First post, by Kahenraz
- Rank
- l33t
I was writing another thread about these two cards where I shared a lot of benchmarks across two motherboards, a Coppermine Pentium 3 and a fast Pentium 4, VIA and Intel chipsets, but then I thought... wait a minute. Maybe the answer is obvious. What about the memory bus?
I remembered reading this quote by Phil:
PhilsComputerLab wrote on 2017-01-25, 00:48:GeForce4 MX 440 is actually a decent card. Do stay away from the 420, that one usually has half the memory bandwidth and also be […]
GeForce4 MX 440 is actually a decent card. Do stay away from the 420, that one usually has half the memory bandwidth and also be careful with SE versions. There are far more variations of these cheap cards than what you can find on wikis.
440 is a great substitute for a GeForce DDR and runs pretty much any Windows 98 game at 1280 x 1024 fluently.
Now the MX 460, that on is already harder to find. It can do 300+ FPS in Quake II at 1024 x 768 and was a short-lived card.
What's cool about these cards is that they are cheap and easy to find, just like Voodoo3 cards used to be a couple of years back.
And Wikipedia also suggests that the MX 420 is a 64-bit memory bus only. And that the 440 is 128-bit only. So I pulled up HWiNFO to check and guess what? My 420 PCI is 128-bit and the 440 is 64-bit! What the heck?
I only have one sample of each but I will be ordering more AGP 440s to investigate this further.
Another important comparison is that the MX 420 has 8 memory chips and the MX 440 only has 4. But I don't know if this is always a good indicator of the memory bus.
Here is a quote from the thread I deleted which has some benchmark scores and details about the hardware I was testing on. It ends abruptly without questions or conclusion because that is the point at which I thought to check the memory bus.
Does anyone have any information on why my two cards have a different memory bus than expected?
Why am I getting worse scores for AGP than PCI? […]
Why am I getting worse scores for AGP than PCI?
I managed to find a GeForce 4 MX 420 on recently and wanted to compare it with the closest card I had for comparison but the results were not what I had expected. I don't have a 420 in AGP so I compared it against a 440 for AGP 8x, which should defeat it handedly.
To add further context, here is a quote by Phil from another thread:
PhilsComputerLab wrote on 2017-01-25, 00:48:GeForce4 MX 440 is actually a decent card. Do stay away from the 420, that one usually has half the memory bandwidth and also be […]
GeForce4 MX 440 is actually a decent card. Do stay away from the 420, that one usually has half the memory bandwidth and also be careful with SE versions. There are far more variations of these cheap cards than what you can find on wikis.
440 is a great substitute for a GeForce DDR and runs pretty much any Windows 98 game at 1280 x 1024 fluently.
Now the MX 460, that on is already harder to find. It can do 300+ FPS in Quake II at 1024 x 768 and was a short-lived card.
What's cool about these cards is that they are cheap and easy to find, just like Voodoo3 cards used to be a couple of years back.
My initial test was with a VIA VT82C691 Apollo Pro motherboard which I had paired with a Pentium 3 667/133Mhz Coppermine CPU (this is the fastest that board can handle). The motherboard supports up to AGP 2x. My comparison benchmarks were done with 3D Mark 99 and 2000. I used the NVIDIA drivers version 56.64 because 3D Mark 2000 would always crash with 45.23 at the very end when benchmarking the 420 PCI.
VIA, P3 667Mhz, WinME, v56.64
GeForce 4 MX 420 PCI
3D Mark 99: 4816
3D Mark 00: 4078
GeForce 4 MX 440 AGP 8x
3D Mark 99: 4316
3D Mark 00: 2362These results were very interesting and unexpected. Of course, I needed more data. Maybe there was some issue with the chipset, maybe the CPU was bottlenecking something in the drivers, or maybe there was a problem with the AGP implementation on this board. I tried several different driver versions between 45.23 and 82.69 with no improvement (just lots of driver bugs after 56.64).
To rule out bottlenecks and the VIA chipset, I ran the same test on a Pentium 4 3Ghz on a motherboard with an Intel 845 chipset. This motherboard supports up to AGP 4x The results improved with the CPU speed but the PCI card was still faster.
Intel, P4 3Ghz, WinME, v56.64
GeForce 4 MX 420 PCI
3D Mark 99: 6025
3D Mark 00: 6087
GeForce 4 MX 440 AGP 8x
3D Mark 99: 6042
3D Mark 00: 3404