NVIDIA Jetson Orin Nano Super: What the 67 TOPS Update Means

The Jetson Orin Nano Super is the NVIDIA Jetson Orin Nano module running the higher-power Super modes that NVIDIA released in JetPack 6.2, which lift the 8GB module from 40 to 67 TOPS of AI performance on the same silicon. It is not a new chip: the GPU, CPU and memory run at higher clocks inside a higher power limit, and the same change reaches the Orin NX.

The name covers two things. NVIDIA launched the Jetson Orin Nano Super Developer Kit in December 2024 at a lower launch price than the original kit, and in January 2025 it brought the same modes to every production Jetson Orin Nano and Orin NX module through a software update. A design already in the field can usually gain the extra performance with a reflash, provided its carrier board and cooling can take the extra power. Our explainer on Jetson Super mode covers the power modes themselves.

This guide is the full reference: original and Super figures for all four Orin Nano and Orin NX modules, how to enable the modes, how much power and cooling they need, what 67 TOPS lets you run, and how to take a design from the developer kit to a deployable industrial system. For the wider platform, start with our overview of NVIDIA Jetson.

NVIDIA published new ratings for all four modules in the Orin Nano and Orin NX families when it added the Super modes. The table shows the original figures and the Super figures side by side, from NVIDIA’s published specifications.

All TOPS figures are NVIDIA’s sparse INT8 peaks, which assume a model pruned for structured sparsity; dense INT8 is about half. Our explainer on what TOPS means sets out why the headline number and real model throughput differ. In practice the uplift on a real network is smaller than 1.7× but still large, often well over 50 percent on GPU-bound work.

The Orin Nano gains the most because its original power limit held it furthest below what the silicon could do. The Jetson Orin NX already ran its memory at full speed, so its gain comes from GPU clock and the higher power ceiling, with its deep learning accelerators adding to the total.

What changed inside the module

Nothing physical. The Orin Nano 8GB keeps its 1,024-core NVIDIA Ampere GPU with 32 Tensor Cores, its six-core Arm Cortex-A78AE CPU and its 8 GB of 128-bit LPDDR5. What JetPack 6.2 changes is the operating point: a higher GPU clock, a higher CPU clock, the memory running at its full rated speed, and a power budget large enough to sustain them. That is why the same carrier boards, connectors and software apply.

With the Super modes enabled, the two Orin Nano modules line up as follows. These are module figures; the carrier board decides which interfaces reach a connector.

Two rows catch people out. The Orin Nano has no hardware video encoder, so recording or re-streaming H.264 and H.265 runs on the CPU cores, and it has no deep learning accelerator, so every model runs on the GPU. Our guide to CUDA cores vs Tensor Cores explains where the AI figure comes from, and MIPI CSI-2 covers the camera lanes.

The SO-DIMM connector is shared with the Orin NX, so a carrier designed for both can move up a tier without a board change, as long as its power design and board support package cover the higher-power module.

Yes. Super mode is software, so every Orin Nano and Orin NX module that has shipped can run it. What it needs is a JetPack 6.2 or later image for the specific carrier board, with the board configuration that includes the Super power modes. Our guide to how JetPack versions decide what a Jetson module can run explains how the JetPack release, the Jetson Linux version and the carrier vendor’s support fit together.

The carrier vendor’s software matters as much as NVIDIA’s. The board support package carries the device tree and power settings for that carrier, which is why a third-party board needs its own JetPack 6.2 release before the Super modes appear. Our explainer on the board support package covers what it contains, and Linux for Tegra explains the operating system underneath.

Major JetPack upgrades are a reflash rather than a package update, because they change the kernel and Ubuntu release. Plan the move to JetPack 6.2 as a validated image change: rebuild the application against the new CUDA and TensorRT versions, retest it, then roll it out.

The module itself draws up to 25 W in its top fixed mode, against 15 W before. The whole system draws more, because the carrier board, NVMe drive, USB devices and cameras all take power on top of the module. NVIDIA lists these power modes with JetPack 6.2:

A worked power budget

Catalog figures show how far system power sits above the module mode. Sintrones measured its IBOX-600 Orin Nano 8GB build at about 29 W from a 24 V input with the module in its 15 W mode, and Yuan rates the Pandora Nano 8GB Super at a maximum of 43.5 W. A reasonable planning figure for an Orin Nano system in a Super mode is therefore 35–45 W at the input, depending on storage, USB loads and any PoE cameras, with the supply and wiring sized above that.

The developer kit avoids the question with its own DC adapter. Forum questions about powering the kit from a battery or a robot’s supply usually come down to the same point: the input has to carry the module’s top mode plus everything else on the board, with headroom for start-up surges.

Heat and fanless enclosures

Every extra watt becomes heat the enclosure has to remove. A fanless system sized for a 15 W module can hold the Orin Nano at its old limit, or let it throttle under sustained load, which gives back much of the Super gain. Our Jetson thermal design guide shows how to size a heatsink and enclosure for a given power mode, and thermal throttling explains what happens when the cooling falls short.

Fanless Orin Nano computers in our range are typically rated from −25 °C up to between 55 and 80 °C, depending on the enclosure and whether airflow is assumed. Check whether the datasheet’s upper limit applies at the Super mode you plan to ship, and treat the inside of a cabinet, not the room, as the ambient temperature.

More of everything the Orin Nano already ran, plus small generative models that were borderline before. The higher GPU clock helps convolutional vision models, and the extra memory bandwidth helps large language models and vision language models, which are often limited by how fast weights can be read rather than by compute.

NVIDIA’s own benchmarks for the Super release show generative AI throughput rising by up to around 1.7× on the same module, and NVIDIA’s Jetson AI Lab publishes containers for small language models, speech and vision language models that run on the 8GB module. Most production gains come from the toolchain rather than the mode: TensorRT compilation and INT8 quantisation typically matter as much as the extra clock.

For language models, memory sets the limit before TOPS do. A 4-bit model in the 7–8 billion parameter range needs roughly 4–5 GB for its weights alone, which leaves little room on an 8 GB module that also runs the operating system and the camera pipeline. Our explainer on running a local LLM covers the trade-offs, and the Orin NX 16GB is the usual next step when the model or the context grows.

Super narrows the gap to the next module but does not close it. At 67 TOPS, the Orin Nano 8GB is about 40 percent of the Orin NX 16GB’s 157 TOPS, with half the memory and no hardware encoder. Our Orin Nano vs Orin NX explainer goes deeper on the two smaller modules, and the Jetson AGX Orin sits above both.

The choice usually follows the camera count and the largest model. Where the workload sits between tiers, a carrier that takes both the Orin Nano and Orin NX keeps the decision open until benchmarks on the real model settle it. Our comparison of Orin NX vs AGX Orin covers the next step up, and the guide to choosing a Jetson module walks the whole range, including Jetson Thor.

The developer kit is the fastest way to benchmark the Super modes and build an application, but it is not what ships. The kit is NVIDIA’s reference carrier with an Orin Nano 8GB module and a fan-cooled heatsink, designed for a bench. We cover the module versus a developer kit in full; the table shows what changes on the way to a product.

A deployed system pairs the bare module with a Jetson carrier board built for the job, in an enclosure that can dissipate 25 W at the site’s hottest temperature. Our guide to choosing a Jetson carrier board covers off-the-shelf, adapted and custom carriers, and the industrial temperature range explainer sets out what the common ratings mean.

Lifecycle is part of the case for Orin Nano. NVIDIA lists the Orin modules in production into the early 2030s, which suits a product that has to ship for years; our answer on how long NVIDIA Jetson is supported sets out the published position. The Orin Nano is also not the same product as the original Jetson Nano: the original Jetson Nano is a Maxwell-generation module that stops at JetPack 4.6, while the Orin Nano is its successor on the Ampere architecture.

Orin Nano systems rated for Super mode

These systems from our range list Super mode support for the Orin Nano in their manufacturers’ specifications. Ratings are the manufacturers’ own.

We configure, image and test each system with the module, storage and JetPack image fitted, and set the power mode to suit the enclosure and the site, so it ships ready to run in the mode it was validated in.