Skip to content

Commit

Permalink
Update Coral TPU and 10G ASUS NIC.
Browse files Browse the repository at this point in the history
  • Loading branch information
geerlingguy committed Oct 11, 2023
1 parent 4488951 commit cc943ce
Show file tree
Hide file tree
Showing 2 changed files with 11 additions and 8 deletions.
14 changes: 7 additions & 7 deletions _cards_m2/coral-accelerator-ae-key.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,19 +2,19 @@
layout: card
title: "Google Coral TPU M.2 Accelerator A+E Key"
picture: "/images/m2-coral-accelerator-ae-key.jpg"
functionality_cm4: "Currently Testing"
functionality_pi5: "Untested"
driver_required: "Maybe"
functionality_cm4: "None"
functionality_pi5: "Currently Testing"
driver_required: "Yes"
github_issue: "https://github.com/geerlingguy/raspberry-pi-pcie-devices/issues/44"
buy_link: https://amzn.to/3jWowRg
videos: []
---
I just bought this card and am currently testing it.
Coral TPU accelerators offer an easy way to add advanced (and fast!) ML and AI coprocessors to embedded devices. You can use them for things like image recognition, and they are integrated into popular software like Frigate.

It seems like the Google Coral Drivers are currently not working on the Pi OS, either 32-bit or 64-bit. In my testing, after installing the drivers, the entire Pi locks up.
The PCIe version of the Google Coral completely locks up on the Compute Module 4, though the [USB TPU version](https://amzn.to/3qGR2sE) works.

Google employees have mentioned they plan on trying to get it working, but there is no specific timeline.
On the Raspberry Pi 5, I am able to follow the [install guide for PCIe](https://coral.ai/docs/m2/get-started), and have the device recognized.

The [USB TPU version](https://amzn.to/3qGR2sE) does work with the Pi already, but with slightly higher latencies due to the overhead of the USB connection.
But with Pi OS 12 (Bookworm), the system Python version is too new for the `pycoral` library. So I've been [testing my Coral TPU using Docker](https://www.jeffgeerling.com/blog/2023/testing-coral-tpu-accelerator-m2-or-pcie-docker), but I am still running into some issues.

Please follow the GitHub issue for more details.
5 changes: 4 additions & 1 deletion _cards_network/asus-xg-c100c-10g.md
Original file line number Diff line number Diff line change
Expand Up @@ -3,11 +3,12 @@ layout: card
title: "ASUS XG-C100C 10G Network Adapter"
picture: "/images/network-asus-xg-c100c.jpg"
functionality_cm4: "Full"
functionality_pi5: "Untested"
functionality_pi5: "Full"
driver_required: "Yes"
github_issue: "https://github.com/geerlingguy/raspberry-pi-pcie-devices/issues/15"
buy_link: https://amzn.to/38wYOiL
videos:
- https://www.youtube.com/watch?v=nBtOEmUqASQ
- https://www.youtube.com/watch?v=FTP5h9jnVx0
---
This 10 Gigabit network card has an Aquantia AQC107 802.3bz NBase-T ethernet controller, which requires the AQtion driver.
Expand All @@ -33,4 +34,6 @@ After copying over the new kernel and installing modules, you should see a new n

On my Compute Module 4, without any overclock and without changing the MTU from the default of 1500, I was able to get about 3.26 Gbps of bandwidth through the card.

On my Pi 5, I was able to get about 6 Gbps of bandwidth if I set the external port to PCIe Gen 3.0 speeds.

Note that you may be able to get 10 gigabit speeds over shorter Cat5e or Cat6 cable runs, but Cat6a or better cabling is recommended for 10G over copper!

0 comments on commit cc943ce

Please sign in to comment.