{
>>> If you wait for the retrace each time, you won't
>>> flicker, but you will limit the speed to the
>>> retrace speed (about 68-70) frames a second.  If
>>> you wanted to truly find the speed, do it without
>>> the retrace wait, and you may get much faster.  You may
>>> already know this, in which case I am just
>>> wasting space...

>> And if you use a tripple buffering you will get no flicker and
>> the maximu speed.

> What the heck is triple buffering?

You are using THREE different screens (no problem in x-mode). One (A) is
displayed, second (B) is waiting, third (C) is being drawn. After third C is
ready B is being displayed and A is being drawn. No flickering, no waiting, no
problem...

>>> By the way, as a general note, I have found that
>>> the single palette entry change also waits for the
>>> retrace.  So if you put that in a program, it will
>>> severely limit the speed (voice of experience).

>> Waits? Only if you are using BIOS for it.
> I am talking about the BIOS op.  Doing it directly takes
> more info than I have.

OK, so take a look - and learn ;-)
}

{$A+,B-,D+,E+,F-,G+,I-,L+,N-,O-,P-,Q-,R-,S-,T-,V+,X+,Y+}
{$M 16384,0,655360}

uses crt;

var
  i       : integer;
  licznik : byte;
  paleta  : array[0..767]of byte;
  screen  : array[0..63999]of byte absolute $A000:0;

{ This is necessaery for drawing plasma. Don't mind. It is the same piece of
code I use in voxel space code posted here for several times, not necessarilly
by me. }

function ncol(mc,n,dvd : integer): integer;
var
  loc : integer;
begin
  loc:=(mc+n-random(2*n)) div dvd;
  ncol:=loc;
  if loc>250 then ncol:=250;
  if loc<5 then ncol:=5
end;

procedure plasma(x1,y1,x2,y2 : word);
var
  xn,yn,dxy,p1,p2,p3,p4 : word;
begin
  if (x2-x1<2) and (y2-y1<2) then EXIT;
  p1:=screen[320*y1+x1];
  p2:=screen[320*y2+x1];
  p3:=screen[320*y1+x2];
  p4:=screen[320*y2+x2];
  xn:=(x2+x1) shr 1;
  yn:=(y2+y1) shr 1;
  dxy:=5*(x2-x1+y2-y1) div 3;
  if screen[320*y1+xn]=0 then screen[320*y1+xn]:=ncol(p1+p3,dxy,2);
  if screen[320*yn+x1]=0 then screen[320*yn+x1]:=ncol(p1+p2,dxy,2);
  if screen[320*yn+x2]=0 then screen[320*yn+x2]:=ncol(p3+p4,dxy,2);
  if screen[320*y2+xn]=0 then screen[320*y2+xn]:=ncol(p2+p4,dxy,2);
  screen[320*yn+xn]:=ncol(p1+p2+p3+p4,dxy,4);
  plasma(x1,y1,xn,yn);
  plasma(xn,y1,x2,yn);
  plasma(x1,yn,xn,y2);
  plasma(xn,yn,x2,y2)
end;

begin
  asm
    mov  ax,13h
    int  10h
  end;
{ Generating palette RGBs }
  for i:=1 to 170 do paleta[3*i]:=round(63*sin(i/170*pi));
  for i:=1 to 170 do paleta[3*i+256]:=round(63*sin(i/170*pi));
  for i:=1 to 170 do paleta[(3*i+512) mod 768]:=round(63*sin(i/170*pi));
  plasma(1,1,319,199);
{ Licznik - it means 'counter' in Polish.  }
  licznik:=0;
  repeat
{ Wait for retrace. }
    repeat until (port[$03DA] and 8)=0;
    repeat until (port[$03DA] and 8)=8;
{ Changing palette - we start with color number licznik }
    port[$3C8]:=licznik;
{ Three outsb are copying whole RGB to VGA register. After those three
instructions value in port $3C8 is incremented. Here I'm redefining whole
palette, but there is no problem in changing only one color. }
    asm
      mov  si,offset paleta
      mov  cx,768
      mov  dx,$3C9
      rep outsb
    end;
    inc(licznik);
  until keypressed;
  asm
    mov  ax,3h
    int  10h
  end;
end.