A CPU architecture is proposed which flexibly allocates chip resources among threads. Execution units (microcores) are arranged in a ring. Instruction fetch units (front-ends) deposit instructions sequentially into storage elements within the microcores. Multiple front-ends can each feed segments of the ring; each such segment is a “smart queue”. If, due to a sustained higher execution rate, a thread catches up to the next thread ahead of it, the slower thread steps aside and lets the faster thread play through. Other circumstances may lead to a thread consuming more than its usual share of resources, possibly even all of the microcores, for a time. The architecture has no instruction set dependencies; it is applicable to existing instruction set architectures and will speed up execution of them significantly as compared to conventional architectures.