Presentation on theme: "Avenues for Research The Microarchitecture of Future Microprocessors."— Presentation transcript:
Avenues for Research The Microarchitecture of Future Microprocessors
Some Items Nanotechnology: –rectangles look more like ovals –storage cells have more than two states 3 years ago everyone said: Homogeneous. Today? How do we interconnect cores and accelerators Verification is a problem – Can we design things in? Alpha particles are flipping bits – Can we design in without slowing Dark silicon, the latest “bug” or the latest feature The Cloud – Can the uarch help? How to get inside Approximate computing –Algorithm side: Optimal is not necessary –Analog values: loss of precision is okay Energy to move data vs energy to recompute Effects of a large local store (i.e., registers) Effects of resource sharing
I am still going against the grain Multicore push has been on throughput –I want to push latency, the time it takes to get the job done (cure cancer, predict tsunamis, play games, simulate) Latency –Look at all aspects of the processing –Approximate results may be good enough (Alg + uarch) –uarch: scratch pad, Milad’s mem.ctl.assist, accelerators –Uarch: asynch for n cycles, context switch on chip –Rethink virtual memory mapping: O/S and microarch –Rethink run time system (speed up lagging threads) –Rethink prefetching, branch prediction –Rethink cache partition to enable better concurrency –Break the layers
Your consent to our cookies if you continue to use this website.