### Table 4. Times for QR factorizations of square matrices. Matrix Cube

1994

"... In PAGE 20: ...1, we ran a sequence of factorizations in which the memory required per processor remained constant, allowing us to compute scaled speedups. The results are presented in Table4 . As before, the e ciencies greater than one are due to longer vectors in the BLAS routines.... ..."

Cited by 60

### Table 4. Times for QR factorizations of square matrices. Matrix Cube

1994

"... In PAGE 20: ...1, we ran a sequence of factorizations in which the memory required per processor remained constant, allowing us to compute scaled speedups. The results are presented in Table4 . As before, the e ciencies greater than one are due to longer vectors in the BLAS routines.... ..."

Cited by 60

### Table 3.7: Results for the QR factorization of square matrices.

### Table 4.2 Single-fault coverage for QR factorization.

### Table 3: Results for QR Factorization on a 17 processor system.

### Table 3: Results for QR Factorization on a 17 processor system.

### Table 2: Performance of out-of-core QR factorization on 64 processors using MB=NB=50.

1997

"... In PAGE 12: ... Note that without this extra reordering cost and assuming perfect speedup from 64 to 256 processors, the out-of-core solver incurs approximately a 18% overhead over in-core solvers ((3502 ? 290)=(681 4) 1:18). Table2 shows the runtime (in seconds) for the out-of-core QR factorization on the Intel Paragon. The eld lwork is the amount of temporary storage (number of double precision numbers) available to the out-of-core routine for panels X and Y.... ..."

Cited by 17

### Table 7 Performance of the QR factorization, SNE and CSNE methods on an iPSC/860 problem p m ops QR SNE C STEP

1994

"... In PAGE 11: ... A running time is obtained by measuring the time spent on each processor and taking the maximum time spent on a processor. The \m ops quot; in Table7 is the execution rate de ned as the number of mega ops performed per second during numerical factorization. The \QR quot; and \SNE quot; in Table 7 represent the running times for the QR factorization method and the SNE method, respectively.... In PAGE 11: ... The \m ops quot; in Table 7 is the execution rate de ned as the number of mega ops performed per second during numerical factorization. The \QR quot; and \SNE quot; in Table7 represent the running times for the QR factorization method and the SNE method, respectively. The \C STEP quot; shows the running time for a correction step in the CSNE method.... ..."

Cited by 3

### Table 2: Performance of out-of-core QR factorization on 64 processors using MB=NB=50.

"... In PAGE 12: ... Note that without this extra reordering cost and assuming perfect speedup from 64 to 256 processors, the out-of-core solver incurs approximately a 18% overhead over in-core solvers ((3502 ? 290)=(681 4) 1:18). Table2 shows the runtime (in seconds) for the out-of-core QR factorization on the Intel Paragon. The eld lwork is the amount of temporary storage (number of double precision numbers) available to the out-of-core routine for panels X and Y.... ..."

### Table 2: Performance of out-of-core QR factorization on 64 processors using MB=NB=50.

1997

"... In PAGE 13: ... Note that without this extra reordering cost and assuming perfect speedup from 64 to 256 processors, the out-of-core solver incurs approximately a 18% overhead over in-core solvers ((3502 ? 290)=(681 4) 1:18). Table2 shows the runtime (in seconds) for the out-of-core QR factorization on the Intel Paragon. The eld lwork is the amount of temporary storage (number of double precision numbers) available to the out-of-core routine for panels X and Y.... ..."