or if it is related to the VIA C3 on its own, or a VIA C3 conflict with the Matrox drivers.

How does Q3 run at lower resolutions?
As a comparison some numbers for Q3 on TNT2 Ultra which should be in the same league as the G400Max:
thandor.net: 1280x1024x32 20.4 fps
http://www.tomshardware.com/reviews/vga-charts-i,453-4.html 1024x768x32 28.2fps

I think the question is not if the C3 and the G400Max are a good match, but what performance is to be expected depending on game and resolution with a G400Max. If you insist on the resolution you should look for other cards.
G550 is indeed usually slower than G400, although IIRC there were a few cases where the G550 wins.

1280x1024x32 is the resolution I am after. I have settled upon a GeForce4 MX440 for this VIA C3-1200 Nehemiah build. I tested the GF6600GT, the GF6200, Quadro FX500, Quadro FX1100, and GF4 MX440, all of which perform at 58 fps in Quake 3 at 1280x1024x32. The CPU is holding back the GF4, but I am not sure by how much. Increasing the Nehemiah's multiplier to 10.5x results in increased frame rates, confirming that the GF4 is being held back by the CPU. At 1400 MHz, Q3 scores 65 fps.

Overall, I am fairly impressed with the VIA C3 Nehemiah for its Quake 1, 2, and 3 performance. Perhaps it would make for a good low-power gaming build. Do Quakes 1 and 2 max out at 60 fps?

By way of comparison, my dual Tualatin 1.5 GHz system with Matrox Parhelia 128 scores 91 fps in Quake III (demo four).

Here's a well packaged comparison of the Tualatin vs. the Nehemiah.
http://ixbtlabs.com/articles2/roundupmobo/via … 3-nehemiah.html

Your MX440 cards are the 128bit variant, according to thandor.net benchmarks a P3 1400 can do 103.8fps in the same conditions. The 64bit model still scores 67.4fps so this would be very balanced.

Q1 and Q2 don't have a built-in frame limiter if that's what you are asking. If you don't get higher measurements I would suspect that vsync is on.

The Parhelia benchmark is probably GPU limited.

feipoa wrote:

Here's a well packaged comparison of the Tualatin vs. the Nehemiah.
http://ixbtlabs.com/articles2/roundupmobo/via … 3-nehemiah.html

Edit, nevermind it is the Nehemiah there indeed, I thought it was the Ezra-T review again.
Thanks for the heads up.

idspispopd wrote:

Your MX440 cards are the 128bit variant, according to thandor.net benchmarks a P3 1400 can do 103.8fps in the same conditions. The 64bit model still scores 67.4fps so this would be very balanced.

Q1 and Q2 don't have a built-in frame limiter if that's what you are asking. If you don't get higher measurements I would suspect that vsync is on.

The Parhelia benchmark is probably GPU limited.

Thank you for your input. You were correct - vsync was enabled for the Q1 and Q2 tests, which limits my frame rate to the monitor's refresh rate. By disabling vsync, I get GLQuake = 117.5 fps; Quake II = 132.9 fps. I did not expect such performance out of a $30 low power VIA/Winchip product.

I did not realise that a 64-bit MX440 card existed. Would that be the MX4000?

I benchmarked my Dual PIII-850 slot 1 440BX system with the Ti4400. Quake III scores 67 fps (all quality settings at max). So the Via C3 Nehemiah at 1400 MHz is scores similarly to a PIII-850, but with a weaker graphics card. The PIII-850 was running WinXP SP3, while the Via C3 Nehemiah was running Win98SE, so I'm sure this has some impact on frame rate.

EDIT: I am beginning the W2K and XP Pro segment of the system setup (NT4/98SE already complete). The TUSL2-C manual mentions there is a Support CD for this motherboard. Does anyone know where to find this disc image?