Performance speedup: Jetson TX2 vs AGX Xavier

Author: Fyodor Serzhenko

Imaging applications benefit from the latest NVIDIA mobile GPUs: Jetson TX2 and AGX Xavier. Nevertheless, general benchmarks can't answer the question about performance speedup for the latest NVIDIA Jetson hardware. Anyway, this is very practical issue for many imaging applications, including aerial imaging, UAV, robotics, self-driving cars, etc. To provide you with real numbers, we've done comparative studies with Fastvideo SDK, which has lots of image processing modules on GPU for camera applications. And this SDK is compatible with full line of Jetson hardwdare.

 

Jetson TX2 vs Xavier

 

Hardware comparison: Jetson TX2 vs AGX Xavier

Hardware feature \ Jetson module Jetson TX2 Jetson AGX Xavier
CPU (ARM) 6-core Denver and A57 @ 2 GHz 8-core Carmel ARM CPU @ 2.26 GHz
Memory 8 GB 128-bit LPDDR4 16 GB 256-bit LPDDR4x @ 2133 MHz
Memory bandwidth 58.4 GB/s 137 GB/s
Storage 32 GB eMMC 32 GB eMMC
GPU 256 Core Pascal @ 1.3 GHz 512 Core Volta @ 1.37 GHz
Tensor cores -- 64
Deep Learning Accelerator -- (2x) NVDLA
Vision Accelerator -- (2x) 7-way VLIW Processor
Video encoding (V4L2) (2x) 4K @30
HEVC
(4x) 4Kp60 / (8x) 4Kp30
HEVC
Video decoding (V4L2) (2x) 4K @30
12-bit support
(2x) 8Kp30 / (6x) 4Kp60
12-bit support
PCI-Express lanes 5 lanes PCIe Gen 2
1x4 + 1x1
16 lanes PCIe Gen 4
1x8 + 1x4 + 1x2 + 2x1
Power 7.5W / 15W 10W / 15W / 30W

 

Jetson AGX Xavier is the most powerful system in comparison with previous Jetson modules. It delivers the capability of a desktop GPU workstation in an embedded module under 30W. It has NVIDIA Volta GPU with Tensor Cores, two NVDLA engines and an 8-core 64-bit ARM CPU.

Jetson TX2 is a fast, power-efficient AI computing device, built around an NVIDIA Pascal GPU and loaded with 8 GB of memory with 58.4 GB/s memory bandwidth.

Jetson TX2i is a hardware module for industrial environments. The rugged design, small form factor and power envelope make the Jetson TX2i module ideal for high performance devices such as robots, machine vision and industrial cameras, portable medical equipment, etc.

Jetson TX2 4GB will allow developers to run neural networks with double the compute performance or double the power efficiency of Jetson TX1 at the same price.

How we've done performance tests for Jetson TX2 vs Xavier

We've done time measurements for most frequently used image processing algorithms on GPU like demosaic, resize, denoise, jpeg encoder and decoder, jpeg2000 codec, etc. This is just a small part of Fastvideo SDK modules, though they could be valuable to understand the performance acceleration with Jetson AGX Xavier vs TX2.

We've utilized the same images and the same parameters for testing. Performance boost at AGX Xavier is very important issue, because in many cases of camera applications we could switch from offline to realtime mode of operation. This is also viable for multiple camera systems both on Jetson TX2 and on AGX Xavier.

 

Jetson Xavier vs TX2 benchmarks

 

We can see that performance speedup is in the range of 1.7 - 3 for imaging applications on Jetson Xavier in comparison with TX2. This is impressive boost for practitioners. Quite often the results of raw image processing go further as the input for AI or DL applications, which have also been significantly accelerated by new Volta hardware cores on Jetson AGX Xavier. According to NVIDIA's measurements for AI applications, Jetson AGX Xavier has 20x performance acceleration in comparison with Jetson TX2.

If you are interested to get detailed benchmarks for image processing on NVIDIA Jetson TX2 and AGX Xavier, or you would be interested to get Fastvideo SDK for evaluation, please fill Contact Form below and send us your letter.

Other blog posts from Fastvideo about Jetson hardware and software

Contact Form

This form collects your name and email. Check out our Privacy Policy on how we protect and manage your personal data.