Luna — latest tests Open measurements of a local AI assistant on AMD https://luna-orbe-orion.com/ 2026-10-10T00:00:00Z g_lejarshttps://x.com/g_lejars Generation speed, Vulkan vs ROCm, 1k to 30k contexthttps://luna-orbe-orion.com/#gen-2026-10-092026-10-09T00:00:00ZVulkan generates +9% (1k) to +23% (30k) faster on an RX 9070 XT. Prompt reading, Vulkan vs ROCm, long promptshttps://luna-orbe-orion.com/#pp-2026-10-092026-10-09T00:00:00ZROCm reads long prompts 13–20% faster.