This shows you the differences between two versions of the page.

Link to this comparison view

seminars:seminar_4_19_17 [2017/09/20 22:02] (current)
Line 1: Line 1:
 +====== LazyPIM: Efficient Support for Cache Coherence in Processing-in-Memory Architectures ======
 +Wednesday April 19, 2017\\
 +Location: CIC Panther Hollow Room\\
 +Time: 4:30PM\\
 +**Amirali Boroumand (CMU)**\\
 +Processing-in-memory (PIM) architectures have seen an increase
 +in popularity recently, as the high internal bandwidth
 +available within 3D-stacked memory provides greater incentive
 +to move some computation into the logic layer of the
 +memory. To maintain program correctness,​ the portions of
 +a program that are executed in memory must remain coherent
 +with the portions of the program that continue to execute within
 +the processor. Unfortunately,​ PIM architectures cannot use
 +traditional approaches to cache coherence due to the high offchip
 +traffic consumed by coherence messages, which, as we
 +illustrate in thiswork, can undo the benefits of PIM execution
 +for many data-intensive applications.
 +We propose¬† LazyPIM, a new hardware cache coherence
 +mechanism designed specifically for PIM. Prior approaches
 +for coherence in PIM are ill-suited to applications that share a
 +large amount of data between the processor and the PIM logic.
 +LazyPIM uses a combination of speculative cache coherence
 +and compressed coherence signatures to greatly reduce the
 +overhead of keeping PIM coherent with the processor, even
 +when a large amount of sharing exists.We find that LazyPIM
 +improves average performance across a range of data-intensive
 +PIM applications by 19.6%, reduces off-chip traffic by 30.9%,
 +and reduces energy consumption by 18.0%, over the best prior
 +approaches to PIM coherence.
 +Amirali Boroumand is a PhD student advised by Onur Mutlu. His research is focused on Processing in Memory.
 +**[[seminars| Back to the seminar page]]**