strange results: writing to pixel data ranges w/ pixperf and nvidia's 41.09 drivers

Has anyone else encountered strange write pdr behavior with the latest (41.09) nvidia drivers for win2k/xp? Running matt craighead’s pixperf (http://ds.dial.pipex.com/town/plaza/yr20/PixPerf.zip) on our system yields very strange results. The read, read pdr, and draw speeds are reasonable, but the write pdr case seems hosed (although drawing with a read pdr seems fine). I would normally assume I was doing somethign foolish, but there is not much to screw up in this case:

pixperf -size 512 -type ubyte -format rgba -read
~45 Mp/s

pixperf -size 512 -type ubyte -format bgra -read
~45 Mp/s

pixperf -size 512 -type ubyte -format rgba -read -readpdr
~47 Mp/s

pixperf -size 512 -type ubyte -format bgra -read -readpdr
~47 Mp/s

pixperf -size 512 -type ubyte -format rgba -draw
~138 Mp/s

pixperf -size 512 -type ubyte -format bgra -draw
~138 Mp/s

pixperf -size 512 -type ubyte -format rgba -draw -writepdr
~4 Mp/s !!!

pixperf -size 512 -type ubyte -format bgra -draw -writepdr
~6 Mp/s !!!

pixperf -size 512 -type ubyte -format rgba -draw -readpdr
~140 Mp/s !!!

pixperf -size 512 -type ubyte -format rgba -draw -readpdr
~140 Mp/s !!!

Any ideas?

Our system:

Dual 1.7 Ghz Xeons
i860-based motherboard (super micro super p4dc6+)
agp 4x
1 GB RDRAM
64 MB agp aperture
geforce 4 ti 4600 (128 MB DDR)

I get the same results as you, the spec suggests that drawpixels is not currently accelerated.
http://www.nvidia.com/dev_content/nvopenglspecs/GL_NV_pixel_data_range.txt

bummer – I thought nvidia might have already implemented it given matt’s comments in http://www.opengl.org/discussion_boards/ubb/Forum3/HTML/007256.html.