Retail prices for AMD RDNA 4 graphics cards have surged by $200 to $300 across major markets. A Sapphire Pulse Radeon RX 9060 XT 16GB owner managed to purchase the card for $575 right before listings jumped past $750, while higher-end Radeon RX 9070 XT models have already reached $1000.
The graphics card is utilized exclusively for running local large language models rather than gaming. Despite the 16GB VRAM limit, the setup efficiently runs the Qwen3.8-27B model in IQ3_XXS quantization with MTP enabled and a quantized KV cache of 100k tokens, alongside MoE architectures backed by system DDR4 memory.
Testing within the llama.cpp framework revealed that the Vulkan backend delivers superior performance compared to AMD ROCm on the Navi 44 GPU with 2048 stream processors. By pairing the hardware with the lightweight Pi harness, users can deploy a fully functional local AI assistant.