NVIDIA Jetson AGX Thor Developer Kit

Jargon buster

Plain-language definitions for the technical terms used above.

Blackwell GPU architecture

A generation of NVIDIA graphics processor design aimed at running very large AI and compute workloads. It matters because software performance, AI model support and power use depend heavily on the GPU architecture inside the developer kit.

DisplayPort

A digital display interface used to connect monitors and other video displays. It matters because the supported DisplayPort version affects display resolution, refresh rate and monitor compatibility.

FP4

A very low-precision 4-bit number format used to run some AI models faster and with less memory. It matters because FP4 performance figures are mainly relevant to compatible AI workloads, not general-purpose computing.

GbE

Gigabit Ethernet, a wired networking speed measured in billions of bits per second. It matters because high-resolution cameras, sensors and edge AI systems may need faster Ethernet links than standard home networking provides.

H.265

A video compression standard, also called HEVC, that reduces video file size and bandwidth compared with older formats. It matters for vision and AI projects because hardware H.265 encode/decode support can handle many camera streams more efficiently.

HDMI

HDMI is a common digital video and audio connection used by computers, media players, and many displays. If a display kit has HDMI input, it is usually much easier to test with a single-board computer because it can act like a normal monitor.

Headers

Rows of metal pins used to plug a module into a breadboard or connect it with jumper wires. Pre-soldered headers make the module easier to use straight away without needing to solder the pins yourself.

JTAG

JTAG is a hardware debugging and programming interface used to inspect and control chips at a low level. It matters for advanced development because it can help diagnose firmware problems that are hard to see through normal serial output.

LiDAR

A sensing method that uses laser light to measure distance and build a 3D picture of the surroundings. It matters for robotics and autonomous systems because LiDAR data can help with mapping, navigation and obstacle detection.

LPDDR5X

A high-speed, low-power type of memory commonly used in advanced mobile and embedded computers. It matters because AI and vision workloads often need both large memory capacity and high memory bandwidth to run smoothly.

M.2

A compact edge-connector format commonly used to plug small modules into a carrier board without soldering. On this product it is the physical connector used by the MicroMod system, so compatibility with the matching processor board is important.

Multi-Instance GPU (MIG)

A feature that can split one physical GPU into several isolated GPU instances for separate workloads. It matters when running multiple AI or robotics tasks at once, because each task can be given its own share of GPU resources.

NVMe

A high-speed storage standard commonly used by modern SSDs. NVMe support matters if you want faster storage for large AI models, video files or operating system images than a typical microSD card can provide.

PCIe Gen5

A high-speed internal expansion bus used to connect devices such as NVMe storage, network cards and accelerator modules. Gen5 offers more bandwidth than earlier PCIe versions, which matters for moving large amounts of sensor or AI data quickly.

QSFP28

A compact high-speed networking connector often used for 25 GbE or 100 GbE links. It matters because it allows very fast camera, sensor or network data transfer, but requires compatible cables and network equipment.

RJ45

The common plug and socket style used for wired Ethernet network cables. If a board has an RJ45 connector, you can usually plug it into standard Ethernet cabling without making a custom connector.

RTC

A Real-Time Clock keeps track of time even when the main processor is asleep or powered down, usually with a small backup battery. It matters for data logging and tracking projects that need accurate timestamps.

Tensor Cores

Specialised processing units inside some GPUs that accelerate the matrix maths used in machine learning. They matter for choosing an AI computer because many neural-network tasks run much faster when software can use them.

TFLOPS

A measure of how many trillion floating-point calculations a processor can perform per second. It helps compare AI and graphics compute capability, but the real speed you get depends on the model, precision format and software support.

USB-C

A modern reversible USB connector used for power and data connections. On this product it matters because it can connect directly to a computer as well as to a microcontroller project.