Entertainment at it's peak. The news is by your side.

Piuma: Programmable Integrated Unified Memory Architecture


[Submitted on 13 Oct 2020]

Obtain PDF

Abstract: High efficiency natty scale graph analytics is necessary to effectively timed analyze
relationships in mountainous data sets. Primitive processor architectures undergo
from inefficient handy resource utilization and contaminated scaling on graph workloads. To enable
ambiance correct and scalable graph prognosis, Intel developed the Programmable
Constructed-in Unified Memory Architecture (PIUMA). PIUMA consists of many
multi-threaded cores, beautiful-grained memory and network accesses, a globally
shared address home and sturdy offload engines. This paper gifts the
PIUMA architecture, and affords initial efficiency estimations, projecting
that a PIUMA node will outperform a outmoded compute node by one to two
orders of magnitude. Furthermore, PIUMA continues to scale all over more than one
nodes, which is a recount in outmoded multinode setups.

Submission history

From: Stijn Eyerman [view email]

Tue, 13 Oct 2020 10: 35: 52 UTC (206 KB)

Read More

Leave A Reply

Your email address will not be published.