Binary Images of Triangle Counts

Mathematica, 126 122 121 89 bytes

Image[1-Thread@IntegerDigits[l=Round[(#+3#~Mod~2)^2/48]&/@Range@##,2,⌈2~Log~Max@l⌉]]&

This defines an unnamed function taking the two integers as parameters and displaying the image on screen. It plots each square as a single pixel, but if you like you can actually zoom in.

I'm now using an explicit formula given in the OEIS article (first one in the Mathematica section, thanks to David Carraher for pointing that out). It's also blazingly fast now.

Here is the indented code with a few comments:

Image[1-Thread@IntegerDigits[   (* 3. Convert each number to padded binary, transpose
                                      invert colours, and render as Image. *)
    l = Round[
      (#+3#~Mod~2)^2/48
    ] & /@ Range@##,            (* 1. Turn input into a range and get the Alcuin
                                      number for each element. *)
    2,
    ⌈2~Log~Max@l⌉               (* 2. Determine the maximum number of binary digits. *)
]] &

Here is the output for 0, 600:

CJam (56 55 53 chars) / GolfScript (64 chars)

CJam:

"P1"q~,>{_1&3*+_*24+48/}%_:e>2b,\2_$#f+2fbz(,@@:~~]N*

GolfScript:

"P1"\~,>{.1&3*+.*24+48/}%.$-1=2base,\{2.$?+2base}%zip(,@@{~}/]n*

Both produce output in NetPBM format, and they're essentially ports of each other.

Dissection

CJam                 GolfScript           Explanation

"P1"                 "P1"\                NetPBM header
q~,>                 ~,>                  Create array [m .. n-1]
{_1&3*+_*24+48/}%    {.1&3*+.*24+48/}%    Map the sequence calculation
_:e>2b,\             .$-1=2base,\         Compute image height H as highest bit
                                          in largest number in sequence
2_$#f+2fb            {2.$?+2base}%        Map sequence to bits, ensuring that
                                          each gives H bits by adding 2^H
z(,@@                zip(,@@              Transpose and pull off dummy row to use
                                          its length as the "width" in the header
:~~                  {~}/                 Flatten double array and dump on stack
]N*                  ]n*                  Separate everything with whitespace

Thanks to Optimizer for CJam 56 -> 53.

Pyth - 101 60 59

Outputs a .pbm. Can likely be golfed more.

Km.B/++24*dd**6%d2d48rvzQJCm+*\0-eSmlkKlddK"P1"lhJlJjbmjbdJ

Highly ungolfed because I will be translating to Pyth.

Explanation coming next. Right now look at the equivalent Python code.

It uses the OEIS algorithm to calculate the sequence and then it converts to binary, pads the numbers, does a matrix rotation, and formats it into a pbm image. Since I'm not using brute force, it is incredibly fast.

         K=
 m          rvzQ      Map from eval input to eval input
  .B                  Binary rep
   /      48          Divided by 48
    ++                Triple sum      
     24               Of 24,
     *dd              Square of d
     **               Triple product
      6               6
      %d2             Modulo d%2
      d               Var d
J                     Set J=
 C                    Matrix rotation from columns of row to rows of columns
  m           K       Map K (This does padding)
   +                  String concat
    *                 String repeat
     \0               "0"
     -     ld         Subtract the length of the column from
      eS              The max
       mlkK           Of all the column lengths
    d                 The column
"P1"                  Print header "P1"
l                     Length of
 hJ                   First row
lJ                    Number of columns
jb                    Join by linebreaks
 m  J                 Map on to J
  jb                  Joined columns by linb
   d

Here is the 600,900 example:

Try it here online.