AGP 2x vs PCI ?
AGP 2x vs PCI ?
AGP 2x, even at 1.0, is expected to perform significantly faster than a standard PCI slot, yet its speed depends on the actual bandwidth available for the GPUs. The 8400GS was offered as a PCI card, while the 9400GT and 9500GT were also PCI-based. These versions were likely considerably slower compared to PCIe cards and marked the final attempt at AGP support. They are newer than any of the 8x AGP Radeon 9000 models, but using an AGP 8x card in a 2x slot could cause problems. It seems more practical to seek a better motherboard or attempt to run DOS and Windows 98 on updated hardware. Back then, I managed to dual boot DOS and Windows XP on an Athlon X2 with an 8800GTS PCIe card without issues. Most Win98 titles should work in compatibility mode under XP, though minor adjustments might be needed. Many classic games still function adequately in Windows 10 and 11, although some outright fail. I know several can be wrapped in a glide wrapper to run older OpenGL/OpenCL games effectively.
The motherboard lacks an AGP slot, making it impossible to install an AGP card. It functions as an internal AGP card. I attempted to boot Windows 98 on an Athlon XP 2000+ system, but the computer froze during the process.
Generally, Super Socket 7 boards suffered from poor AGP implementations, causing AGP-capable cards like ATI to face significant stability issues (since Intel never released a 100MHz FSB Socket 7 chipset). At that time, the preferred choice was a manufacturer that didn’t use any AGP features—such as 3DFX. This explains why a Voodoo 3 3000 PCI performed just as well as its AGP counterpart (and Voodoo 2 was purely PCI-based). Sadly, the prices for these older pieces of hardware are now extremely high, and it’s hard to believe how many I overlooked when they were only $1.95 used. Considering how DOS and 2D Windows performance has been neglected in newer cards, it might not be wise to chase the latest PCI options unless you’re prioritizing top DX9 speeds in Windows 98 at the cost of stability. I’ve definitely run into major headaches with cards that included PCIe-to-AGP bridge chips, and I doubt PCIe-to-PCI would improve things much.
BTW, if you were aiming for the highest DX9 speed possible with a 32-bit PCI card, you'd likely need either a rare Zotac GT430 or the more typical GT520/GT610—cards that are essentially the same but offer about half the capabilities of the GT430. These models typically drop 30 to 60% performance on the 133MB/s PCI bus compared to x1 PCIe 1.1, which at 250MB/s performs similarly to AGP 1x at 266MB/s.
View: https://www.youtube.com/watch?v=yNEge5r6-mg
I’m not sure how these would function well on Super Socket 7, which often had subpar PCI support, or how reliable the bridge chip would be. I anticipate many timing problems for a GPU built to run on a port that uses a bridge to a bus, forcing it to share and wait its turn. Of course, such a setup would have never been tested when these GPUs were first released, since Super Socket 7 was already over a decade old at the time. At least the Fermi architecture was the last NVIDIA platform to fully implement hardware triangle setup, which should have eased the burden on weaker CPUs.
FWIW, the Trident 9880 performed adequately for 640x480 gaming in OpenGL, but the company shut down shortly after its release, leaving driver issues unresolved and DX6 performance heavily reliant on AGP features your system might lack. If you manage to source a 8400GS or 9500GT at a low cost, the results would be significantly better than what you could achieve with PCI options, especially if you can run games in 800x600 resolution—they’re quite comparable to integrated graphics from the later Core 2 generation, such as the GMA X4500HD.