Repository navigation
VASP: Genoa/Milan and GPU examples, NUMA settings, shorter page - #1484
Merged
Merged
Conversation
- Example batch scripts in VASP6 and VASP5 tabs, each with Genoa and Milan tabs. VASP6 also has GPU node scripts: H100 on Genoa and A100 80 GB on Milan, using VASP/6.5.1-nvofbf-2023b. - Scripts follow the NUMA Domains page: --cores-per-socket from slurm_limits, --threads-per-core 1, --distribution *:block:*, --mem-bind local, OMP_PROC_BIND, OMP_PLACES, and its print_binding_stats step. The job-log and "Finished printing" echo lines are removed. - GPU builds are the -nvofbf- modules, replacing the -NVHPC- note. - The theory section links to the NUMA Domains page instead of repeating it, is shorter, and moves to the end of the page. - Repetition removed from the efficiency, GPU and useful information sections. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Contributor
|
Test deployment successful!! (2026-11-16T01:57:51Z) See all deployed demo sites |
Contributor
|
| Line | Check | Message |
|---|---|---|
| 59 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
| 90 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
| 121 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
| 153 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
| 185 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
| 214 | meta: code_line_length | Code line is 356 characters long, so it will scroll sideways. Code blocks fit about 80 characters. |
ℹ️ 14 Notices
docs/Software/Available_Applications/VASP.md (14)
| Line | Check | Message |
|---|---|---|
| 54 | meta: module_version_hardcoded | 'VASP/6.5.1-foss-2023a' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 85 | meta: module_version_hardcoded | 'VASP/6.5.1-nvofbf-2023b' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 116 | meta: module_version_hardcoded | 'VASP/6.5.1-foss-2023a' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 148 | meta: module_version_hardcoded | 'VASP/6.5.1-nvofbf-2023b' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 180 | meta: module_version_hardcoded | 'VASP/5.4.4-intel-2020a' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 209 | meta: module_version_hardcoded | 'VASP/5.4.4-intel-2020a' has the version typed in, so it will go out of date. If any version will do, use 'VASP/{{ applications.VASP.default }}'. |
| 234 | meta: walk_toc | Header 'Step 1: INCAR & KPOINTS Benchmarking' is too long. Try to keep h3 headers under 26 characters to avoid word wrapping in the toc. |
| 271 | meta: section_length | Section 'VASP extensions' is only ~362 chars. Sections under 400 will be lumped into the next section when parsed by the RAG. |
| 275 | meta: walk_toc | Header 'The theory behind VASP efficiency' is too long. Try to keep h2 headers under 28 characters to avoid word wrapping in the toc. |
| 275 | meta: section_length | Section 'The theory behind VASP efficiency' is only ~186 chars. Sections under 400 will be lumped into the next section when parsed by the RAG. |
| 287 | meta: walk_toc | Header 'How parallelisation works in VASP' is too long. Try to keep h3 headers under 26 characters to avoid word wrapping in the toc. |
| 299 | meta: walk_toc | Header 'What happens when calculating an orbital in VASP' is too long. Try to keep h3 headers under 26 characters to avoid word wrapping in the toc. |
| 305 | meta: walk_toc | Header 'How do we decrease the latency (and increase speed) of calculations?' is too long. Try to keep h3 headers under 26 characters to avoid word wrapping in the toc. |
| 341 | meta: walk_toc | Header 'My job is spread across different nodes — is this a problem?' is too long. Try to keep h3 headers under 26 characters to avoid word wrapping in the toc. |
See the Checks tab for the full output.
geoffreyweal
marked this pull request as ready for review
November 16, 2026 02:20
CallumWalley
approved these changes
Nov 19, 2026
geoffreyweal
added a commit
that referenced
this pull request
Nov 19, 2026
) ## What changed and why The ORCA page's example script predates the NUMA Domains page. This brings the page in line with it, the same way #1484 does for VASP, and shortens the rest of the page. - **Example batch scripts:** one for Genoa and one for Milan, with the NUMA settings from the NUMA Domains page (the four `#SBATCH` options, the two `export` lines and the `print_binding_stats` step). They also gain `--account`, `module purge` and the default-version macro. - **Input file settings (was "Requesting a parallel run"):** moved up next to the examples, and now sets `%maxcore` as well as `%pal`. ORCA's default `%maxcore` is 4096 MB per core, and ORCA can use more than `%maxcore`, while the examples ask for 1 GB per core. Without it, larger calculations could run out of memory. - **NUMA settings (new section):** Genoa and Milan tabs with the NUMA lines from the example scripts, what they are for, and the `print_binding_stats` command with a link to how to read its output. The detail stays on the NUMA Domains page. - **Benchmark your ORCA jobs (new section):** points to the [ORCA benchmarking tool](https://github.com/geoffreyweal/orca-benchmarking), which holds its own instructions. - **Checkpointing:** now a top-level section, as "Further notes" is gone. Shorter, filenames in backticks, the restart lines in the manual's form, and a link to the manual's restart section. Its anchor is unchanged, so the link from Job Checkpointing still works. - **Removed:** the intro paragraph and homepage link. The app header already shows both, so the page showed them twice. - **Licensing:** shorter, with the same meaning. Its "ORCA licence" link returned 404 and is removed. Page to look at on the preview site: **Software → Supported Applications → ORCA**. ## Needs a decision The page says ORCA's licence allows **academic research only**. The [FACCTs ORCA page](https://www.faccts.de/orca/) now says "ORCA is and will remain free for academic and personal use". I left the page's wording as it was. I also found no current URL for the licence text itself to replace the dead link. ## Sources | Fact | Source | | - | - | | The four `#SBATCH` options, the two `export` lines and the `print_binding_stats` command | `docs/Software/Parallel_Computing/NUMA_Domains.md`. The command is copied byte for byte. | | Cores per NUMA domain (21 on Genoa, 16 on Milan) | `slurm_limits.cpu_nodes` macros, from `docs/assets/slurm-limits.json` | | ORCA module version | `applications.ORCA.default` macro, from `docs/assets/module-list.json` | | NUMA domains can slow a job or make its runtime vary | NUMA Domains page, "When NUMA domains matter" | | `%maxcore` is per core in MB, defaults to 4096, ORCA can use more, give it about 80% of the memory | [ORCA 6.1 tutorial: Memory Settings](https://www.faccts.de/docs/orca/6.1/tutorials/first_steps/memory.html) | | With `--threads-per-core 1`, `--mem-per-cpu` is memory per physical core | `docs/Software/Parallel_Computing/Simultaneous_Multithreading.md` | | GBW renaming, `! moread` / `%moinp`, older GBW files | [ORCA 6.1.1 manual, 2.20.5 Restarting SCF Calculations](https://orca-manual.mpi-muelheim.mpg.de/contents/essentialelements/initialguess.html#restarting-scf-calculations) | | The app header already shows the description and homepage | `overrides/partials/app_header.html`, and the built page, which had each twice | | What the benchmarking tool does | Its README | The links into the NUMA Domains page use anchors that exist both on `main` and in #1483. --------- Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What this changes
Brings the VASP page (Software > Available Applications > VASP) in line with the new NUMA Domains page (#1478),
adds Genoa, Milan and GPU example batch scripts, and removes repetition without dropping information.
VASP6 also has CPU and GPU node tabs: an H100 script for Genoa and an A100 80 GB script for Milan,
using
VASP/6.5.1-nvofbf-2023bwith one MPI task per GPU.--partition,--cores-per-socket(from
slurm-limits.json: 21 on Genoa, 16 on Milan),OMP_PROC_BINDandOMP_PLACES,use
--threads-per-core 1in place of--extra-node-info=1:*:1, and keep--distribution *:block:*and--mem-bind local.The Slurm header now uses whitespace as the delimiter, and each script runs the NUMA page's
print_binding_statsstep beforesrun vasp_std.One warning above the tabs gives the
--cores-per-socketvalue for each partition.-nvofbf-modules are the GPU builds, replacing the note about-NVHPC-modules.for how jobs lose time across NUMA domains and what each option does, instead of repeating it, and shows the options in Genoa and Milan tabs.
"Some CPUs do not need to lie as close to each other as others" is merged into "My job is spread across different nodes".
The theory section moves to the end of the page, above References.
(the
KPARnote is now part of its bullet). Prose is down from 2,164 words on main to 1,958, despite the new scripts and NUMA guidance.Removed from the example scripts: the job-log
echo(~/VASP_job_log.txt) and the "Finished printing CPU binding information"echo.Page to look at on the preview site: VASP.
Sources
slurm-limits.json, throughslurm_limits.cpu_nodes.OMP_*variables, theprint_binding_statsline, and the GPU options(
--gpus-per-node,--constraint a100_80gb,--gres-flags enforce-binding,--gpu-bind closest): the NUMA Domains page.module-list.json(VASP/6.5.1-foss-2023a,VASP/6.5.1-nvofbf-2023b,VASP/5.4.4-intel-2020a).-nvofbf-modules being the GPU builds, and the-NVHPC-note no longer being correct: the author.