The Gauss-Seidel Method requires the matrix to be in diagonally dominant form.
This matrix is not diagonally dominant and G-S does not converge (sometimes it still may).
The first step is to put the matrix in D-D form so we have $Ax = b$ as
$$A = \begin{pmatrix}200 & -3 & 2 \\
1 & -500 & 2 \\
1 & -3 & 100 \\
\end{pmatrix}, b = \begin{pmatrix}765\\
987 \\
123 \\
\end{pmatrix}$$
You can figure out how many iterations you need to meet your accuracy requirement
$$\begin{align}x_0 &= (3.5,-2,1)\\
x_1 &= (3.785,-1.96243,1.1332771)\\
x_2 &= (3.784230779,-1.96189843,1.133300739)\\
x_3 &= (3.784238516,-1.96189832,1.133300665)\\
x_4 &= (3.784238519,-1.96189832,1.133300665)\end{align}$$
The exact results are
$$x = \begin{pmatrix}
\dfrac{900785}{238036} \\
-\dfrac{3269017}{1666252} \\
\dfrac{3776729}{3332504} \\
\end{pmatrix} \approx
\begin{pmatrix}
3.784238518543414\\-1.961898320302091\\1.133300665205503 \end{pmatrix}$$