Skip to content
Snippets Groups Projects
Commit e13f0537 authored by Johannes Wasmer's avatar Johannes Wasmer
Browse files

update

parent 8f76f8a2
No related branches found
No related tags found
No related merge requests found
Hide whitespace changes
Inline Side-by-side
notes/hdslee-retreat-talk-notes.org
+ 125
− 14
View file @ e13f0537
...@@ -18,9 +18,12 @@ ...@@ -18,9 +18,12 @@
- [[#event-info][Event info]] - [[#event-info][Event info]]
- [[#workflow][Workflow]] - [[#workflow][Workflow]]
- [[#slides-templates][Slides templates]] - [[#slides-templates][Slides templates]]
- [[#checklists-00][Checklists]] - [[#checklists-07][Checklists]]
- [[#slides][Slides]] - [[#slides][Slides]]
- [[#journal][Journal]] - [[#journal][Journal]]
- [[#initial-draft][Initial draft]]
- [[#research][Research]]
- [[#energy-consumption][Energy consumption]]
* Project management * Project management
:PROPERTIES: :PROPERTIES:
...@@ -100,28 +103,136 @@ There is a 2023 slide template for PowerPoint in the HDSLEE sciebo folder. ...@@ -100,28 +103,136 @@ There is a 2023 slide template for PowerPoint in the HDSLEE sciebo folder.
Probably, the LaTeX version should be updated to that. Probably, the LaTeX version should be updated to that.
- Retreat 2024 slides template ppt. [[https://fz-juelich.sciebo.de/apps/files/?dir=/HDS-LEE%20PhD%20Students/Presentation%20slides%20of%20the%20events/Retreat%202024/Talks&fileid=1451917185][URL]]. - Retreat 2024 slides template ppt. [[https://fz-juelich.sciebo.de/apps/files/?dir=/HDS-LEE%20PhD%20Students/Presentation%20slides%20of%20the%20events/Retreat%202024/Talks&fileid=1451917185][URL]].
** Checklists [0/0] ** Checklists [0/7]
:PROPERTIES: :PROPERTIES:
:CUSTOM_ID: h-2E64B36D-89F7-489B-BBFA-37AE2EC0B74E :CUSTOM_ID: h-2E64B36D-89F7-489B-BBFA-37AE2EC0B74E
:END: :END:
- [ ] Section Frontmatter
- [ ] titlepage
- [ ] slides repo
- [ ] outline
- [ ] Section Introduction
- [ ] the most important problem
- [ ] Corporate research AI4Science
- [ ] WFT and DFT
- [ ] 5th paradigm
- [ ] Quantum materials
- [ ] JuDFT
- [ ] AiiDA
- [ ] The full pipeline
- [ ] Section Machine learning exchange interactions
- [ ] AiiDA workflows
- [ ] results Jijs
- [ ] model selection
- [ ] results SOAP
- [ ] Section Machine learning all-electron potential
- [ ] JLCDM
- [ ] KKR-JLCDM
- [ ] plots
- [ ] Section Deep Spin Models
- [ ] Idea
- [ ] Section Community
- [ ] iffAiiDA
- [ ] best-of-aml
- [ ] funding
- [ ] Section Discussion
* Slides * Slides
:PROPERTIES: :PROPERTIES:
:CUSTOM_ID: h-09980DCF-1F56-492B-BED2-FF33078B4927 :CUSTOM_ID: h-09980DCF-1F56-492B-BED2-FF33078B4927
:END: :END:
General ideas of slide arrangement per section. General ideas of slide arrangement per section.
| No | Section | Title | Intent |
| No | Section | Title | Intent | |----+------------------+-------------------------------+--------------------------------|
|----+--------------+-------------+--------| | -2 | frontmatter | titlepage | |
| -2 | frontmatter | titlepage | | | -1 | frontmatter | slides repo | |
| -1 | frontmatter | slides repo | | | 0 | frontmatter | outline | |
| 0 | frontmatter | outline | | |----+------------------+-------------------------------+--------------------------------|
|----+--------------+-------------+--------| | 1 | introduction | the most important problem | personal motivation |
| | introduction | | | | 2 | introduction | Corporate research AI4Science | business motivation |
|----+--------------+-------------+--------| | 3 | introduction | WFT and DFT | definition of first principles |
| | | | | | 4 | introduction | 5th paradigm | perfectly labeled data |
|----+--------------+-------------+--------| | 5 | introduction | Quantum materials | spintronics NC, TQC |
| | discussion | | | | 6 | introduction | JuDFT | all-electron DFT |
| 7 | introduction | AiiDA | workflows |
| 8 | introduction | The full pipeline | PhD project |
|----+------------------+-------------------------------+--------------------------------|
| 9 | jij-prediction | AiiDA workflows | |
| 10 | jij-prediction | results Jijs | |
| 11 | jij-prediction | model selection | |
| 12 | jij-prediction | results SOAP | |
|----+------------------+-------------------------------+--------------------------------|
| 13 | vimp-prediction | JLCDM | |
| 14 | vimp-prediction | KKR-JLCDM | |
| 15 | vimp-prediction | plots | |
|----+------------------+-------------------------------+--------------------------------|
| 16 | depp-spin-models | Idea | |
|----+------------------+-------------------------------+--------------------------------|
| 17 | community | iffAiiDA | |
| 18 | community | best-of-aml | |
| 19 | community | funding | |
|----+------------------+-------------------------------+--------------------------------|
| | discussion | | |
* Journal * Journal
:PROPERTIES: :PROPERTIES:
:CUSTOM_ID: h-B08CDFB2-F592-4DE9-8F46-01D6C52E9514 :CUSTOM_ID: h-B08CDFB2-F592-4DE9-8F46-01D6C52E9514
:END: :END:
** Initial draft
:PROPERTIES:
:CUSTOM_ID: h-DB34FB5C-6C27-490A-A002-4614CDEB4AAC
:END:
* Research
:PROPERTIES:
:CUSTOM_ID: h-A89373D7-965C-4485-BDFE-1AEA3EA45E42
:END:
** Energy consumption
:PROPERTIES:
:CUSTOM_ID: h-9208995B-3865-49D3-AF61-34EA2DB32517
:END:
- How much electricity (kWh and cost) is used per FLOP of computation?
- needed to interpret "epoch-ml-trends" training compute (FLOP) vs publicaton
date (year).
- Result from below to use in talk: example gemini ultra 5e25 FLOP training
operations, 30 mio USD training electricity cost, 15 MW power use.
- references
- ai models FLOPs and cost consumption
- https://epochai.org/data/epochdb/visualization
- https://ourworldindata.org/grapher/artificial-intelligence-training-computation
- https://ourworldindata.org/brief-history-of-ai
- https://ourworldindata.org/grapher/hardware-and-energy-cost-to-train-notable-ai-systems
- energy efficiency
- https://en.wikipedia.org/wiki/Performance_per_watt#FLOPS_per_watt
- https://top500.org/lists/green500/2024/06/
- https://en.wikipedia.org/wiki/Cost_of_electricity_by_source#Capital_costs
- https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Electricity_price_statistics
- https://en.wikipedia.org/wiki/European_countries_by_electricity_consumption_per_person
- From above, can get a rough estimate of electricity cost as 2 mio. USD per 1
MW of power for one year, or equivalently, 20 Cent per kWh. Conversion
between the two: 1 year has 1e4 h, so 1 MW * 1y = 1 MW * 1e4 h = 1e4 MWh =
1e7 kWh. Then, 2e6 / 1e7 USD/kWh = ca. 20 Cent/kWh. Formula: C = M/(P*t) =
M/E = 2e-1 USD/kWh for assumption 1 EUR = 1 USD.
- How much MW to power a city? Assume 7e3 kWh /d/cap for Germany. That's
- Based on the references, current data suggests that energy efficiency 50
GFlops / Watt = 5e10 FLOP/Watt = 5e16 FLOP/MW for HPC systems is a good bet.
That's EFF=O/P (operations per power). we want e=energy. we have p=e/t, so
e=p*t. So, eff=o/p -> p*eff=o -> e/t*eff=o -> E=O*T/EFF or P=O/EFF.
- Based on reference, largest consumption in FLOP for notable AI models
training is around o=1e26 FLOP (epochai), 1e25 (ourworldindata).
- Based on reference, largest consumption in USD for notable AI models
training is around 100 million (ourworldindata).
- example training stats from epochai:
| Model | Publication | O | T | M | E | P | P2 | E2 |
|------------------+-------------+--------+-------+--------+-------+-------+-----+--------|
| Gemini 1.0 Ultra | 2023 | 5.0e25 | 1.0e2 | 30.0e6 | 1.5e8 | 1.5e1 | 1e9 | 2.4e15 |
#+TBLFM: $6=$5/0.2;s3::$7=$6*(1e3/(1e4*1e6));s3::$8=$3/5e16;s3::$9=$3*($4*24)/(1e3 * 5e10)
- Table legend.
- O = training compute (FLOP)
- T = training time (d)
- M = training electricity cost (2023 USD)
- E = energy electrity use (kWh). Formula from above: E = M/C, C=0.2
USD/kWh. Answer 150 GWh.
- P = power use (MW). Formula: P/y = E*t, conversion d/y*kWh/y -> W =
*1e3/(1e4*1e6) (khW, h/y, MW). Answer 15 MW. This is what we expected.
- P2 = power use (MW). With formula from above P=O/EFF. Answer 1e9 MW. This is wrong.
- E2 = energy use (kW), based on formula 2: E=O*T/EFF, conversion *24/1e3
(d, W->kW). Answer 2.4e15 kWh. This is wrong.
presentation/config/header.tex
+ 2
− 2
View file @ e13f0537
...@@ -253,8 +253,8 @@ aspectratio=169,% common values: 169, 1610, 43 ...@@ -253,8 +253,8 @@ aspectratio=169,% common values: 169, 1610, 43
\sisetup{output-exponent-marker=\ensuremath{\mathrm{e}}} % print 1e6 as 1e6, not \sisetup{output-exponent-marker=\ensuremath{\mathrm{e}}} % print 1e6 as 1e6, not
% 1 x 10^6 % 1 x 10^6
\usepackage{physics} % \usepackage{physics}
% \usepackage{braket} % Dirac notation \usepackage{braket} % Dirac notation
% commented out package braket, because ceriotti's cosmo/dirac-rep.tex for % commented out package braket, because ceriotti's cosmo/dirac-rep.tex for
% structural representations uses the physics package whose dirac not. syntax % structural representations uses the physics package whose dirac not. syntax
......
presentation/mainmatter/introduction.tex 0 → 100644
+ 277
− 0
View file @ e13f0537
\section{The most important problem}
\label{sec:intro}
\begin{frame}[c,plain]
% \frametitle{}
\begin{center}
{\Huge \textcolor{fzjblue}{What is the most important problem right
now?}}
\end{center}
\end{frame}
% Slide The most important problem 1 The Great Acceleration
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-1.pdf}
\end{frame}
% Slide Anthropocene 1
\begin{frame}%[plain,c]
\frametitle{Socio-economic trends\footcite{steffenTrajectoryAnthropoceneGreat2015}}
% \framesubtitle{}
\vspace*{0em}
\begin{center}
\includegraphics[width=0.9\linewidth]{../resources/fig/external/papers/steffenTrajectoryAnthropoceneGreat2015/figure1-subselection-2x3-a.pdf}%
\end{center}
\vspace*{0em}
\end{frame}
% Slide Anthropocene 2
\begin{frame}%[plain,c]
\frametitle{Earth system trends\footcite{steffenTrajectoryAnthropoceneGreat2015}}
% \framesubtitle{}
\vspace*{0em}
\begin{center}
\includegraphics[width=0.9\linewidth]{../resources/fig/external/papers/steffenTrajectoryAnthropoceneGreat2015/figure3-subselection-2x3-a.pdf}%
\end{center}
\vspace*{0em}
\end{frame}
% Slide The most important problem 2 Changes Earth's Atmosphere
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-2.pdf}
\end{frame}
% Slide Carbon Duration
\begin{frame}%[plain,c]
\frametitle{How long will the change last?}
% \framesubtitle{}
\vspace*{0em}
\begin{columns}[c]
\begin{column}{0.75\linewidth}
\begin{center}
\includegraphics[width=0.9\linewidth]{../resources/fig/external/papers/inmanCarbonForever2008/figure1.pdf}%
\end{center}
\end{column}
\begin{column}{0.25\linewidth}
Model simulation of atmospheric \(\text{CO}_2\) concentration from
combustion of fossil fuels\footcite{inmanCarbonForever2008}.
\end{column}
\end{columns}
\vspace*{0em}
\end{frame}
% Slide The most important problem 3 Decarbonize by 2050
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-3.pdf}
\end{frame}
% Slide Energy mix
\begin{frame}%[plain,c]
\frametitle{We are 20 percent done}
% \framesubtitle{}
\vspace*{0em}
\begin{columns}[c]
\begin{column}{0.75\linewidth}
\begin{center}
\includegraphics[width=0.9\linewidth]{../resources/fig/external/papers/ritchieEnergyMix2024/sub-energy-fossil-renewables-nuclear-cropped2.pdf}%
\end{center}
\end{column}
\begin{column}{0.25\linewidth}
Primary energy consumption from fossil fuels, nuclear and renewables, World
\footcite{ritchieEnergyMix2024}.
\end{column}
\end{columns}
\vspace*{0em}
\end{frame}
% Slide The most important problem 4 The energy challenge
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-4.pdf}
\end{frame}
% Slide IT energy use 1
\begin{frame}%[plain,c]
\frametitle{{\small Example: The Energy Challenge in Information Technology}}
\framesubtitle{Overall power use}
\vspace*{0em}
\begin{columns}[c]
\begin{column}{0.75\linewidth}
\begin{center}
\includegraphics[width=0.9\linewidth]{../resources/fig/external/papers/andraeHypothesesPrimaryEnergy2020/fig1-no-title.pdf}%
\end{center}
\end{column}
\begin{column}{0.25\linewidth}
Computing power trends in relation to global power consumption 2010 to 2030.
\footcite{andraeHypothesesPrimaryEnergy2020}.
\end{column}
\end{columns}
\vspace*{0em}
\end{frame}
% Slide IT energy use 2
\begin{frame}[plain,c]
\frametitle{{\small Example: The Energy Challenge in Information Technology}}
% \framesubtitle{AI power use\footcite{sevillaComputeTrendsThree2022}}
\vspace*{0em}
\begin{center}
\includegraphics[width=0.8\linewidth]{../resources/fig/external/papers/sevillaComputeTrendsThree2022/epoch-ai.org/ml-trends-2024-09-08.png}%
\footcite{sevillaComputeTrendsThree2022}
\end{center}
\vspace*{0em}
\end{frame}
% Slide Energy mix
\begin{frame}%[plain,c]
\frametitle{A \textit{disruptive} technology: Quantum Materials}
% \framesubtitle{}
\vspace*{0em}
Precise control of the electron spin will enable ultra-low power, neuromorphic and
quantum computing
\begin{columns}[c]
\begin{column}{0.5\linewidth}
\begin{center}
\includegraphics[width=0.6\linewidth]{../resources/fig/external/papers/hoffmannAntiskyrmionsStabilizedInterfaces2017/fig1-Skyrmion_and_antiskyrmion_magnetic_configurations.jpg}\footcite{hoffmannAntiskyrmionsStabilizedInterfaces2017}%
Magnetic skyrmion
\end{center}
\end{column}
\begin{column}{0.5\linewidth}
\includegraphics[width=0.7\linewidth]{../resources/fig/external/papers/russmannDensityfunctionalDescriptionMaterials2023/fig6a1.pdf}\footcite{russmannDensityfunctionalDescriptionMaterials2023}%
Majorana zero modes
\end{column}
\end{columns}
\vspace*{0em}
\end{frame}
% Slide The most important problem 5 Is a materials challenge
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-5.pdf}
\end{frame}
% Slide materials development duration
\begin{frame}[plain,c]
\frametitle{Materials discovery is too slow}
% \framesubtitle{}
\vspace*{0em}
\begin{center}
\includegraphics[width=0.8\linewidth]{../resources/fig/external/papers/reinhardtGettingMaterialsOut2024/materials-take-decades-to-make-it-out-of-the-lab.png}%
\footcite{reinhardtGettingMaterialsOut2024}
\end{center}
\vspace*{0em}
\end{frame}
% Slide The most important problem 6 Accelerate discovery
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-6.pdf}
\end{frame}
% Slide 5th paradigm
\begin{frame}%[c,plain]
\frametitle{Accelerate materials discovery}
\framesubtitle{With emerging scientific paradigms}
\includegraphics[width=1.0\textwidth]{../resources/fig/fifth-paradigm/fifth-paradigm-helmholtz-blue6-background3-mint.pdf}
\end{frame}
% Slide The most important problem 7 Computational materials design
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-7.pdf}
\end{frame}
% Slide First Principles WFT and DFT
\begin{frame}%[c,plain]
\frametitle{First-principles electronic structure methods}
% \framesubtitle{}
\begin{columns}[t]
\begin{column}{0.75\linewidth}
\begin{quote}
\enquote{The underlying laws necessary for the mathematical theory of a large part of physics and the whole of chemistry are thus completely known, and the difficulty is only that exact applications of these laws lead to equations which are too complicated to be soluble.}%
\vspace*{1em}
P.M.A. Dirac, Proceedings of the Royal Society A123, 714 (1929)
\end{quote}
\end{column}
\begin{column}{0.25\linewidth}
\includegraphics[width=0.5\textwidth]{../resources/fig/external/historical-figures/paul-dirac/Paul_Dirac,_1933.jpg}%
\end{column}
\end{columns}
\vspace*{2em}
\begin{columns}[t]
\begin{column}{0.3\linewidth}
\textbf{Schrödinger equation}
\vspace*{1em}
\hspace*{4em} {\large \textcolor{fzjgreen}{\faArrowAltCircleDown}}
\vspace*{1em}
\textbf{Density functional theory}
\end{column}
\begin{column}{0.4\linewidth}
\(
\mathcal{H} \ket{\Psi} = E \ket{\Psi}, \quad \Psi(\boldsymbol{r}, \boldsymbol{r}_2, \dots, \boldsymbol{r}_N)
% \mathcal{H} = - \sum_a^M \nabla_a^2 - \sum_i^N \frac{1}{2} \nabla_i^2 + \sum_{a < b}^M \frac{Z_a Z_b}{r_{ab}} - \sum_i^{M,N} \frac{Z_a}{r_{ia}} + \sum_{i < j}^N \frac{1}{r_{ij}}
\)
\vspace*{3em}
\(
\left( - \nabla_i^2 + v_{\text{eff}}(\boldsymbol{r}) \right) \varphi_i(\boldsymbol{r}) = \varepsilon_i \varphi_i(\boldsymbol{r})
\)
\end{column}
\begin{column}{0.3\linewidth}
\(\mathcal{O}(3^N)\)
\vspace*{1em}
\hspace*{0.5em} {\large \textcolor{fzjgreen}{\faArrowAltCircleDown}}
\vspace*{1em}
\(\mathcal{O}(N_{el}^3)\)
\end{column}
\end{columns}
\end{frame}
% Slide The most important problem 8 Is not fast enough alone
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-8.pdf}
\end{frame}
% Slide The most important problem 9 AI to the rescue!
\begin{frame}%[c,plain]
\frametitle{The most important problem}
% \framesubtitle{}
\includegraphics[width=1.0\textwidth]{../resources/fig/presentation-2024-09/the-most-important-problem/mip-9.pdf}
\end{frame}
%%% Local Variables:
%%% mode: latex
%%% TeX-master: "../presentation"
%%% mode: flyspell
%%% ispell-local-dictionary: "english"
%%% End:
presentation/presentation.tex
+ 1
− 1
View file @ e13f0537
...@@ -175,7 +175,7 @@ ...@@ -175,7 +175,7 @@
% MAINMATTER % MAINMATTER
% %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% \input{mainmatter/introduction} \input{mainmatter/introduction}
% \input{mainmatter/kkr-theory} % \input{mainmatter/kkr-theory}
% \input{mainmatter/kkr-jlcdm} % \input{mainmatter/kkr-jlcdm}
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment