Operation: C = A B + C

C – m x n, A – m x k, B – k x n

Platform: Pentium II 233 MHz Dell Inspiron 3200 laptop

We report performance for m = 64:64:512, n = 64:64:512, k = 64:64:512

Figure 1: Mflop/sec attained for indicated matrix dimensions by current version of TXBLAS matrix-matrix multiply.

Figure 2: Mflop/sec attained for indicated matrix dimensions by ATLAS Release 2.0 matrix-matrix multiply.

Figure 3: Difference in Mflop/sec attained: TXBLAS – ATLAS reporting only those cases where TXBLAS outperforms ATLAS.

Figure 4: Difference in Mflop/sec attained: ATLAS - TXBLAS reporting only those cases where ATLAS outperforms TXBLAS.

Figure 5: Percent of cases measured where implementation outperforms the Mflop/sec rate indicated on the x-axis.

Figure 6: Percent of cases measured where implementation outperforms the indicated percent of peak, where peak is taken to equal 233 Mflop/sec.