I see you still haven't lost the art form of replying with some
amazingly irrelevant statement while still presenting it as if it's
exactly what the other person needed to hear to fully comprehend your
logic.

Really though, the UP1500's AGP will never work like a regular AGP,
because it fundamentally can't. (Not that what I'm going to tell you
has anything to do with SRM's ability to initialize AGP video cards,
but just disproves your silly statements about following the spec "to
the letter")

Take a quote from Ivan Kokshaysky, maybe the only guy to try to
actually try to program it,
"""
There is quite fundamental conflict between the Alpha architecture and
x86 AGP implementation - Alpha is entirely cache coherent by design,
while x86 AGP is not (I mean native AGP DMA transactions, not a PCI
over AGP). There are no such things as non-cacheable mappings or
software support for cache flushing/invalidation on Alpha, so x86 AGP
code won't work on Nautilus.
"""
http://alphalinux.org/wiki/index.php/UP1500

It's like this because it would have been a ton of work for anything
else. When DEC/Compaq/HP designed the Marvel and Titan systems and
their chipsets, they had the flexibility to implement what they
needed, unrestricted by a drop-in AMD chipset. Take another quote from
Ivan: "they are sort of AGP extension of traditional Alpha PCI IOMMU
and are fully cache coherent."

Don't bother repeating that around, because inevitably it'll get
mangled into some anecdote about your old 500 MHz Celeron system again.