Uploaded September 2025 | Updated September 2026, 2 weeks ago
In this episode of Product of the Week, Robin takes a look at the Pocket AI, a portable AI inference system built around Nvidia RTX GPU technology. With up to 100 TOPS of performance, 2048 CUDA cores, 64 Tensor cores, and 16 RT cores, the Pocket AI is designed for serious local AI applications. It connects over Thunderbolt 3, supports Windows and Linux, and runs at a 25W power envelope, making it a compact yet capable alternative to larger GPUs.
Robin explores its potential uses, from autonomous vehicles to drone surveillance, AI assistants, robotics, and industrial anomaly detection. With 4GB GDDR6 memory, 96 GB/s bandwidth, and clock speeds up to 1335 MHz, the Pocket AI offers desktop-class performance in a device that fits in your hand!
📌 Chapters:
0:00 - Introducing the Pocket AI Accelerator
0:36 - Why Pocket AI Was Built
1:02 - Specs and Performance Overview
1:50 - TOPS, CUDA, Tensor, and RT Core Breakdown
2:48 - Memory, Bandwidth, and Clock Speeds
3:22 - Power Consumption and USB PD
3:38 - Thunderbolt 3 and OS Compatibility
3:56 - Size and Mounting Options
4:00 - Project Idea: Drone Surveillance
4:15 - Project Idea: Offline AI Assistant
4:33 - Project Idea: Robotics and Gesture Recognition
4:51 - Project Idea: Industrial Anomaly Detection
5:10 - Pocket AI as a Portable Data Center
💡 Key Features:
💡Nvidia RTX GPU architecture
💡Up to 100 TOPS at INT8
💡2048 CUDA cores, 64 Tensor cores, 16 RT cores
💡4GB GDDR6 memory, 96 GB/s bandwidth
💡Base clock 435 MHz, boost clock 1335 MHz
💡25W TDP with USB PD power input
💡Thunderbolt 3 connection
💡Windows and Linux support
💡Compact 110 x 76 x 25 mm case
👉 Buy yours now from the Electromaker Store:
electromaker.io/shop/product/egpu-with-nvidia-quadro-rtx-a500-embedded-graphics
❤️ Support Us!
Stay updated with the latest products, projects, and news at Electromaker.io
electromaker.io
🌍 **Stay Connected with Electromaker!**
youtube.com/channel/UCiMO2NHYWNiVTzyGsPYn4DA?sub_confirmation=1
facebook.com/electromaker.io
twitter.com/ElectromakerIO
linkedin.com/company/electromaker
In this episode of Product of the Week, Robin takes a look at the Pocket AI, a portable AI inference system built around Nvidia RTX GPU technology. With up to 100 TOPS of performance, 2048 CUDA cores, 64 Tensor cores, and 16 RT cores, the Pocket AI is designed for serious local AI applications. It connects over Thunderbolt 3, supports Windows and Linux, and runs at a 25W power envelope, making it a compact yet capable alternative to larger GPUs.
Robin explores its potential uses, from autonomous vehicles to drone surveillance, AI assistants, robotics, and industrial anomaly detection. With 4GB GDDR6 memory, 96 GB/s bandwidth, and clock speeds up to 1335 MHz, the Pocket AI offers desktop-class performance in a device that fits in your hand!
📌 Chapters:
0:00 - Introducing the Pocket AI Accelerator
0:36 - Why Pocket AI Was Built
1:02 - Specs and Performance Overview
1:50 - TOPS, CUDA, Tensor, and RT Core Breakdown
2:48 - Memory, Bandwidth, and Clock Speeds
3:22 - Power Consumption and USB PD
3:38 - Thunderbolt 3 and OS Compatibility
3:56 - Size and Mounting Options
4:00 - Project Idea: Drone Surveillance
4:15 - Project Idea: Offline AI Assistant
4:33 - Project Idea: Robotics and Gesture Recognition
4:51 - Project Idea: Industrial Anomaly Detection
5:10 - Pocket AI as a Portable Data Center
💡 Key Features:
💡Nvidia RTX GPU architecture
💡Up to 100 TOPS at INT8
💡2048 CUDA cores, 64 Tensor cores, 16 RT cores
💡4GB GDDR6 memory, 96 GB/s bandwidth
💡Base clock 435 MHz, boost clock 1335 MHz
💡25W TDP with USB PD power input
💡Thunderbolt 3 connection
💡Windows and Linux support
💡Compact 110 x 76 x 25 mm case
👉 Buy yours now from the Electromaker Store:
electromaker.io/shop/product/egpu-with-nvidia-quadro-rtx-a500-embedded-graphics
❤️ Support Us!
Stay updated with the latest products, projects, and news at Electromaker.io
electromaker.io
🌍 **Stay Connected with Electromaker!**
youtube.com/channel/UCiMO2NHYWNiVTzyGsPYn4DA?sub_confirmation=1
facebook.com/electromaker.io
twitter.com/ElectromakerIO
linkedin.com/company/electromaker
