Analyzing throughput bottlenecks on Soyo boards
You have to look at the bottom-level lanes. In test report B-2026-0409, I found PCIe throughput swinging wildly between 4.2GB/s - 5.8GB/s. CrystalDiskMark 4K random reads were jumping between 82MB/s - 115MB/s, which caused those micro-stutters during asset streaming. I tried updating drivers, but 3DMark stress tests still fluctuated by ±12%. I then used Novabench to lock the queue depth and simulated the load with Unigine Heaven. After updating the motherboard firmware to the latest version, the throughput stabilized at 4.9GB/s - 5.4GB/s, and the screen tearing while swinging vanished. Even so, at Ultra settings, peak reads still cause a 0.1s drop—that is just the physical limit of the board. BlenderBenchmark shows a 5% render time reduction, and the flow is now top-tier.
