DGEMM on Integer Tensor Cores - NHR Perflab Seminar

Explore a cutting-edge seminar talk on double-precision matrix multiplication using Int8 Tensor Cores and the Ozaki scheme. Delve into the world of specialized computing units for matrix multiplication, focusing on their applications in deep learning and machine learning inference. Learn about the development of processors with fast integer matrix multiplication units and their significance in meeting the increasing demand for dense matrix-matrix multiplication. Discover how low-precision operations and fixed-point value computation are utilized in deep learning, and understand the importance of efficient matrix multiplication in this context. Gain insights from speaker Hiroyuki Ootomo, a Ph.D. candidate at Tokyo Institute of Technology, as he presents his research on high-precision matrix multiplication using lower-precision computing units. Examine the potential applications of this technology in high performance computing, mixed-precision computing, randomized numerical linear algebra, and quantum circuit simulation.