I think that it is big posability that AGP GPU transfer registers are different in different cards.
But i found out how to accelerate CPU access to FrameBuffer.
Steps:
1. Enable write posting
2. Enable Write-Combining for linear frame buffer.
3. Enable AGP fastest speed.
And without AGP i got 178MB/sec Frame buffer access.
Then i turned on AGP 4x and it was in front of me 720MB/sec.
So in VESA 118h mode resolution 1024x768x32bit true color
my simple bootable OperSysTest Was going 240FPS.
God Thank you.
1. write posting is something like when CPU decide to send something through PCI bus it just sends command to Northbridge and continue to go it's way, not waiting for PCI transaction to compleate.
2. writhe combining is CPU caching technique where is it waiting for for some big block amount of data to be written , not just one byte, but bad thing is that data would not be written in right order , but who cares that for FrameBuffer.
3. Agp is AGP
Did i just Googled something new for this forum or this was in some topic already. I am sure that 10% of people on this forum knew this but they wouldn't tell. Well Big Brother is watching.
My 1024x768 os is not finished yet, i would call it PilotOS but if somebody wants to see FPS-es like PROVE, i can send link to not BETA but GAMA version

and put it in new topic, ofcourse with source.
How much will it be on PCI-E x16 about 1000FPS

.
This FPS is only MAX value with 100% of CPU usage. You probbably wouldn't need that much.
I lost all hope for Vesa LFB acceleration but suddenlly...
I would like to thanks to John Hinkley from
http://www.fastgraphics.com on his FAST VID 1.10 wich shows how to enable write-combining.
Thanks to this site info
http://www.osdev.org/osfaq2/index.php/AGP%20information
And i would like to thanks to AGP Specs.
And to all good people here who want to create something, and to help in rising humankind to higher level.
Now i am going to have some sleep.

.