An Nvidia Tech Brief reports on speeding-up SLB’s Intersect fluid-flow reservoir modeler using the latest A100 80 GB Tensor Core GPU. The test compared runs of SLB’s Intersect simulator on single-node servers powered by Tensor Core GPU processors against ‘servers powered by CPU processors’ (although no information was provided on the CPUs or how they were used). A 2.3 million cell black oil model showed a 4.5x speedup on a single simulation. This increased to 5.7x throughput when leveraging the Tensor’s multi-instance GPU (MIG) function. A further speedup (to 7.5x) was achieved using the Tensor’s Multi-Process Service (MPS) capability. MPS, is described as an ‘alternative, binary-compatible implementation’ of Nvidia’s CUDA API that enables CUDA kernels to be processed concurrently on the same GPU.
More from Nvidia Energy and the A100 datasheet.
© Oil IT Journal - all rights reserved.