nesi · CallumWalley · Aug 11, 2026 · Aug 12, 2026 · Aug 12, 2026 · Aug 12, 2026
diff --git a/checks/run_test_build.py b/checks/run_test_build.py
@@ -4,17 +4,35 @@
 from mkdocs.commands import build, serve
 from mkdocs.config.base import Config, load_config
 import logging
+import os
 import sys
 import re
+import tempfile
 import time
 
+import requests
 
-""" 
+
+"""
 This works but is a bit messy
 """
 
 msg_count = {"DEBUG": 0, "NOTICE": 0, "WARNING": 0, "ERROR": 0}
 
+MODULES_LIST_URL = "https://raw.githubusercontent.com/nesi/modules-list/main/module-list.json"
+
+
+def fetch_module_list():
+    """Fetch the latest module-list.json contents, or None if unavailable."""
+    try:
+        response = requests.get(MODULES_LIST_URL, timeout=10)
+        response.raise_for_status()
+        return response.content
+    except requests.RequestException as e:
+        print(f"::WARNING file={__file__},title=module_list_fetch_failed,col=0,endColumn=0,line=0::"
+              f"Could not fetch latest module-list.json ({e}); using committed copy instead.")
+        return None
+
 
 def parse_macro(record):
 
@@ -48,12 +66,6 @@ def parse_macro(record):
     return True
 
 
-def count_msg(record):
-    msg_count[record.levelname] += 1
-
-    return True
-
-
 if __name__ == '__main__':
     # Github uses 'NOTICE' rather than 'INFO'
     # This should overwrite existing INFO level.
@@ -62,10 +74,20 @@ def count_msg(record):
     log.setLevel(logging.INFO)
     sh = logging.StreamHandler(sys.stdout)
     sh.addFilter(parse_macro)
-    sh.addFilter(count_msg)
     sh.setFormatter(logging.Formatter(
         '::%(levelname)s file=%(filename)s,title=%(name)s,col=0,endColumn=0,line=%(lineno)s::%(message)s'))
     log.addHandler(sh)
+
+    module_list = fetch_module_list()
+    tmp_module_list_path = None
+    if module_list is not None:
+        # mkdocs_hooks.py / macro_hooks.py read MODULE_LIST_PATH at import time,
+        # so it must be set before load_config() pulls those in.
+        fd, tmp_module_list_path = tempfile.mkstemp(suffix=".json", prefix="module-list-")
+        with os.fdopen(fd, "wb") as f:
+            f.write(module_list)
+        os.environ["MODULE_LIST_PATH"] = tmp_module_list_path
+
     config = load_config(config_file_path="./mkdocs.yml")
     config.plugins.on_startup(command='build', dirty=True)
     try:
@@ -75,6 +97,16 @@ def count_msg(record):
         sys.exit(1)
     finally:
         config.plugins.on_shutdown()
+        if tmp_module_list_path:
+            os.remove(tmp_module_list_path)
+
+    if module_list is not None:
+        # Overwrite the stale copy mkdocs just copied from docs/assets/module-list.json,
+        # so the client-side app filter (supportedApplications.js) also sees fresh data.
+        site_module_list_path = os.path.join(config.site_dir, "assets", "module-list.json")
+        os.makedirs(os.path.dirname(site_module_list_path), exist_ok=True)
+        with open(site_module_list_path, "wb") as f:
+            f.write(module_list)
 
     time.sleep(5)
     # exit(100 < msg_count["NOTICE"] + (30 * msg_count["WARNING"] + (100 * msg_count["ERROR"])))
diff --git a/docs/Batch_Computing/Hardware.md b/docs/Batch_Computing/Hardware.md
@@ -22,81 +22,90 @@
 <table>
     <tr>
         <td>Architecture</td>
-        <td>Core</td>
+        <td>Cores</td>
         <td colspan="2">Memory</td>
         <td>GPU</td>
         <td>Nodes</td>
     </tr>
     <tr>
-        <td rowspan="3">2 x AMD Milan 7713 CPU</br>└ 8 x Chiplets<br>&nbsp;&nbsp;&nbsp;&nbsp;└ 8 x Cores</td>
-        <td rowspan="3">126</td>
+        <td rowspan="2">2 x AMD Milan 7713 CPU<br>└ 8 x Chiplets<br>&nbsp;&nbsp;&nbsp;&nbsp;└ 8 x Cores</td>
+        <td rowspan="2">128</td>
         <td>512GB</td>
         <td><em>(4GB / Core)</em></td>
         <td>-</td>
-        <td>54</td>
+        <td>55</td>
     </tr>
     <tr>
-        <td rowspan="2">1024GB</td>
-        <td rowspan="2"><em>(8GB / Core)<em></td>
+        <td>1024GB</td>
+        <td><em>(8GB / Core)</em></td>
         <td>-</td>
         <td>8</td>
     </tr>
     <tr id="gpu-milan-a100">
+        <td>1 x AMD Milan 7713P CPU<br>└ 8 x Chiplets<br>&nbsp;&nbsp;&nbsp;&nbsp;└ 8 x Cores</td>
+        <td>64</td>
+        <td>512GB</td>
+        <td><em>(8GB / Core)</em></td>
         <td>4 x NVIDIA HGX A100</td>
         <td>4</td>
     </tr>
     <tr>
-        <td rowspan="5">2 x AMD Genoa 9634 CPU</br>└ 12 x Chiplets</br>&nbsp;&nbsp;&nbsp;&nbsp;└ 7 x Cores</td>
-        <td rowspan="5">166</td>
-        <td>358GB</td>
-        <td><em>(1GB / Core)</em></td>
+        <td rowspan="5">2 x AMD Genoa 9634 CPU<br>└ 12 x Chiplets<br>&nbsp;&nbsp;&nbsp;&nbsp;└ 7 x Cores</td>
+        <td rowspan="5">168</td>
+        <td>384GB</td>
+        <td><em>(2GB / Core)</em></td>
         <td>-</td>
         <td>44</td>
     </tr>
-    <tr id="gpu-genoa-a100">
-        <td>716GB</td>
-        <td><em>(2GB / Core)</em></td>
-        <td>2 x NVIDIA A100</td>
+    <tr id="gpu-genoa-rtx6000">
+        <td>768GB</td>
+        <td><em>(4GB / Core)</em></td>
+        <td>2 x NVIDIA RTX PRO 6000</td>
         <td>4</td>
     </tr>
     <tr>
-        <td rowspan="3">1432GB</td>
-        <td rowspan="3"><em>(4GB / Core)</em></td>
+        <td rowspan="3">1536GB</td>
+        <td rowspan="3"><em>(8GB / Core)</em></td>
         <td>-</td>
         <td>8</td>
     </tr>
     <tr id="gpu-genoa-h100">
-        <td>2 x NVIDIA H100</td>
+        <td>2 x NVIDIA H100 NVL</td>
         <td>4</td>
     </tr>
     <tr id="gpu-genoa-l4">
         <td>4 x NVIDIA L4</td>
         <td>4</td>
     </tr>
     <tr>
-    </tr>
-    <tr>
-        <td rowspan=2>Intel Xeon Gold</td>
-        <td>80</td>
+        <td>2 x Intel Xeon Gold 6230 CPU<br>&nbsp;&nbsp;&nbsp;&nbsp;(Cascade Lake)</td>
+        <td>40</td>
         <td>1.5TB</td>
-        <td><em>(18GB / Core)</em></td>
+        <td><em>(38GB / Core)</em></td>
         <td>-</td>
         <td>2</td>
     </tr>
     <tr>
-        <td>176</td>
+        <td>4 x Intel Xeon Gold 6238M CPU<br>&nbsp;&nbsp;&nbsp;&nbsp;(Cascade Lake)</td>
+        <td>88</td>
         <td>6TB</td>
-        <td><em>(346GB / Core)</em></td>
+        <td><em>(69GB / Core)</em></td>
         <td>-</td>
         <td>1</td>
     </tr>
 </table>
-
+
+!!! note "Memory figures"
+    Memory shown is the amount physically installed. A small amount is reserved for the operating system,
+    so the memory actually available to jobs is a few percent lower — for example a 512GB Milan node offers
+    480GB to Slurm. A job requesting exactly the full per-core ratio across every core of a node will
+    therefore not fit. Run `sinfo -o '%n %m'` for the exact schedulable figures.
+
 !!! warning "hugemem"
     Jobs will not automatically land on the Intel 'hugemem' nodes. You must specifically request `--partition hugemem`.
     The CPU architecture is different enough from the milan and genoa nodes, you will probably have to recompile your software.


 ## GPUs

 REANNZ HPC has a range of Graphical Processing Units (GPUs) to accelerate compute-intensive research and support more analysis at scale.
@@ -115,23 +124,25 @@
         <td colspan="2">Nodes</td>
     </tr>
     <tr>
-        <td rowspan="2">NVIDIA A100</td>
-        <td rowspan="2"></td>
+        <td>NVIDIA A100 SXM4</td>
+        <td></td>
         <td>80GB</td>
         <td>4</td>
         <td><a href="#gpu-milan-a100">Milan</a></td>
         <td>4</td>
     </tr>
     <tr>
-        <td>40GB</td>
+        <td>NVIDIA RTX PRO 6000</td>
+        <td></td>
+        <td>96GB</td>
         <td>2</td>
-        <td><a href="#gpu-genoa-a100">Genoa</a></td>
+        <td><a href="#gpu-genoa-rtx6000">Genoa</a></td>
         <td>4</td>
     </tr>
     <tr>
-        <td>NVIDIA H100</td>
+        <td>NVIDIA H100 NVL</td>
         <td></td>
-        <td>96GB</td>
+        <td>94GB</td>
         <td>2</td>
         <td><a href="#gpu-genoa-h100">Genoa</a></td>
         <td>4</td>

diff --git a/docs/Batch_Computing/Job_Limits.md b/docs/Batch_Computing/Job_Limits.md
@@ -16,8 +16,8 @@ These are open for review if you find any of them unreasonable or inefficient.
 ![job limits](../assets/images/job_limits.png){ align=right width=75% }
 
 - 10 nodes
-- 21 node-days
-- 21 days
+- 21 days walltime
+- 21 node-days (walltime x nodes)
 
 <hr width=100%>
 

diff --git a/docs/Batch_Computing/Using_GPUs.md b/docs/Batch_Computing/Using_GPUs.md
@@ -31,7 +31,7 @@
 the following option in the header of your submission script:

 ```sl
 #SBATCH --gpus-per-node=<gpu_type>:<gpu_number>
 ```

 where `<gpu_type>` is the type of gpu you want to use (either 'h100', 'a100', or 'l4'), and `<gpu_number>` is the number of gpus you would like to request for your job.
@@ -48,16 +48,18 @@
         <td>Slurm Header</td>
     </tr>
     <tr>
-        <td rowspan="2">NVIDIA A100</td>
-        <td rowspan="2"></td>
+        <td>NVIDIA A100</td>
+        <td></td>
         <td>80GB</td>
         <td>4</td>
-        <td><pre><code>#SBATCH --partition=milan<br>#SBATCH --gpus-per-node=a100:1</code></pre></td>
+        <td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
     </tr>
-    <tr>
-        <td>40GB</td>
+        <tr>
+        <td>RTX 6000</td>
+        <td></td>
+        <td>96GB</td>
         <td>2</td>
-        <td><pre><code>#SBATCH --partition=genoa<br>#SBATCH --gpus-per-node=a100:1</code></pre></td>
+        <td><pre><code>#SBATCH --gpus-per-node=rtx6000:1</code></pre></td>
     </tr>
     <tr>
         <td>NVIDIA H100</td>
@@ -80,7 +82,7 @@
 with the `srun` and `salloc` commands. For example:
 
 ``` sh
-srun --job-name "InteractiveGPU" --gpus-per-node L4:1 --partition genoa --cpus-per-task 8 --mem 2GB --time 00:30:00 --pty bash
+srun --job-name "InteractiveGPU" --gpus-per-node L4:1 --cpus-per-task 8 --mem 2GB --time 00:30:00 --pty bash
 ```
 
 will request and then start a bash session with access to a L4 GPU, for a
@@ -122,14 +124,14 @@
 Please {% include "partials/support_request.html" %} if you need a version not
 available on the platform.


 The CUDA module also provides access to additional command line tools:

 - [nvidia-smi](https://developer.nvidia.com/nvidia-system-management-interface)
            to directly monitor GPU resource utilisation,
 - [nvcc](https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/index.html)
            to compile CUDA programs,
 - [cuda-gdb](https://docs.nvidia.com/cuda/cuda-gdb/index.html)
            to debug CUDA applications.

 In addition, the [cuDNN](https://developer.nvidia.com/cudnn) (NVIDIA
@@ -159,12 +161,11 @@
 #SBATCH --job-name       GPUJob      # job name (shows up in the queue)
 #SBATCH --account        nesi99991   # Your account
 #SBATCH --time           00-00:10:00 # Walltime (DD-HH:MM:SS)
-#SBATCH --partition      genoa       # This means the job will land on A100 with 40GB VRAM
 #SBATCH --gpus-per-node  A100:1      # GPU resources required per node
 #SBATCH --cpus-per-task  2           # number of CPUs per task (1 by default)
 #SBATCH --mem            512MB       # amount of memory per node (1 by default)

 # load CUDA module
 module purge
 module load CUDA/11.0.2

@@ -192,7 +193,7 @@
   (Use "module --force purge" to unload all):

  1) slurm   2) NeSI
 Wed May 12 12:08:27 2021
 +-----------------------------------------------------------------------------+
 | NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |
 |-------------------------------+----------------------+----------------------+
@@ -200,7 +201,7 @@
 | Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
 |                               |                      |               MIG M. |
 |===============================+======================+======================|
 |   0  Tesla P100-PCIE...  On   | 00000000:05:00.0 Off |                    0 |
 | N/A   29C    P0    23W / 250W |      0MiB / 12198MiB |      0%      Default |
 |                               |                      |                  N/A |
 +-------------------------------+----------------------+----------------------+
@@ -208,7 +209,7 @@
 +-----------------------------------------------------------------------------+
 | Processes:                                                                  |
 |  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
 |        ID   ID                                                   Usage      |
 |=============================================================================|
 |  No running processes found                                                 |
 +-----------------------------------------------------------------------------+
@@ -227,7 +228,7 @@

    ```bash
    user.name@login03:$ squeue --me
    JOBID         USER     ACCOUNT   NAME        CPUS MIN_MEM PARTITI START_TIME     TIME_LEFT STATE    NODELIST(REASON)    
    1234567       user.nam nesi99999 Example_GPU_   8     24G genoa   Apr 30 17:36    23:58:08 RUNNING  g09               
    ```

@@ -289,7 +290,7 @@

 ![5831962_profile.png](../assets/images/GPU_5831962_profile.png)

 See [Slurm Native Profiling](../Software/Profiling_and_Debugging/Slurm_Native_Profiling.md) for more information on this feature. 

 ## How to determine which GPU is best for your job

@@ -300,11 +301,13 @@
 When running a 15-minute test job, add the following settings in your Slurm submission script:
 
 ```sl
-#SBATCH --time=00:15:00
-#SBATCH --gpu-per-node=<gpu-type>:1
-#SBATCH --qos=debug
-#SBATCH --profile=task # Only for testing
-#SBATCH --acctg-freq=1 # Only for testing
+#!/bin/bash -e
+
+#SBATCH --time          00:15:00
+#SBATCH --gpu-per-node  <gpu-type>:1
+#SBATCH --qos           debug
+#SBATCH --profile       task # Only for testing
+#SBATCH --acctg-freq    1 # Only for testing
 ```
 
 To record the GPU utilisation and GPU memory, see [Measuring GPU efficiency after a job has finished](./Using_GPUs.md#measuring-gpu-efficiency-after-a-job-has-finished) for more information.
@@ -312,5 +315,5 @@
 ## Application and toolbox specific support pages

 {% for p in pages_with_tag("gpu") %}
 - [{{ p.title }}]({{ p.path }})
 {% endfor %}
diff --git a/docs/FORMAT.md b/docs/FORMAT.md
@@ -12,7 +12,7 @@

 This page is an overview of the Markdown syntax supported in this documentation.

 ## Headers

 Headers should be surrounded by blank lines.

@@ -68,7 +68,7 @@

 _italic_: `_italic`

 ## Tab Containers

 Tab Containers should be surrounded by blank lines.

@@ -84,7 +84,7 @@
    something else
 ```

 ## Admonitions

 Admonitions should be surrounded by blank lines.

@@ -125,8 +125,8 @@
    Wheeee
 ```

 Consider making a Admonition collapsable-open (`???+`) if it is particularly long.
 An Admonition **shouldn't be made collapsable-closed** (`???`) unless it has been given an **optional title explaining it's contents**.

 ### Admonition Flavours

@@ -140,7 +140,7 @@

 !!! tip "with title"
    For tangential actionable advice.  
    *`x` other command may be useful here.*

 !!! info
    Use this to provide (optional) additional context or make an in depth explanation.
@@ -530,12 +530,17 @@
 {% endraw %}
 ```
 
+```{% raw %}{% include %}{% endraw %}``` fails soft in this project: if the included template errors while
+rendering (e.g. an app is missing an expected field in module-list.json), that
+include renders as nothing instead of taking out the whole page. No special
+syntax needed - this is patched into both Jinja environments in `mkdocs_hooks.py`.
+
 There are a few includes you may want to use.
 
 | Path | content | usage |
 | ---- | ------- | ----- |
 | ```{% raw %}{% include "partials/support_request.html" %}{% endraw %}``` | ```<a href="mailto:support@nesi.org.nz">Contact our Support Team</a>``` | Anywhere the user is told to contact support. |
-| ```{% raw %}{% include "partials/appHeader.html" %}{% endraw %}``` | Info block | At the top of documents about particular software (TODO: elaborate) |
+| ```{% raw %}{% include "partials/app_header.html" %}{% endraw %}``` | Info block | At the top of documents about particular software (TODO: elaborate) |
 | ```{% raw %}{% include "partials/app/app_network_licence.html" %}{% endraw %}``` | List of network licences | When dynamic licence info is required (used in `appHeader.html`)  |
 | ```{% raw %}{% include "partials/app/app_version.html" %}{% endraw %}``` | List of versions and a 'module load' code-block. | When dynamic version info is required |
 

diff --git a/docs/Getting_Started/FAQs/How_Do_I_Run_My_Python_Notebook_Through_SLURM.md b/docs/Getting_Started/FAQs/How_Do_I_Run_My_Python_Notebook_Through_SLURM.md
@@ -32,7 +32,7 @@ This option might be less convenient as the exporter saves the python
 file to your local computer, meaning you will have to drag it back into
 the file explorer in Jupyter from your downloads folder.
 
-This script can then be run as a regular python script as described in
+This file can then be run as a regular python script inside a Slurm job as described in
 our
 [Python](../../Software/Available_Applications/Python.md)
 documentation.
diff --git a/docs/Storage/Data_Recovery.md b/docs/Storage/Data_Recovery.md
@@ -5,9 +5,11 @@ tags:
 - storage
 ---
 
+Deleted files may be retrieved from a snapshot.
+
 Snapshots are read only copies of the filesystems at a point in time.
 They are taken daily (or weekly for `/nesi/nobackup`) and retained for
-at least seven days.  
+at least seven days.
 
 Files from your home directory can be found
 in `/home/.snapshots/` followed by a snapshot timestamp and

diff --git a/docs/Storage/Models.md b/docs/Storage/Models.md
@@ -31,8 +31,7 @@ If you need a model that is not listed here, please {% include "partials/support
 </tr>
 <tr>
 <td><pre><code>/opt/nesi/model/gguf/llama3.1/llama3.1-70b.gguf</code></pre></td>
-<td><pre><code>#SBATCH --partition=milan
-#SBATCH --gpus-per-node=a100:1</code></pre></td>
+<td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
 </tr>
 <tr>
 <td rowspan="3"><a href="https://ollama.com/library/deepseek-r1">DeepSeek-R1</a></td>
@@ -42,13 +41,11 @@ If you need a model that is not listed here, please {% include "partials/support
 </tr>
 <tr>
 <td><pre><code>/opt/nesi/model/gguf/deepseek-r1/deepseek-r1-32b.gguf</code></pre></td>
-<td><pre><code>#SBATCH --partition=genoa
-#SBATCH --gpus-per-node=a100:1</code></pre></td>
+<td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
 </tr>
 <tr>
 <td><pre><code>/opt/nesi/model/gguf/deepseek-r1/deepseek-r1-70b.gguf</code></pre></td>
-<td><pre><code>#SBATCH --partition=milan
-#SBATCH --gpus-per-node=a100:1</code></pre></td>
+<td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
 </tr>
 <tr>
 <td rowspan="2"><a href="https://ollama.com/library/qwen3">Qwen3</a></td>
@@ -58,8 +55,7 @@ If you need a model that is not listed here, please {% include "partials/support
 </tr>
 <tr>
 <td><pre><code>/opt/nesi/model/gguf/qwen3/qwen3-32b.gguf</code></pre></td>
-<td><pre><code>#SBATCH --partition=genoa
-#SBATCH --gpus-per-node=a100:1</code></pre></td>
+<td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
 </tr>
 <tr>
 <td rowspan="2"><a href="https://ollama.com/library/qwen2.5">Qwen2.5</a></td>
@@ -75,8 +71,7 @@ If you need a model that is not listed here, please {% include "partials/support
 <td><a href="https://ollama.com/library/gemma3">Gemma 3</a></td>
 <td><a href="https://huggingface.co/google/gemma-3-27b/blob/main/LICENSE">Gemma</a></td>
 <td><pre><code>/opt/nesi/model/gguf/gemma3/gemma3-27b.gguf</code></pre></td>
-<td><pre><code>#SBATCH --partition=genoa
-#SBATCH --gpus-per-node=a100:1</code></pre></td>
+<td><pre><code>#SBATCH --gpus-per-node=a100:1</code></pre></td>
 </tr>
 </tbody>
 </table>