NVIDIA Tegra Architecture

NVIDIA Jetson Nano: A Feature-Packed Arm Developer Kit For $99 USD
With this low-cost Jetson board, the Nano is using a Tegra chip similar to what was found in the Jetson TX1 a few years back. This Tegra X1 SoC has a quad-core Cortex-A57 processor and 128-core NVIDIA Maxwell graphics... Not nearly as interesting as the X2 or AGX Xavier, but still not bad considering the SoCs usually found in sub-$100 Arm developer boards.
...
The Jetson Nano also offers 4GB of LPDDR4 memory, Gigabit Ethernet, 12 MIPI lanes, four USB ports, and can drive up to two simultaneous displays.
Unlike the higher-end Jetson boards featuring eMMC storage, the Jetson Nano relies upon a microSD card for storage. The connectivity on the developer kit includes four USB 3.0 Type-A ports, HDMI 2.0, DisplayPort 1.2, 40-pin header, MIPI CSI camera connector, micro-SD slot, M.2 WiFi slot, and Gigabit Ethernet. That's one of the shortcuts on this board is there is no integrated WiFi but does require an external card if you are interested in wireless connectivity.

embed.php

https://www.phoronix.com/scan.php?page=article&item=nvidia-jetson-nano&num=1


JetBot Is NVIDIA's Newest DIY Robot: Open-Source, Ubuntu-Powered, Built Around The Jetson Nano
NVIDIA won't be selling the Jetbot as a retail product, but it's something you can build yourself with the Jetbot Nano. NVIDIA will be providing detailed instructions and parts lists on GitHub along with all of the necessary software resources.
https://www.phoronix.com/scan.php?page=article&item=nvidia-jetson-jetbot&num=1
 
Miracles happen: Michael has found that there are different modes on TX2! Now he has to find what the fastest mode is and for MT tasks (hint: it's neither Max-Q or Max-P [and there are 3 Max-P modes]). And neither Max-Q or Max-P is the fastest for GPU tests.

The problem is not that he got it wrong several times, the problem is that he denies being wrong and that certainly reinforced my doubts about any benchmark he posts.

Sorry for the rant...
 
Miracles happen: Michael has found that there are different modes on TX2! Now he has to find what the fastest mode is and for MT tasks (hint: it's neither Max-Q or Max-P [and there are 3 Max-P modes]). And neither Max-Q or Max-P is the fastest for GPU tests.

The problem is not that he got it wrong several times, the problem is that he denies being wrong and that certainly reinforced my doubts about any benchmark he posts.

Sorry for the rant...

Is that exposed in the driver he benchmarked?
 
According to Chrome, my Nexus 9 has hardware acceleration for VP9. It's listed in chrome://gpu. Kepler era SOC with that eh? You need the last Maxwell chips like GM206 to get that on their PC products right?
 
It's funny I don't find vp9 hardware acceleration for VP9 for K1. I just found this : "One thing Tegra K1 won't be able to do is to decompress Google's new low-bandwidth 4K video codec, VP9, in hardware. The codec just wasn't available early enough, Wuebbling said. The new chip will have to use a combination of hardware and software for VP9;"
 
Screenshot_20190609-081117.png


However when I try watching VP9 YouTube videos it seems barely able to play 1080p30 without dropping frames and it gets hot. 1440p drops lots of frames. So yeah seems like it's being done on the CPU.
 
Last edited:
NVIDIA Launches Jetson Xavier NX As 70x45mm 10~15 Watt "AI Supercomputer"
November 6, 2019
NVIDIA announced today the newest member of the Jetson family: the Xavier NX as "the world's smallest supercomputer" coming in at smaller than the size of a credit/debit card. This mini supercomputer can deliver 21 TOPS for modern AI workloads while consuming less than 10 Watts or optionally a higher-performance 15 Watt mode.
image.php

The Jetson Xavier NX is powered by a low-power version of the Tegra Xavier SoC. The Jetson Xavier NX offers six NVIDIA Carmel ARMv8.2 cores, a 384-core Volta GPU with 48 Tensor cores, dual NVDLA engines, 8GB of LPDDR4x memory, 16GB eMMC, Gigabit Ethernet, USB 3.1, and other functionality all off a 70x45 mm PCB and running off a +5V line.

https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Jetson-Xavier-NX

Edit: Price for Xavier NX is $399.
 
Last edited:
I'm guessing these are very low binned chips, and/or Xavier was clearly designed for 30W operation.
General efficiency (power-per-processing-throughput) is lower for Xavier NX than it is for Xavier AGX.
In the 15W case, a 50% cut in power consumption results in 55% lower CPU throughput, 42% lower GPU throughput, 63% lower memory bandwidth and 35% lower tensor throughput.
 
Found this amusing and interesting. Shopping around for a toaster oven/air fryer and discovered the awesome June oven which has a list of full specifications, including an Nvidia Tegra K1 processor.
 
Found this amusing and interesting. Shopping around for a toaster oven/air fryer and discovered the awesome June oven which has a list of full specifications, including an Nvidia Tegra K1 processor.
Surely if you want to cook you'd want the Tegra K1 Denver. It could totally fulfill that "warming drawer" role.
 
Last edited:
Aren't the
Shopping around for a toaster oven/air fryer
I was reading some of the reviews and this doesn't have a self-cleaning feature, though I'm not sure that's the norm. Aren't these appliances usually difficult to clean?
 
Real-Time Object Detection in 10 Lines of Python on Jetson Nano
In this hands-on tutorial, you’ll learn how to:
  • Setup your NVIDIA Jetson Nano and coding environment by installing prerequisite libraries and downloading DNN models such as SSD-Mobilenet and SSD-Inception, pre-trained on the 90-class MS-COCO dataset
  • Run several object detection examples with NVIDIA TensorRT
  • Code your own real-time object detection program in Python from a live camera feed.
You can then use this 10-line Python program for object detection in different settings using other pre-trained DNN models. The code for this and other Hello AI world tutorials is available on GitHub.


https://news.developer.nvidia.com/realtime-object-detection-in-10-lines-of-python-on-jetson-nano/
 
Last edited:
Back
Top