Graphics card question.

First of all, no, there's a whole lot more to graphics performance than VRAM. VRAM is used for certain specific things and will make a big difference if there isn't enough for the particular game/3d operation/whatever in question, but otherwise by far the biggest factor is the horsepower of the graphics processor itself. There's also software--the driver for the card can make a very large difference, as can a particular program being tuned or otherwise better suited to a particular graphics card/chip/architecture.

All that said, a 9200 should outperform a Geforce2MX by a wide margin doing just about any operation. It's possible there's something physically wrong with the ATI card, but the fact that the 9200 is aftermarket and the MX isn't makes me think that it's a software issue.

A cursory Google search lead me to this thread, which is describing a problem much like yours:
http://discussions.apple.com/thread.jspa?threadID=1227727

...the solution apparently being to install the appropriate ROM update and software from ATI:
http://support.amd.com/us/gpudownload/Pages/macosx10-4x-3x-radeon.aspx

10.5 is supposed to have built-in drivers for that card, but it sounds like maybe it doesn't always get recognized by the system. Read through that thread and try what's suggested there, and if that fails, come back and ask.

 

You are indeed wrong. The GPU chip is what will determine how fast a scene can be rendered.

 

Whatever the problem is, that's not it--the 2MX is nowhere near even the 9200.

As for where to go next, I'm not sure. Have you tried removing the Nvidia card from the computer entirely, just to see if the problem is due to some sort of conflict there? There's some chance of that--perhaps the system is intentionally hamstringing itself to match the lesser of the GPUs available.

I'll give you one related example from my own experience: I have a mini running as a server/HTPC, connected to my TV. Unless I'm watching a movie off it, I usually only connect via Remote Desktop. One day the graphics performance went completely to hell--videos were tearing, skipping, and generally behaving as if the graphics card had suddenly stopped supporting Quartz Extreme.

Turned out the issue was that I had powered up the computer while the TV was off, and the MacOS, not seeing any attached monitor, had set the graphics to some sort of minimal mode, presumably to save resources (which you'd want in a headless server). Except it didn't kick back in when a monitor became available--you need to boot with the screen present.

Just one example of how a software problem can produce symptoms exactly like you're seeing.

A second thing you can try would be to get a spare hard drive and do a completely clean installation of the MacOS, with no additional drivers and none of your software and settings. If that works (either with or without the Nvidia card installed), then you've narrowed it down. If not, then either you do need some extra software or there's a hardware problem.

Also, the second suggestion in that thread was to update the firmware on the card--I believe one of the two downloads at the ATI site will do this for you, if your card needs it. If it hasn't been updated, you certainly should--it won't hurt, and may help.

There are usually specs for the GPU published somewhere obscure--polygons per second, memory throughput, number of processing units--but in practice the way you do it is you find a review with real-world benchmarks and compare.

Try http://www.barefeats.com/ -- they have plenty of useful benchmarks of graphics performance dating back quite a while.

 

A good way of knowing if one graphics card is better in theory than another one is to check it's bandwidth (in gigabytes per second GB/s)

To calculate the maximum bandwidth of a GPU you need to know 2 things about the card:
1) Memory clock (for example 650MHz)
2) Memory bus width (for example 64 bits)

Knowing these two numbers you can calculate them and divide them with 8 (because there are 8 bits in a byte). The resulting number is the max bandwidth of a graphics card in Gigabytes (or Megabytes) per second

Example:
GeForce4 MX 440
Memory clock : 400 MHz
Memory bus width: 128 bits

(400*128)/8 = 6400 MB/s

Divided with 1000 gives us 6.4 GB/s

 

That will give you a vaguely comperable number to compare, but in reality the drivers, other factors in the hardware, and the particular demands of whatever application you're running can make a big difference.

Example, using specs from the manufacturers: Try comparing the GT8800 (57.6GB/s, or as high as 103.7GB/s for some variants, though I think the Mac version is a standard GT), Radeon HD 4870 (115.2GB/s), FX 5600 (76.8 GB/sec), Radeon HD 3870 (72GB/s), and an older ATI X1900 XT (49.6Gb/s). Most of these have 512MB of VRAM, though the workstation-class Quadro FX 5600 has a whopping 1.5GB.

Based on the memory bandwidth, you'd expect the HD 4870 to be faster than everything else by far, followed by the FX 5600 (maybe better, considering its RAM advantage), R3870, GT8800, and the X1900 bringing up the rear.

But in practice, here's a real-world test in a pro app using the card to render a bunch of video effects:
http://www.barefeats.com/harper10.html

...in which the X1900 beats both the FX 5600 and GT8800 by a whopping 30-40%. It also outperforms the theoretically much slower R2600XT in some cases, lags somewhat in others.

This doesn't necessarily mean that the X1900 is, from a purely theoretical standpoint, faster than the Nvidia cards, but in this particular application either the hardware or software (probably software) is FAR better optimized for the ATI card.

But if you look at similar cards running gaming benchmarks under Windows:
http://www.barefeats.com/harper11.html

...you've got the GT8800 running literally three times faster than the R2600 XT.

Or this test of core image and Motion effects:
http://www.barefeats.com/nehal06.html

...in which, installed in the same machine, the GT8800 is faster than every other card on that list by a small margin, including the R4870 and FX 5600.

Point here being not that memory bandwidth isn't a useful number for general comparisons, but that it doesn't tell anywhere near the whole story.