G in case of irregular memory access patterns

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: es "   y intersect: multiplicative factor "   slope: exponent à༎ linear: 450 slope Typical Roofline Plot Low Operational Intensity: - very few Flops per byte - memory bandwidth is limiting factor - linear slope behavior min(32n,2048) G F l o p s / s e c 2048 1024 512 256 128 64 32 16 8 4 2 1 min(32n,2048) 0.5 2 8 32 128 512 2048 Operational intensity (GFlops/byte) 8192 High Operational Intensity: - many Flops per byte - machine peak Flop rate is limiting factor - constant performance ridge point, where slope meets horizontal: minimum operational intensity to get maximal performance Example: Opteron X2 vs. Opteron X4 "   Both in same socket, so same memory behavior 64 "   X4: 4x higher Gflops rate 32 double # cores...
View Full Document

Ask a homework question - tutors are online