i.MX93 and AM62x SoM benchmarks for industrial AI
Israeli module maker Variscite is showing its two latest system on module boards for the first time at Embedded World (EW2023) today with industrial AI benchmarking data.
Both SoMs, launched earlier this year, are compatible with the VAR-SOM Pin2Pin product family, allowing Variscite’s customers to easily scale at any point of the product lifecycle while using the same carrier board for all platforms. The Pin2Pin family provides an extended lifespan, reduced development time, costs, and risks as well as advanced scalability options from entry-level to high-performance modules.
The VAR-SOM-MX93, based on NXP’s i.MX93 microcontroller is aimed at energy-efficient machine learning edge devices, with a rich set of features for markets like industrial, IoT, and smart devices at an attractive price point, starting at only $39.
- i.MX93 System on Module targets edge AI
- Two team for secure embedded IoT on NXP’s i.MX modules
- System on Module uses TI AM62x processor
The 1.7GHz Dual Cortex-A55 i.MX 93 is the industry’s first implementation of ARM’s Ethos-U65 microNPU neural processing unit (NPU) to accelerate machine learning (ML) workloads with a flexible energy sub-system. The SoM offers an additional 250MHz Cortex-M33 real-time co-processor, AI/ML capabilities, built-in security and a wide range of industrial features.
The $33 VAR-SOM-AM62 is powered by Texas Instruments’ AM62x for cost-sensitive embedded products that require low power, high performance and a GPU. The VAR-SOM-AM62 runs on 1.4 GHz Quad-core Cortex-A53 AM625x with 400MHz Cortex-M4F and additional 333 MHz PRU real-time co-processors. It offers rich connectivity options like a camera interface, dual LVDS display, certified dual-band Wi-Fi, BT/BLE 5.2, 3x CAN bus, dual USB, and dual GbE.
In a performance benchmark, Variscite compared three of its modules with different CPU performance, pricing, power efficiency level and ML options.
|
SoM |
Applications Processor |
Average Inference Time (ms) |
Average Video (FPS) |
Remarks |
Starting Price Point |
||||||
|
VAR-SOM-MX93 |
2 × Cortex-A55 1.7GHz |
4.9 |
30 |
Utilizing Ethos-U65 NPU |
$39 |
||||||
|
VAR-SOM-MX8M-NANO |
4 × Cortex-A53 1.5GHz |
300 |
1 |
No NPU |
$44 |
||||||
|
VAR-SOM-MX8M-PLUS |
4 × Cortex-A53 1.8GHz? |
2.9 |
30 |
Using NPU |
$62 |
In internal benchmarking, running video-based object classification tests, the VAR-SOM-MX93 utilizing the Ethos-U65 NPU performed with an average inference time of 4.9 ms and an average video throughput of 30 fps on a 640×480 input video. Running the same tests on the VAR-SOM-MX8M-NANO with no NPU demonstrated an average inference time of 300 ms and average video throughput of 1 fps. Finally, the VAR-SOM-MX8M-PLUS uses its NPU yields an average inference time of 2.9 ms and an average video throughput of 30 fps.
“Our newest modules represent the next generation of Variscite’s technology and provide state-of-the-art platforms for robust, reliable, power efficient and cost-effective embedded computing devices,” said Ofer Austerlitz, VP Business Development and Sales of Variscite. “We look forward to presenting show attendees the module’s capabilities.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
