ethernet topology configuration tool for Tenstorrent silicon
Project description
TT-Topology
Tenstorrent Topology (TT-Topology) is a command line utility used to flash multiple n150 or n300 cards on a system to use specific single-host ETH routing configurations.
It currently supports three configurations: mesh, linear, and torus.
[!WARNING]
tt-topologyis not designed to be used with the following products:
- BH PCIe cards
- WH 6U Galaxy systems
- BH 6U Galaxy systems
The tool will throw an error if used with unsupported products.
Additionally,
tt-topologyis designed to be used only in a single-host context. Multi-host topologies will not be discovered.
Official Repository
https://github.com/tenstorrent/tt-topology/
Getting started
Install Rust (if you don't already have it)
If Rust isn't already installed on your system, you can install it through either of the following methods:
Using Distribution packages (preferred)
-
Fedora / EL9
sudo dnf install cargo -
Ubuntu / Debian
sudo apt install cargo
Using Rustup
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"
Installation (for users)
tt-topology is available on PyPi and can be installed in your Python (v3.10 and up) environment using pip.
pip install tt-topology
[!IMPORTANT] It is always recommended to manage, build, and install Python packages within a virtual environment.
A virtual environment can be created using
venv:python -m venv .venv source .venv/bin/activate
Installation (for developers)
Clone the repository
git clone https://github.com/tenstorrent/tt-topology.git
cd tt-topology
Install
pip install .
or for users who would like to edit the code without re-building, install tt-topology in editable mode.
pip install --editable .
Recommended: install the pre-commit hooks so there is auto formatting for all files on committing.
pre-commit install
Usage
Command line arguments
usage: tt-topology [-h] [-v] [-l {linear,torus,mesh,isolated}] [-o] [-f [filename]] [-g] [-ls] [--log [log]] [-p [plot]] [-r [config.json ...]]
Tenstorrent Topology (TT-Topology) is a command line utility to flash ethernet coordinates when multiple NB's are connected together.
options:
-h, --help show this help message and exit
-v, --version show program's version number and exit
-l {linear,torus,mesh,isolated}, --layout {linear,torus,mesh,isolated}
Select the layout (linear, torus, mesh, isolated). Default is linear.
-o, --octopus
-f [filename], --filename [filename]
Change filename for test log. Default: ~/tt_smi/<timestamp>_snapshot.json
-g, --generate_reset_json
Generate default reset json file that reset consumes. Update the generated file and use it as an input for the --reset option
-ls, --list List out all the boards on host with their coordinates and layout.
--log [log] Change filename for the topology flash log. Default: ~/tt_topology_logs/<timestamp>_log.json
-p [plot], --plot_filename [plot]
Change the plot of the png that will have the graph layout of the chips. Default: chip_layout.png
-r [config.json ...], --reset [config.json ...]
Provide a valid reset JSON
TT-Topology Procedure
TT-Topology does the following when calculating and flashing the coordinates:
- Flash all the boards to default - set all eth port disables to 0 and reset coordinates to (0,0) for local chips and (1,0) for n300 remote chips.
- Issue a board level reset to apply the new flash to the chips.
- Generate a mapping of all possible connections and their type between the available chips.
- Using a graph algorithm generate coordinates for each chip based on user input. These layouts are discussed in detail in the sections below.
- Write the new coordinates to the chips.
- Issue a board level reset to apply the new flash to the chips.
- Return a png with a graphic representation of the layout and a .json log file with details of the above steps.
Chip layouts
TT-Topology can be used to flash one of the three chip layouts: mesh, linear and torus.
Mesh
The mesh layout is a trivalent graph where each node can have a maximum of three connections. A BFS algorithm is used to assign the coordinates.
The command to generate a mesh layout is:
$ tt-topology -l mesh -p mesh_layout.png
For a host with two n300 cards and four n300 cards, the command will generate layouts that look like the following:
Linear
The linear layout, as the name suggests, is a layout where all chips are connected in a single line. The coordinates are assigned by finding a cycle in the graph and then assigning coordinates in order.
The command to generate a linear layout is:
$ tt-topology -l linear -p linear_layout.png
For a host with two n300 cards and four n300 cards, the command will generate layouts that look like the following:
Torus
The torus layout is a cyclic graph where a single line connects all nodes.
The coordinates are assigned by finding a cycle in the graph and then assigning coordinates in order.
The command to generate a torus layout is:
$ tt-topology -l torus -p torus_layout.png
For a host with two n300 cards and four n300 cards, the command will generate layouts that look like the following:
Octopus (TGG/TG) Support in TT-Topology
- TGG setting: 8 n150 cards connected to 2 Galaxy 4U systems
- TG setting: 8 n150 cards connected to 1 Galaxy 4U system
Usage
-
Generate a default mobo reset json file saved at
~/.config/tenstorrent/reset_config.jsonby running the following command$ tt-topology -g -
Fill in "mobo", "credo", and "disabled_ports" under "wh_mobo_reset"
Here is an example of what your reset_config.json file may look like:
{ "time": "2024-03-06T20:12:27.640859", "host_name": "yyz-lab-212", "gs_tensix_reset": { "pci_index": [] }, "wh_link_reset": { "pci_index": [ 0, 1, 2, 3 ] }, "re_init_devices": true, "wh_mobo_reset": [ { "nb_host_pci_idx": [ 0, 1, 2, 3 ], "mobo": "mobo-ce-44", "credo": [ "6:0", "6:1", "7:0", "7:1" ], "disabled_ports": [ "0:2", "1:2", "6:2", "7:2" ] } ] } -
Flashing multiple NB cards to use specific eth routing configurations by running the following command
$ tt-topology -o -r ~/.config/tenstorrent/reset_config.json
Internal Procedure
- Setup
mobo_eth_enon every local n150 to train with the Galaxy - Program the shelf/rack of the Galaxies
- Program all local n150s to rack 0, shelf 0, x 0, y 0
- Reset with the following
retimer_selanddisable_seland wait for trainingretimer_sel: From thecredofield of the reset json file for the specific Galaxydisable_sel: All the other ports not specified by theretimer_sel
- Check QSFP link and change shelf number for each n150 according to the shelf on the connected Galaxy
- Program the x, y coords of the local n150s based on the other side of the link
- Reset again with the
retimer_selanddisable_seland wait for training, and verify all chips show upretimer_sel: From thecredofield of the reset json file for the specific Galaxydisable_sel: From thedisabled_portsfield of the reset json file for the specific Galaxy
Logging
TT-Topology records the pre- and post-flash relevant SPI parameter values, connection maps, and coordinates of the chips in a .json file for record keeping and debugging.
By default it is stored at ~/tt_topology_logs/<timestamp>_log.json. This can be changed by using the --log CLI arg:
$ tt-topology --log new_log.json ...
License
Apache 2.0 - https://www.apache.org/licenses/LICENSE-2.0.txt
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file tt_topology-1.2.19.tar.gz.
File metadata
- Download URL: tt_topology-1.2.19.tar.gz
- Upload date:
- Size: 108.0 kB
- Tags: Source
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
91034ae9b222def5c03f949610d4adfc01fdfc843274f96f8501384b20655ded
|
|
| MD5 |
0322a652a614206bc2f76b6e4cd18b07
|
|
| BLAKE2b-256 |
fedb47db54dd53d50ec0715a8ae455c770345c7040c451e0b5a665d50ed28775
|
Provenance
The following attestation bundles were made for tt_topology-1.2.19.tar.gz:
Publisher:
release.yml on tenstorrent/tt-topology
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
tt_topology-1.2.19.tar.gz -
Subject digest:
91034ae9b222def5c03f949610d4adfc01fdfc843274f96f8501384b20655ded - Sigstore transparency entry: 999876226
- Sigstore integration time:
-
Permalink:
tenstorrent/tt-topology@aa6d7123446c64d3ca59e4b2c5c3027782c224a9 -
Branch / Tag:
refs/heads/main - Owner: https://github.com/tenstorrent
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
release.yml@aa6d7123446c64d3ca59e4b2c5c3027782c224a9 -
Trigger Event:
workflow_dispatch
-
Statement type:
File details
Details for the file tt_topology-1.2.19-py3-none-any.whl.
File metadata
- Download URL: tt_topology-1.2.19-py3-none-any.whl
- Upload date:
- Size: 102.9 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? Yes
- Uploaded via: twine/6.1.0 CPython/3.13.7
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
4de7d2924ce9593ac13287e077395aa9dc972ee94f620069fcc391784ce690c8
|
|
| MD5 |
d6ad61197845dab047d24319d0e5596c
|
|
| BLAKE2b-256 |
7d3f5ac4730d8aaaded7792a6a1f423ab49f327b8bf66f8c767cfb0a5736e87d
|
Provenance
The following attestation bundles were made for tt_topology-1.2.19-py3-none-any.whl:
Publisher:
release.yml on tenstorrent/tt-topology
-
Statement:
-
Statement type:
https://in-toto.io/Statement/v1 -
Predicate type:
https://docs.pypi.org/attestations/publish/v1 -
Subject name:
tt_topology-1.2.19-py3-none-any.whl -
Subject digest:
4de7d2924ce9593ac13287e077395aa9dc972ee94f620069fcc391784ce690c8 - Sigstore transparency entry: 999876329
- Sigstore integration time:
-
Permalink:
tenstorrent/tt-topology@aa6d7123446c64d3ca59e4b2c5c3027782c224a9 -
Branch / Tag:
refs/heads/main - Owner: https://github.com/tenstorrent
-
Access:
public
-
Token Issuer:
https://token.actions.githubusercontent.com -
Runner Environment:
github-hosted -
Publication workflow:
release.yml@aa6d7123446c64d3ca59e4b2c5c3027782c224a9 -
Trigger Event:
workflow_dispatch
-
Statement type: