https://community.intel.com/t5/Intel-Embree-Ray-Tracing-Kernels/knl/m-p/1113766#M626 <P>Hi,</P> <P>Providing a single mrays/s number here is quite difficult as the ray tracing performance depends on so many factors, e.g. scene complexity, primitive distribution in the scene, coherent/incoherent ray distribution, whether single, packet, or ray streams are used, closest-hit or any-hit rays, SIMD utilization during shading and to generate rays, HW core/thread count, sustained core turbo frequency, work distribution among the threads etc. It might be better to benchmark the entire rendering application to get a meaningful number.&nbsp; Assuming (for the sake of simplicity) that the renderer is equally efficient per clock on both a Xeon and Xeon Phi system and the work distribution scheme achieves linear scaling than the relative performance can be very roughly compared by: #cores * sustained core frequency between the two platforms.</P> <P>&nbsp;</P> <P>&nbsp;</P> <P>&nbsp;</P> Mon, 04 Jul 2016 14:25:03 GMT BenthinC_Intel 2016-07-04T14:25:03Z https://community.intel.com/t5/Intel-Embree-Ray-Tracing-Kernels/knl/m-p/1113765#M625 <P>hallo, i have seen support for knl in embree. how is the knl</P> <P>performance compared to std. server cpus like haswell or broadwell ?</P> <P>what are the embree benchmark numbers in terms of mrays/s ?</P> Thu, 30 Jun 2016 08:36:02 GMT https://community.intel.com/t5/Intel-Embree-Ray-Tracing-Kernels/knl/m-p/1113765#M625 f__n_ 2016-06-30T08:36:02Z https://community.intel.com/t5/Intel-Embree-Ray-Tracing-Kernels/knl/m-p/1113766#M626 <P>Hi,</P> <P>Providing a single mrays/s number here is quite difficult as the ray tracing performance depends on so many factors, e.g. scene complexity, primitive distribution in the scene, coherent/incoherent ray distribution, whether single, packet, or ray streams are used, closest-hit or any-hit rays, SIMD utilization during shading and to generate rays, HW core/thread count, sustained core turbo frequency, work distribution among the threads etc. It might be better to benchmark the entire rendering application to get a meaningful number.&nbsp; Assuming (for the sake of simplicity) that the renderer is equally efficient per clock on both a Xeon and Xeon Phi system and the work distribution scheme achieves linear scaling than the relative performance can be very roughly compared by: #cores * sustained core frequency between the two platforms.</P> <P>&nbsp;</P> <P>&nbsp;</P> <P>&nbsp;</P> Mon, 04 Jul 2016 14:25:03 GMT https://community.intel.com/t5/Intel-Embree-Ray-Tracing-Kernels/knl/m-p/1113766#M626 BenthinC_Intel 2016-07-04T14:25:03Z