Qwen3.6-35B vs Gemma4-26B: Real Workload Benchmarks on Radeon 7900 XTX30 May 2026·11 minsBenchmarks Rocm 7900xtx Qwen3.6 Gemma4 Inference Local-Llm