Benchmarks surrounding Intel’s upcoming Knights Landing Xeon Phi cards has managed to court some controversy as Nvidia claims that Intel has published some incorrect “facts” about the performance of the Xeon Phi.
The problem stems from the slide above that was presented by Intel at a high performance computing conference called ISC 2016. According to the slide, Intel claims that the Xeon Phi provides 2.3x faster training for neural networks and 38% better scaling across nodes. Nvidia believes that the chart above is misleading as Intel has supposedly used an old version of the Caffe AlexNet benchmark, resulting in benchmarks that favours Intel. Nvidia further states that if Intel used an up-to-date version of Caffe AlexNet, the Maxwell system would actually perform 30% faster, and for Pascal-based systems, could potentially be 90% faster.
Not surprisingly, Intel believes that they did not do anything wrong. “It is completely understandable that Nvidia is concerned about Intel in this space. Intel routinely publishes performance claims based on publicly available solutions at the time, and we stand by our data,” says Intel in a response to Ars Technica.
Source: Ars Technica
