About
Intel systems featuring Optane DC DIMMs offer a vast amount of memory at low TCO. Without further aid, having more than one memory subsystem leaves application developers with the responsibility of deciding where to allocate their different memory objects, which is far from an easy task and may pose severe performance implications. The hardware-assisted 2LM mode palliates this situation, exposing a single memory address space and performing automatic data movements at page-level granularity. In this session we present a software-based placement solution which outperforms the 2LM mode by leveraging the semantics of memory objects and exploiting previously captured access patterns. We aim at moving a big step forward and developing technology to build an innovative generic software ecosystem to facilitate the efficient use of heterogeneous memory systems, what will be crucial to leverage the full potential of exascale platforms for compelling HPC and AI workloads..
0:00:00 Slide 1
0:00:08 Slide 2
0:00:20 Slide 3
0:00:35 Slide 4
0:01:21 Slide 5
0:02:33 Slide 6
0:03:35 Slide 7
0:03:53 Slide 8
0:04:10 Slide 9
0:04:59 Slide 10
0:05:20 Slide 11
The talk
Credits and licence
- Institution
- CERN
- Department
- other events or meetings
- Speaker
- Marc Jorda
- Year
- 2019
- Language
- English
- Source
- CERN
- Licence
- CERN, as CERN states it. CERN's terms
- Attribution
- © CERN. CERN provides its audiovisual media free of charge for educational and informational use, credited to CERN; this lecture is shown in CERN's own player.
- Who can watch
- CERN gives no rating. Invisible Network shows it to Teens profiles and older, judged from what the source says about it. How courses are placed