different results with and without SSE ( float arrays multiplication)

I have two functions of 2d arrays multiplication. One of them with SSE. Another function without any optimization. Both functions work well. But the results are slightly different. For example 20.333334 and 20.333332.

Can you explain why the results are different? And what can I do with functions to have the same result?

function with SSE

float** sse_multiplication(float** array1, float** array2, float** arraycheck){    int i, j, k;    float *ms1, *ms2, result;    float *end_loop;    for( i = 0; i < rows1; i++)    {        for( j = 0; j < columns2; j++)        {            result = 0;            ms1 = array1[i];            ms2 = array2[j];            end_loop = &array1[i][columns1];            __asm{                     mov rax, ms1                     mov rbx, ms2                     mov rdx, end_loop                     xorps xmm2, xmm2                loop:                     movups xmm0, [rax]                     movups xmm1, [rbx]                     movups xmm3, [rax+16]                     movups xmm4, [rbx+16]                     mulps xmm0, xmm1                     mulps xmm3, xmm4                     addps xmm2, xmm0                     add rax, 32                     add rbx, 32                     cmp rdx, rax                     jne loop                     haddps xmm2, xmm2                     haddps xmm2, xmm2                     movups result, xmm2               }             arraycheck[i][j] = result;        }    }    return arraycheck;}

function without any optimization

float** multiplication(float** array1, float** array2, float** arraycheck){    for (int i = 0; i < rows1; i++)        for (int j = 0; j < columns2; j++)            for (int k = 0; k < rows1; k++)                arraycheck[i][j] += array1[i][k] * array2[k][j];    return arraycheck;}