Running Stellar Blade on a Zotac RTX 5070 Ti SOLID CORE and the benchmark is all over the place during boss fights. Do I need to quantify the bottleneck for a report?

For test environment 2026-BENC-08, I ran multiple stress tests on a Zotac NVIDIA RTX 5070 Ti 16GB. Initially, the 3DMark frame time graphs looked like jagged saws, swinging between 14ms and 22ms, making it impossible to get a clean read. Suspecting VRAM scheduling, I went to the NVIDIA Control Panel and set Power Management to 'Prefer Maximum Performance,' then verified the VRAM clock was steady around 2400MHz in GPU-Z. After 5 loops, the standard deviation dropped below 3%, pinning the bottleneck on the CPU's single-core scheduling. Weirdly, if the room temp hits 28℃, the score dips by 5% instantly. Without a beefy radiator, these benchmarks are basically useless in the summer.