7900 XTX + 32/64GB RAM for Qwen 3.8 Flash Next?

A user is planning a PC build primarily for local LLMs/coding agents and is considering a 7900 XTX 24GB GPU due to the difficulty in finding 3090s and the high cost of newer Nvidia GPUs. They are seeking information on the performance of Qwen 3.8 Flash Next on the 7900 XTX, specifically regarding decode and prefill tokens/second, and the quality of AMD/ROCm support. Additionally, they are asking if 32GB system RAM is sufficient or if 64GB is more beneficial, especially if SSD streaming/offloading is necessary for Q4.

This post specifically asks for Qwen 3.8 Flash Next benchmarks on the 7900 XTX, unlike other discussions focusing on the 3090.