Coral Edge TPU Libraries for Legacy CPUs (SSSE3)

Pre-compiled Google Coral Edge TPU libraries for CPUs that only support SSE/SSE2/SSSE3 instructions — no SSE4.1, SSE4.2, or AVX required.

Who This Is For

If you get Illegal instruction when trying to use a Google Coral PCIe TPU on older hardware, this is for you. The official Google prebuilt libraries require SSE4.2 minimum. These binaries target -march=bonnell (Intel Atom microarchitecture) and work on any x86_64 CPU with SSSE3 support.

Tested Hardware

• CPU: Intel Atom D2700 (Thecus N5550 NAS)
• OS: Debian 12 Bookworm (inside Docker container)
• Kernel: 6.18.x with patched gasket-dkms
• Frigate: 0.17.0

Check If You Need This

grep -m1 flags /proc/cpuinfo | tr ' ' '\n' | grep -E 'sse4|avx'

If this returns nothing — you need these binaries. If it returns sse4_1, sse4_2, or avx — the official Google binaries will work.

Prerequisites

1. Install gasket driver (kernel module for Coral PCIe):

   • For modern kernels (6.4+), use the patched version: https://github.com/google/gasket-driver
   • Apply these patches for kernel 6.18+:

     sed -i -e 's/no_llseek/noop_llseek/g' gasket_core.c
     sed -i -e 's/MODULE_IMPORT_NS(DMA_BUF)/MODULE_IMPORT_NS("DMA_BUF")/g' gasket_page_table.c
     sed -i 's/^[[:space:]]*\.llseek[[:space:]]*=[[:space:]]*no_llseek,[[:space:]]*$//' gasket_core.c
     dpkg-buildpackage -us -uc -tc -b
     sudo dpkg -i ../gasket-dkms_1.0-18_all.deb

   • Verify: ls /dev/apex_0

2. Docker with Frigate container

Installation

Option A: Volume Mount (Recommended — survives container restarts)

1. Download the binaries:

   mkdir -p /opt/coral-libs
   wget https://github.com/thatSFguy/coral-edgetpu-legacy-cpu/raw/main/binaries/libedgetpu.so.1.0 \
     -O /opt/coral-libs/libedgetpu.so.1.0
   wget https://github.com/thatSFguy/coral-edgetpu-legacy-cpu/raw/main/binaries/_pywrap_tensorflow_interpreter_wrapper.so \
     -O /opt/coral-libs/_pywrap_tensorflow_interpreter_wrapper.so

2. Add to your Frigate docker-compose.yml volumes section:

   volumes:
     - /opt/coral-libs/libedgetpu.so.1.0:/usr/lib/x86_64-linux-gnu/libedgetpu.so.1.0
     - /opt/coral-libs/_pywrap_tensorflow_interpreter_wrapper.so:/usr/local/lib/python3.11/dist-packages/tflite_runtime/_pywrap_tensorflow_interpreter_wrapper.so

3. Add Coral detector to Frigate config.yml:

   detectors:
     coral:
       type: edgetpu
       device: pci

4. Restart Frigate:

   docker compose down && docker compose up -d

5. Verify in logs:

   docker exec frigate cat /dev/shm/logs/frigate/current | grep -i "tpu\|edgetpu"

   Should show: TPU found with no errors after it.

Option B: Install Script

curl -sSL https://raw.githubusercontent.com/thatSFguy/coral-edgetpu-legacy-cpu/main/install/install.sh | bash

Build Details

Component	Version	Source
libedgetpu	master @ e35aed1	https://github.com/google-coral/libedgetpu
tflite_runtime	2.17.1	https://github.com/tensorflow/tensorflow @ v2.17.1
Build container	debian:bookworm	Debian 12 Bookworm
glibc	2.36	Debian 12 Bookworm
Python	3.11	-
Compiler flags	-march=bonnell -mno-sse4.1 -mno-sse4.2 -mno-avx -mno-avx2	-

Compatibility

CPU Generation	SSE Support	Works?
Intel Atom (Bonnell/Saltwell)	SSE, SSE2, SSSE3	✅ Yes
Intel Core 2 (pre-2010)	SSE, SSE2, SSSE3	✅ Yes
Intel Nehalem+ (2010+)	SSE4.1, SSE4.2	Use official binaries
Intel Sandy Bridge+ (2011+)	AVX	Use official binaries

Frigate Version Compatibility

These binaries are compiled for Frigate 0.17.0 (tflite_runtime 2.17.1). If Frigate updates and changes its tflite_runtime version, a rebuild will be needed.

Related Issues

These GitHub issues describe the problem this solves:

• google-coral/edgetpu#808
• google/gasket-driver#55
• blakeblackshear/frigate#5932

License

• libedgetpu: Apache 2.0 (https://github.com/google-coral/libedgetpu/blob/master/LICENSE)
• tflite_runtime: Apache 2.0 (https://github.com/tensorflow/tensorflow/blob/master/LICENSE)