The Geforce 4 MX 4000 PCI is about the same speed as a Voodoo 2 SLI on a Pentium 3 700MHz and a little faster than a Kyro 2:

3DMark99 640x480
5637 | 16bit | Geforce 4 MX 4000 PCI (Jaton, 128MB) | Dual P III 1000MHz | VIA 694 | Win2k | Forceware lowest settings

3DMark99 800x600
5540 | 16bit | Geforce 4 MX 4000 PCI (Jaton, 128MB) | Dual P III 1000MHz | VIA 694 | Win2k | Forceware lowest settings
5920 | 16bit | Hercules 3D Prophet 4500 (Kyro 2, 64MB) | Dual P III 1000MHz | VIA 694 | Win2k | Last official "Kyro 2" driver (PowerVR)

3DMark2000 1024x768
5066 | 16bit | Geforce 4 MX 4000 PCI (Jaton, 128MB) | Dual P III 1000MHz | VIA 694 | Win2k | Forceware lowest settings
4924 | 16bit | Hercules 3D Prophet 4500 (Kyro 2, 64MB) | Dual P III 1000MHz | VIA 694 | Win2k | Last official "Kyro 2" driver (PowerVR)

F2bnp wrote:

My bad then. Very thorough explanation, thanks!

Worth pointing out: that Anandtech article is based on AGP cards. While the PCI FX 5200 was able to hold its own with AGP GeForce 2 Ultra (in my admittedly limited test), I'm not sure to what extent it is or isn't being bottlenecked on PCI (e.g. would an AGP card score much higher on the same platform/same circumstances), so the PCI FX 5600 in practice may not have as much of an advantage.

Arctic: Very interesting; thanks for posting the #s. 😀

I get more than 6000 3dmark in 800x600 using a Diamond TNT-1-PCI 16megabyte card, in this little machine...
I just fire up 3dMark99 in default settings.

I really don't know how and why it is so. My first guess is that the cpu is feeding the TNT in a really heavy way.

Edit:
The machine is running with 512mb dual side PC-133 and a Diamond TNT1-16mb-PCI and everything else stock.
If this clears up anything, as to why the TNT1 is just hammering away like this.

Well that is probably due to Windows 2000. Also the TNT should have a 128Bit Interface. 😀

Arctic wrote:

Well that is probably due to Windows 2000. Also the TNT should have a 128Bit Interface. 😀

It was tested in Win98SE. Installing Win2K as of now, to test out what differences there are.

All of the FX 5200 cards I'm seeing these days are gimped 64-bit memory bus. The Dell OEM GeForce4 MX 440 card is very nice (I have acquired several). Good build quality, fast, 64MB RAM, 128bit bus and inexpensive. Anything from the Ti series is nice, I am waiting for my Quadro4 700 XGL's to arrive, same GPU as the GeForce4 series.

squareguy wrote:

All of the FX 5200 cards I'm seeing these days are gimped 64-bit memory bus. The Dell OEM GeForce4 MX 440 card is very nice (I have acquired several). Good build quality, fast, 64MB RAM, 128bit bus and inexpensive. Anything from the Ti series is nice, I am waiting for my Quadro4 700 XGL's to arrive, same GPU as the GeForce4 series.

They come in PCI edition's right? And not AGP-Only.

brostenen wrote:

squareguy wrote:

All of the FX 5200 cards I'm seeing these days are gimped 64-bit memory bus. The Dell OEM GeForce4 MX 440 card is very nice (I have acquired several). Good build quality, fast, 64MB RAM, 128bit bus and inexpensive. Anything from the Ti series is nice, I am waiting for my Quadro4 700 XGL's to arrive, same GPU as the GeForce4 series.

They come in PCI edition's right? And not AGP-Only.

As far as I'm aware there is no NV2x PCI, but there are MX (NV1x) cards for PCI.

My bad I wasn't thinking about PCI only.

Keep an eye out for a Quadro FX 600 also. Think of it as a turbocharged GeForce FX 5200 and yes it is PCI!

EDIT: this card only comes in 128-bit memory bus and PCI so you know you will get what you need

I own a couple of Dell Optiplex computers (GX1 and GX110) without AGP. I still like them quite a lot because of their neat modular construction and ability to turn off internal cache on the fly with a key combo.

I was surprised how capable PCI graphics can be, and now I am trying to find the best PCI card as well.

The 3DMark01 for the FX5500 is not too shabby (run on the GX110 -- a 933MHz P3 on an 810 board):

All in all, I think that you can have a very respectable DX8 gaming experience with a PCI card. I have a 128-bit Radeon 9250 on the way, and I'm curious to find out how it stacks up. Just to reiterate what others have said, you need to get a 128-bit card. I have another FX5500 that is 64-bit, and it is terrible.

boxpressed wrote:

All in all, I think that you can have a very respectable DX8 gaming experience with a PCI card. I have a 128-bit Radeon 9250 on the way, and I'm curious to find out how it stacks up. Just to reiterate what others have said, you need to get a 128-bit card. I have another FX5500 that is 64-bit, and it is terrible.

The 128-bit 9250 PCI is actually not too bad. It easily outperforms a 64-bit AGP FX5200:

Standard Def Steve wrote:

The 128-bit 9250 PCI is actually not too bad. It easily outperforms a 64-bit AGP FX5200:

Cool. Which version of Catalyst are you using?

I took that screenshot a year ago, but iirc it was Catalyst 5.6

I've used a GeCube Radeon 9200SE 128MB in a Socket 7 machine with good success. It played everything fine, at least on the level of a V3 I would say, but not sure exactly.

The SE is, of course that's the one I have, the gimped version with half the memory width 🤣

@Standard Def Steve

Is your PCI clock running at 33MHz?
Because your FSB is 151MHz. If your PCI is overclocked then you don't have the control values, this will increase your result in 3DMark.

boxpressed wrote:

Standard Def Steve wrote:

The 128-bit 9250 PCI is actually not too bad. It easily outperforms a 64-bit AGP FX5200:

Cool. Which version of Catalyst are you using?

According to GPU-z it's 6.14.10.6542

Are you guys able to get GPU-Z running in 98, if so how?

Get an older version ^^.

(I believe 1.57 is the last to work in Win98SE).

I am using 1.40 on my win98 systems.
It works but the font is tiny and hard to read.

I have an old version but it complains about a missing DLL from Windows NT 4.0 or Windows 2000, I will try a different download.

UPDATE: Please disregard the numbers below. Vsync was still on for D3D, which is why the results are so similar between the Ti4600 and Quadro2. Will update soon.

Following my previous post, I want to post some benchmarks of the FX5500 128-bit PCI done on a common system.

INTEL SE440BX-2
PIII-700 SLOT 1 COPPERMINE
128MB PC100 SDRAM
AUREAL VORTEX 2
AWE64 VALUE

1UNREAL GOLD                   1600X1200     1024X768
2
3QUADRO 2 PRO AGP (45.23)      47.5          55.9
4GF4 TI4600 AGP (45.23)        48.3          56.5
5GF FX5500 PCI (56.64)         39.4          48.5

1QUAKE SW 1.06                 1024X768     640X480
2
3QUADRO 2 PRO AGP              23.3          50.2
4GF4 TI4600 AGP                23.2          50.1
5GF FX5500 PCI                 20.6          44.9

As I hinted, the 128-bit FX5500 acquits itself well compared to its AGP rivals. The PCI bus costs the FX5500 a few FPS in software rendering with Quake, as both AGP cards turn in nearly identical scores.

I have some OpenGL benchmarks for Quake 2, but I don't think that they're accurate for the FX5500. This is because the 56.64 drivers don't really turn off vsync even though the control panel says it is turned off. So the results hover near the 60 and 75 FPS marks (whereas the Quadro2 and Ti4600 turn in scores between 145-184 using the 45.23 drivers). If anyone knows why this is happening, please let me know. It's driving me crazy.