$199+ NVIDIA Jetson Orin Nano system-on-module delivers up to 40 TOPS

NVIDIA Jetson Orin Nano system-on-module (SoM) is an update to the Jetson Nano entry-level Edge AI and robotics module that delivers up to 40 TOPS of AI performance, meaning it’s up to 80 times faster than the original module.

The new SoM features an hexa-core Arm Cortex-A78AE processor, an up to 1024-core NVIDIA Ampere architecture GPU with 32 Tensor cores, up to 8GB RAM, and the same 260-pin SO-DIMM connector found in the Jetson Orin NX modules.

NVIDIA Jetson Orin Nano
Jetson Nano Orin with 8GB RAM (left) and 4GB RAM (right)

Two versions are offered with the following specifications:

Jetson Orin Nano 4GB
Jetson Orin Nano 8GB
CPU
6-core Arm Cortex-A78AE v8.2 64-bit CPU @ 1.5 GHz with 1.5 MB L2 + 4 MB L3
GPU
512-core NVIDIA Ampere GPU @ 625 MHz with 16 Tensor cores
1024-core NVIDIA Ampere GPU @ 625 MHz with 32 Tensor cores
AI performance
20 Sparse TOPs | 10 Dense TOPs
40 Sparse TOPs | 20 Dense TOPs
Memory
4GB 64-bit LPDDR5 34 GB/s
8GB 128-bit LPDDR5 68 GB/s
Storage
Supports external NVMe
Video Decode
1x 4K60 (H.265) | 2x 4K30 (H.265) | 5x 1080p60 (H.265) | 11x 1080p30 (H.265)
Video Encode
1080p30 supported by 1-2 CPU cores (i.e. software encoding)
Display

1x 4K30 multimode DisplayPort 1.2 (+MST)
1x embedded DisplayPort 1.4
1x HDMI 1.4
Camera
Up to 4x cameras (8x through virtual channels)
8 lanes MIPI CSI-2 D-PHY 2.1 (up to 20 Gbps)
Networking
Gigabit Ethernet
USB
3x USB 3.2 Gen2 (10 Gbps)
3x USB 2.0
PCIe
1 x4 + 3 x1 PCIe Gen3, Root Port, & Endpoint
Other I/Os
3x UART, 2x SPI, 2x I2S, 4x I2C, 1x CAN, DMIC and DSPK, PWM, GPIOs
Power modes
5W or 10W
7W or 15W
Dimensions
69.6 mm x 45 mm 260-pin SO-DIMM connector

Jetson Orin Nano Block Diagram

That means the Jetson Orin family has now six modules ranging from 20 TOPS to 275 TOPS. There’s no specific development kit for the Jetson Orin Nano SoM since it can be emulated on the NVIDIA Jetson AGX Orin developer kit, and supported by the JetPack 5.0.2 SDK based on Ubuntu 20.04.

NVIDIA has tested some dense INT8 and FP16 pre-trained models from NGC and a standard ResNet-50 model on the new module and compared the results against the earlier generation entry-level modules such as the Jetson Nano and Jetson TX2 NX.

Jetson Nano vs TX2 NX vs Jetson Orin Nano

That’s a significant increase in performance, as well as in price since the Jetson Orin Nano series production modules will be available in January for $199 for the Jetson Orin Nano 4GB and $299 for the Jetson Orin Nano 8GB. Those are unit prices for 1K orders. More details may be found on the announcement and developer website.

Share this:

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress

ROCK Pi 4C Plus
Subscribe
Notify of
guest
The comment form collects your name, email and content to allow us keep track of the comments placed on the website. Please read and accept our website Terms and Privacy Policy to post a comment.
9 Comments
oldest
newest
Joe
Joe
1 year ago

Ouch, such performance will run circles around the Rk3588 based boards, not only in CPU performance, but also the additional powerful npu. Just wonder if I can put it in my original Jetson nano developer carrier board? They only say it’s compatible with the Orin nx, which may be different from the original nano?

tkaiser
tkaiser
1 year ago

> run circles around the Rk3588 based boards, not only in CPU performance

The A78AE are only clocked at 1.5GHz and there are just six of them (most probably the result of binning/QA at the factory).

If if it’s not about GPGPU/NPU single-threaded performance is lower compared to RK3588’s A76 @ ~2.3 GHz and with most CPU-bound workloads even multi-threaded could be lower.

Joe
Joe
1 year ago

Have you seen benchmarks? I’ve estimated about 20-30% better performance per core of A-76 vs A-78, although the clock speed may eat that up. But then there are six vs four performance cores. Also possibly faster memory.
Maybe I should withdraw “run circles”, but the nano certainly is interesting, especially when running machine learning models.

tkaiser
tkaiser
1 year ago

> Have you seen benchmarks?

Both times comparing the 12-core Orin at 2.2 GHz with RK3588 at 2350/1800 MHz: Geekbench numbers here, 7-zip, tinymembench, ramlat, openssl and cpuminer there (ARMv8 Crypto Extensions perform identical on A72-A78 and scale linearly with clockspeeds).

Joe
Joe
1 year ago

That does look very good for the Rk3588. Given that the 12-core Orin is in a completely different price league. Now my board just needs to be shipped.

bernstein
bernstein
1 year ago

as always, thanks! awesome

tkaiser
tkaiser
1 year ago

> Have you seen benchmarks?

Sure, Results.md in my sbc-bench repo list 7-zip, tinymembench, ramlat, openssl and cpuminer scores. And this here is Geekbench.

Always comparing the 12-core Orin at 2.2 GHz with RK3588 at 2350/1800 MHz. I wouldn’t be too surprised if memory performance with little Jetson Orin Nano is (significantly) lower.

IM2323
IM2323
1 year ago

Only software encode? very unlike previous Jetson series…

David Jashi
1 year ago

This looks to me as a software/Jetpack issue, as I cannot imagine, why GPU cannot be used for it.

Khadas VIM4 SBC