1 PFLOP is the peak number of operations per second that can be executed by the GPU. It is derived using NVFP4 with the Sparsity feature enabled. Here’s an example of DGX Spark users seeing 1 PLFOP throughput: GB10 really does hit ~1 PFLOP NVFP4 (2:4 sparse) — measured, with an open-source tool to reproduce it
Neill
4