DOS ain't dead

> > Please remove my email from the post.
>
> Oops -- i apologize for that. I just tried to edit the original post but
> got the message:
>
> Entry cannot be edited any longer after 1440 minutes.
>
> I don't know what else to do. Perhaps RR can edit it? (I contacted
> admin.)

Done, though I think it's almost futile to expect no spam these days. :-/

> > Hm strange. Try last ver of vesatest.
> > http://rayer.g6.cz/1tmp/vesatest.zip
>
> Ok, will do.

It looks like VESATEST can successfully autodetect LFB and set WC on the card. So it must be a bug in MTRRLFBE? Here's the relevant comparison:

VideoModeNumber = 4112h
921600 B allocated for FrameBuffer
LFB Selector = C7h
LFB PhysBasePtr = E0000000h
Measured refresh rate = 60.4 Hz
Measured 81 FPS (FrameBuffer>VRAM)
(transfer speed 71 MB/s)

VideoModeNumber = 4112h, MTRR WC set
921600 B allocated for FrameBuffer
LFB Selector = C7h
LFB PhysBasePtr = E0000000h
Measured refresh rate = 60.1 Hz
Measured 227 FPS (FrameBuffer>VRAM)
(transfer speed 200 MB/s)

Also, if of any help, here's the debug dump from MTRRLFBE 1.4:

MTRR-WC enabler for VESA LFB 1.4 (C) 2005-2012 by Martin Rehak
Compiled by GCC 4.7.0 at 09:54:27, May  9 2012
Host machine CPU vendor: GenuineIntel, ID: F27h

VESA 2.0 ATI RAGE128 [16384 kB]
ATI Technologies Inc.
R128
01.00
LFB address: 00000000h
ERROR: invalid address or LFB not supported!
MTRR #0: base = 000000000h (    0MB), mask = FE0000000h (  512MB), WB, used
MTRR #1: base = 01FF80000h (  511MB), mask = FFFF80000h (    0MB), UC, used
MTRR #2: base = 000000000h (    0MB), mask = 000000000h (    0MB), UC, unused
MTRR #3: base = 000000000h (    0MB), mask = 000000000h (    0MB), UC, unused
MTRR #4: base = 000000000h (    0MB), mask = 000000000h (    0MB), UC, unused
MTRR #5: base = 000000000h (    0MB), mask = 000000000h (    0MB), UC, unused
MTRR #6: base = 000000000h (    0MB), mask = 000000000h (    0MB), UC, unused
MTRR #7: base = 000000000h (    0MB), mask = FFF000000h (   16MB), WC, used
Setting MTRR #7...
MTRR area 00000000-00FFFFFFh was set to mode: WC

Upon further testing, i discovered that MTRRLFBE actually slowed down measured throughput on this card, both BS and LFB. Using the VSpeed test program (from the FastVid package):

SYSTEM DEFAULT:
  Copy DRAM to banked VGA:           70.41 million bytes per second
  Copy DRAM to linear framebuffer:   75.57 million bytes per second

MRTTLFBE LFB WC:
  Copy DRAM to banked VGA:           13.64 million bytes per second
  Copy DRAM to linear framebuffer:   17.69 million bytes per second

VESAMTRR (reports MSRs 204-205 modified):
  Copy DRAM to banked VGA:           70.41 million bytes per second
  Copy DRAM to linear framebuffer:  213.07 million bytes per second

FASTVID x11 16 E0000000:
  Copy DRAM to banked VGA:          210.49 million bytes per second
  Copy DRAM to linear framebuffer:  212.71 million bytes per second

- Doug B.