Research Paper
DGEMM using Tensor Cores, and Its Accurate and Reproducible Versions
Event Type
Research Paper
TimeTuesday, June 23rd1:50pm - 2:15pm
DescriptionThis paper proposes a method for implementing dense matrix multiplication on FP64 (DGEMM) and FP32 (SGEMM) using Tensor Cores on NVIDIA's graphics processing units (GPUs). Tensor Cores are special processing units that perform 4x4 matrix multiplications on FP16 inputs with FP32 precision, and return the result on FP32. The proposed method adopts the Ozaki scheme, an accurate matrix multiplication algorithm based on error-free transformation for matrix multiplication. The proposed method has three prominent advantages: first, it can be built upon the cublasGemmEx routine using Tensor Core operations; second, it achieves higher accuracy than standard DGEMM, including the correctly-rounded result; third, it ensures bit-level reproducibility even for different numbers of cores and threads. The achievable performance of the method depends on the absolute-value range of each element of the input matrices. For example, when the matrices were initialized with random numbers over a dynamic range of 1E+9, our DGEMM-equivalent implementation achieved up to approximately 3.7 TFlops of FP64 operation on the Titan RTX GPU (with 130 TFlops on Tensor Cores), although the Titan RTX GPU can achieve only 510 GFlops on FP64 floating-point units. Our results reveal the possibility of utilizing hardware with limited FP32/FP64 resources and fast low-precision processing units (such as AI-oriented processors) for general-purpose workloads.