My ASUS TUF throughput is all over the place during benchmarks. Should I export a baseline?

I tested two different paths. Path one was just messing with queue depth using CrystalDiskMark set to Queue 32; speed went up, but the scores were still jumping around. Path two was a deep dive into the BIOS where I enabled Re-Size BAR (Above 4G Decoding). This dropped the storage response latency significantly, and the throughput curve finally smoothed out. I then used 3DMark's stress test to confirm that the conflict between the CPU scheduler and storage was gone. After a final check with Cinebench, the scores were consistent. Path two is the only way to go; software tweaks are useless if the underlying PCIe communication is bottlenecked. It's a night and day difference.