Why measuring performance is becoming so difficult.
Raw performance used to be the main focus of benchmarks, but they may have outlived their usefulness for many applications. Dana McCarty, vice president of sales and marketing for AI Inference Products at Flex Logix, talks about why companies need to develop and utilize their own specific models to accurately gauge hardware and software performance, which can be slowed by bottlenecks in I/O and across different convolution layers.
 |
 |
 |
 |
 |
 |
 |
Leave a Reply