CASA - Workshop on JUPITER and NVIDIA Grace Hopper

7 Oct 2024, 12:00 → 11 Oct 2024, 16:00 Europe/Berlin

Rotunda (Building 16.4/Room 301) (Jülich Supercompting Centre)

In preparation for the arrival of JUPITER, JSC’s new exascale system, this workshop will help users ready their software for the Big Flop/s!

We introduce the NVIDIA Grace Hopper architecture, profiling, and JSC’s setup for continuous integration and benchmarking.

Interested users should bring their own laptops, ideally have code running on JUWELS, or participate in the JUREAP program.

Zoom-Link for the presentations: We will be streaming the presentations using the zoom-link here.

Monday, 7 October

12:00 Lunch & Check-in

Announcements

Presentations: JUPITER - JSC new Exascale system

GraceHopper.pdf

JUPITER.pdf

Profiling.pdf

Participants: Team intros

15:00 Coffee

Presentations: Grace Hopper: Soft- and Hardware Architecture

GraceHopper.pdf

JUPITER.pdf

Profiling.pdf

Hands-on: Code setup on JEDI

Tuesday, 8 October

Announcements: Starting remarks

Participants: Team plans

Presentations: Profiling on Grace Hopper

GraceHopper.pdf

JUPITER.pdf

Profiling.pdf

Hands-on: Profiling

12:00 Lunch

Hands-on: Profiling & Porting

15:00 Coffee

Hands-on: Profiling & Porting

Wednesday, 9 October

Hands-on: Porting & Testing

Participants: Team plans

Hands-on: Porting & Testing

12:00 Lunch

Hands-on: Porting & Testing

15:00 Coffee

Hands-on: Porting & Testing

Presentations: CPU-GPU System Memory

GraceHopper.pdf

JUPITER.pdf

Profiling.pdf

1 Harnessing Integrated CPU-GPU System Memory for HPC: a first look into Grace Hopper

Abstract:

Memory management across discrete CPU and GPU physical memory is traditionally achieved through explicit GPU allocations and data copy or unified virtual memory. The Grace Hopper Superchip, for the first time, supports an integrated CPU-GPU system page table, hardware-level addressing of system allocated memory, and cache-coherent NVLink-C2C interconnect, bringing an alternative solution for enabling a Unified Memory system. In this presentation, we provide an in-depth analysis of the memory system of the Grace Hopper Superchip. We further detail how this hardware can be leveraged by programmers through various programming interfaces, comparing system-allocated memory and managed memory. We evaluate the performance of a suite of six representative applications, including the Qiskit quantum computing simulator, using system memory and managed memory. Using our memory utilization profiler and hardware counters, we quantify and characterize the impact of the integrated CPU-GPU system page table on GPU applications. Our study focuses on first-touch policy, page table entry initialization, page sizes, and page migration. Our results show that as a new solution for unified memory, the system-allocated memory can benefit most use cases with minimal porting efforts.

The talk will be held in English | Join Zoom Meeting
https://fz-juelich-de.zoom.us/j/63570550943?pwd=S1rwqKaB6ayfMKW8kkaQK3unkK4iKl.1

Speaker: Gabin Schieffer (KTH)

Social event

Thursday, 10 October

Announcements: Starting remarks

Participants: Team plans

Presentations: Continuous Integration and Benchmarking

GraceHopper.pdf

JUPITER.pdf

Profiling.pdf

Hands-on: CI&CB

Photo

12:00 Lunch

Hands-on

JSC Tour: (optional)

We will tour the different systems as JSC

15:00 Coffee

Hands-on: Benchmarking

Friday, 11 October

Announcements: Starting remarks

Hands-on

12:00 Lunch

Hands-on

Participants: Final presentations

Announcements: Workshop close