A Performance Study of a Dual Xeon-Phi Cluster for the Forward Modelling of Gravitational Fields

<div>Behaviour of the computing time obtained for the estimation of <svg height="12.0184pt" id="M120" style="vertical-align:-3.382501pt" version="1.1" viewbox="-0.0498162 -8.6359 18.6833 12.0184" width="18.6833pt" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink"><g transform="matrix(.013,0,0,-0.013,0,0)"><path d="M673 297H417L411 270C517 261 522 258 506 183L487 92C476 38 428 19 360 19C207 19 123 132 123 287C123 472 243 631 441 631C557 631 617 583 617 469L647 472C650 545 655 604 658 632C625 643 554 666 467 666C201 666 23 502 23 278C23 90 156 -16 339 -16C410 -16 501 6 569 24C566 43 567 70 573 101L589 183C604 259 607 262 667 271L673 297Z" id="g113-72"></path><glyph.data ascent="3473" descent="-2876" horiz-adv-x="684" vert-adv-y="684"></glyph.data></g><g transform="matrix(.0091,0,0,-0.0091,8.827,3.132)"><path d="M481 445L456 468C426 437 403 427 388 427C372 427 356 428 302 438C256 447 221 451 187 451C159 451 130 428 105 384C93 363 80 340 65 305L96 290C122 342 149 381 193 381C216 381 267 377 294 374S346 372 365 376L75 105C37 70 24 35 24 16C24 2 30 -5 35 -8C42 -8 46 -6 48 -2C58 14 75 35 90 41C102 46 116 46 147 32C200 8 267 -21 298 -21C347 -21 390 12 448 126L420 151C369 77 338 56 311 56C287 56 249 68 212 81C180 92 149 100 127 98C212 172 353 308 413 372L481 445Z" id="g50-123"></path><glyph.data ascent="3443" descent="-2856" horiz-adv-x="505" vert-adv-y="505"></glyph.data></g><g transform="matrix(.0091,0,0,-0.0091,13.423,3.132)"><path d="M481 445L456 468C426 437 403 427 388 427C372 427 356 428 302 438C256 447 221 451 187 451C159 451 130 428 105 384C93 363 80 340 65 305L96 290C122 342 149 381 193 381C216 381 267 377 294 374S346 372 365 376L75 105C37 70 24 35 24 16C24 2 30 -5 35 -8C42 -8 46 -6 48 -2C58 14 75 35 90 41C102 46 116 46 147 32C200 8 267 -21 298 -21C347 -21 390 12 448 126L420 151C369 77 338 56 311 56C287 56 249 68 212 81C180 92 149 100 127 98C212 172 353 308 413 372L481 445Z" id="g50-123"></path><glyph.data ascent="3443" descent="-2856" horiz-adv-x="505" vert-adv-y="505"></glyph.data></g></svg> when using 10 nodes of the cluster (the scale is semilogarithmic base 10 on the time axis). The curves compare the behaviour of the times obtained using 16 threads on the CPU versus 236 threads for each Xeon-Phi (these are the best results for both cases). As expected, the behaviour is not completely linear (but almost) as we can see when the speed-up is calculated (see Figure <a href="../fig17/">17</a>).</div>

Scientific Programming

fig16

Figure 16

Figure 16: A Performance Study of a Dual Xeon-Phi Cluster for the Forward Modelling of Gravitational Fields 