Skip to content

Commit 125140f

Browse files
dohyun-csedohyun-cse
committed
spencer's talk
1 parent df6f0da commit 125140f

File tree

2 files changed

+93
-0
lines changed

2 files changed

+93
-0
lines changed

‎_news/seminar_2.md‎

Lines changed: 21 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,21 @@
1+
---
2+
layout: post
3+
date: 2025-10-29 11:00:00-0500
4+
inline: true
5+
related_posts: false
6+
---
7+
8+
#### _Shocks without tracking or capturing: Exceeding 1 quadrillion degrees of freedom via inviscid regularization of the compressible Navier-Stokes_
9+
10+
- **Speaker**: [Spencer Bryngelson](https://comp-physics.group/) (Georgia Institute of Technology)
11+
- **Date**: October 29, 2025
12+
- **Time**: 11:00 -- 12:00 (EST)
13+
14+
- **Abstract**:
15+
A method for solving multi-species and shock-laden flow at unprecedented problem sizes and time-to-solution is presented. The first inviscid regularization of the Navier-Stokes-like PDE is performed. This enables linear and well-conditioned numerics suitable for mixed-precision computation. A unified memory implementation is crafted for tightly coupled CPU-GPU and APU architectures (e.g., MI250X, GH200, MI300A), now standard on flagship machines like El Capitan and Frontier. With this trio, we improve on state-of-the-art CFD techniques with order of magnitude improvements along computational cost and memory footprint axes. The reduced memory footprint compared to baselines enables, for example, 25-times larger simulations, here shown to exceed 1 quadrillion degrees of freedom (200T grid points) with per-grid-point cost speedups. The method strong scales from 8 nodes to the full systems (>10K nodes) with better than 50% efficiency. This enables, for example, a typical 200B grid point CFD simulation in less than one wall clock minute on OLCF Frontier. Early results suggest increased robustness compared to ENO/limiter-type shock capturing schemes for high-Mach flows and strong discontinuities. Results are shown for a Mach 14 many-rocket-engine configuration that nominally matches the SpaceX Super Heavy.
16+
17+
- **Location**:
18+
Virtual (Zoom)
19+
20+
- **Zoom**: [Zoom Link](https://brown.zoom.us/j/92988066475?pwd=xhn5ibHydwlXsnGdLyHGWwXsBUGaac.1)
21+
- [**Add to Calendar**](/assets/calendar/seminar_2.ics)

‎assets/calendar/seminar_2.ics‎

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,72 @@
1+
BEGIN:VCALENDAR
2+
CALSCALE:GREGORIAN
3+
VERSION:2.0
4+
X-WR-CALNAME:CIGMO Seminar - Spencer Bryngelson
5+
METHOD:PUBLISH
6+
PRODID:-//Apple Inc.//macOS 15.6.1//EN
7+
BEGIN:VTIMEZONE
8+
TZID:America/New_York
9+
BEGIN:DAYLIGHT
10+
TZOFFSETFROM:-0500
11+
DTSTART:20070311T020000
12+
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
13+
TZNAME:EDT
14+
TZOFFSETTO:-0400
15+
END:DAYLIGHT
16+
BEGIN:STANDARD
17+
TZOFFSETFROM:-0400
18+
DTSTART:20071104T020000
19+
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
20+
TZNAME:EST
21+
TZOFFSETTO:-0500
22+
END:STANDARD
23+
END:VTIMEZONE
24+
BEGIN:VEVENT
25+
TRANSP:OPAQUE
26+
DTEND;TZID=America/New_York:20251029T120000
27+
UID:30173891-0693-4C33-9090-BBEE62C42F52
28+
DTSTAMP:20251015T150739Z
29+
LOCATION:https://brown.zoom.us/j/9011251684?omn=93423404156
30+
DESCRIPTION:CIGMO Seminar Series - 2\n\nTitle: Shocks without tracking o
31+
r capturing: Exceeding 1 quadrillion degrees of freedom via inviscid reg
32+
ularization of the compressible Navier-Stokes\nSpeaker: Spencer Bryngels
33+
on\n\nAbstract:\nA method for solving multi-species and shock-laden flow
34+
at unprecedented problem sizes and time-to-solution is presented. The f
35+
irst inviscid regularization of the Navier-Stokes-like PDE is performed.
36+
This enables linear and well-conditioned numerics suitable for mixed-pr
37+
ecision computation. A unified memory implementation is crafted for tigh
38+
tly coupled CPU-GPU and APU architectures (e.g.\, MI250X\, GH200\, MI300
39+
A)\, now standard on flagship machines like El Capitan and Frontier. Wit
40+
h this trio\, we improve on state-of-the-art CFD techniques with order o
41+
f magnitude improvements along computational cost and memory footprint a
42+
xes. The reduced memory footprint compared to baselines enables\, for ex
43+
ample\, 25-times larger simulations\, here shown to exceed 1 quadrillion
44+
degrees of freedom (200T grid points) with per-grid-point cost speedups
45+
. The method strong scales from 8 nodes to the full systems (>10K nodes)
46+
with better than 50% efficiency. This enables\, for example\, a typical
47+
200B grid point CFD simulation in less than one wall clock minute on OL
48+
CF Frontier. Early results suggest increased robustness compared to ENO/
49+
limiter-type shock capturing schemes for high-Mach flows and strong disc
50+
ontinuities. Results are shown for a Mach 14 many-rocket-engine configur
51+
ation that nominally matches the SpaceX Super Heavy.
52+
X-APPLE-CREATOR-IDENTITY:com.apple.calendar
53+
STATUS:CONFIRMED
54+
SEQUENCE:1
55+
SUMMARY:CIGMO Seminar - Spencer Bryngelson
56+
DTSTART;TZID=America/New_York:20251029T110000
57+
LAST-MODIFIED:20251015T150416Z
58+
CREATED:20251015T150415Z
59+
X-APPLE-CREATOR-TEAM-IDENTITY:0000000000
60+
BEGIN:VALARM
61+
UID:560E4484-7AA1-41B8-9E3D-64C596CBB1C2
62+
X-WR-ALARMUID:560E4484-7AA1-41B8-9E3D-64C596CBB1C2
63+
TRIGGER:-PT10M
64+
DESCRIPTION:This is an event reminder
65+
ACTION:DISPLAY
66+
END:VALARM
67+
BEGIN:VALARM
68+
TRIGGER;VALUE=DATE-TIME:19760401T005545Z
69+
ACTION:NONE
70+
END:VALARM
71+
END:VEVENT
72+
END:VCALENDAR

0 commit comments

Comments
 (0)