r300g on AGP is known to be slow. I don't think I made any optimizations to make AGP any faster, as I don't have such a system and neither do I want one. When adding optimizations, I was mainly testing R500 with PCIe, where most of the user base is. Not sure if anybody else was trying to optimize for AGP...
Also, using Unity might have had a negative impact on the results.
Michael says in the article that the mGPU was PCI-E, which matches what my old Inspiron 6000 had as well (128MB x300 PCI-E). That being said, if it was missing Hi-Z and any other off-by-default optimizations, that would explain part of the performance shortfall.
Of course it's also possible that there's room for improvement in the shader compiler, state trackers, drm/kernel-level, and TGSI to hardware back-end.
When adding optimizations, I was mainly testing R500 with PCIe
I'm really curious about how optimized is R500 with heavy loads. I'm planning to benchmark the fastest R500 card (X1950XTX) which I've already bought some months ago with doom3 and unigine heaven, mesa against catalyst. To avoid cpu bottlenecks I will use a six-cores Intel i7 3930 @5GHz.
You should be bloody ashamed of yourself about how ignorant you are. Without those people, you wouldn't even have glxgears running.
As always, patches welcome.
He stated earlier that he was being sarcastic, reading his earlier comments I think this can be safely taken as a joke.
For the record, I benefit from your work and would like to thank you. If we ever meet in an establishment dedicated or allowing the consumption of alcoholic beverages, I would like to buy you a (several) pint(s).