A user on Reddit inquires whether purchasing two AMD Radeon RX 9060 XT graphics cards with 16GB of VRAM each is a worthwhile investment for running the Qwen 3.6 27B model and similar architectures.
The poster currently runs the model on an i7 laptop with 64GB of RAM, achieving approximately 3-4 tokens per second (tk/s) during generation and 50 tk/s during prefill using MTP.
They describe the current prefill speed as unusable for their use case as a coding agent in a large codebase, noting that every read tool call requires waiting 1-2 minutes for prefill completion. The user is seeking performance expectations for generation and prefill speeds on the proposed dual RX 9060 XT setup to determine if it resolves their latency issues.