PCI Express Bandwidth Testing and Video Performance
Serious Magic, a vendor of video editing software and other related tools, created a benchmark a couple of years ago that tested GPU-to-host writeback performance which back then was seriously limited by the graphics card's driver. It caused The Tech Report to investigate and write a famous article on driver performance when doing host writebacks. That same tool is a fine stress of the GPU-to-host interface today, even on PCI Express.TexBench 1.3 is used to test host-to-GPU texture uploads, to figure out what the bandwidth available is there. I use a quad-texture situation where the host uploads 8MB of 128x128 textures in burst mode, hopefully revealing any limitations in the bus bandwidth.
Host-to-GPU texture upload performance
Compared to 6600 GT, X700 XT has no more real-world bandwidth when writing to the GPU from the host. Infact, all tested hardware sits at the same levels in this test, indicating the limitating lies in the PCI Express controller itself.
GPU-to-host texture download performance
The AGP8X bridged NVIDIA GPUs show a set limit in their GPU-to-host writeback performance. The native PCI Express boards, 6600 GT and X700 XT, have no such limit, comfortably enjoying twice the bandwidth. That's some way off any theoretical maximum, but a consistent, reproducible difference between the two interconnects.
Video Performance
Video decode performance on RV410 is some 40% more efficient, in terms of the CPU hit while doing the decode, than NV43. It's done entirely on the CPU in NV45 and NV35. Decoding video in the pixel shader, as NVIDIA's GPU's seem to do, isn't trivial. ATI's GPUs have more dedicated decode paths for video it seems. A topic for a future article?
