What is GTC?

GTC is NVIDIA's international conference that brings together the latest GPU (Graphics Processing Unit) technology, which has become indispensable as a computational resource for building AI.

You can learn how the field of AI and its related technologies are currently being used, and how they will be used in the future, such as the latest GPUs, development cases using AI, and research cases of AI itself. increase.

It is truly a must-see event for everyone involved in AI.

Atmosphere of the venue

The venue is the Convention Center in San Jose, California.
This place is wide, so you don't have to be full of people and not be able to move.
NVIDIA's sponsors and GTC exhibitors are introduced from the ceiling, and they are all outstanding.

However, the opening keynote of GTC by NVIDIA CEO Jen Sun Huang held at another venue (San Jose State University) was full of such crowds and people's heat (+ the sun was strong on this day...). It was painful.

I was surprised that there were so many people.

In addition, coffee (the brand of the cup was Su*ba) is all-you-can-drink, and dispensers are placed here and there.
At night, you can have a proper meal while going around the exhibition booths. (I wish I had taken a picture...)
In addition, you can get special backpacks and various conversion cables, and it's a fun event that makes you happy.

Keynote point

I will skip from the beginning, but the attention is the announcement of "Jetson Nano".

The performance of Jetson TX1, which has built an era as an AI module for edges, is now even smaller and (quite) cheaper.
Jetson Nano in the palm of Mr. Jen Sun Huang's hand... it's pretty small!

In addition to the introduction of modules, "Kaya", which combines Jetson Nano with a stereo camera and depth sensor, and AGV "Carter", which is equipped with Jetson AGX Xavier, were introduced as reference models for self-propelled robots.
There was also a demonstration at the exhibition booth, so I will introduce it later.

Right after Keynote ended, there was a sales event for the developer kits, and there was a huge queue. I managed to get one for evaluation. . .
I may write an article about AI implementation using Jetson Nano at a later date, so please look forward to it.
For more information on Jetson, please see the link at the bottom of the blog!

Undecided name as an analysis machine? There was an introduction of a workstation for data scientists (RTX 8000 x2) and a T4 server specialized for data science.
The Turing architecture (the latest GPU) will be used for analysis in earnest, so AI learning efficiency will continue to improve!

There were also many announcements about toolkits (software) for AI development.

The AI development toolkit "CLARA" for radiology covers all the latest AI development flows: annotation, trained models and transfer learning support, and implementation.
I definitely want to use this... I feel like it can be applied to the development of appearance inspection systems if you take a broader view of it for images! Dreams come true.

Also, there was an announcement of a library called CUDA-X that accelerates development by unifying various CUDA libraries that have existed separately, such as RAPIDS and cu○○.
By simplifying the divisions that were difficult to understand, you will be able to get the parallel operations you want without having to pay attention to the details!
I have thoroughly researched RAPIDS, so I will introduce it later.

exhibition booth

At the GTC 2019 exhibition booth, AI startup companies, AI solutions of each exhibitor and sponsor, and NVIDIA products that were also in Keynote were introduced.

Among the startup companies, there were several companies that introduced solutions related to AI development and image annotation.
It was interesting to see the introduction of a tool that allows annotation simply by specifying a point cloud, as a means of reducing the human burden of annotation for region segmentation tasks.

At the exhibition booth, "Kaya" and "Carter" introduced in the above Keynote were demonstrating.

"Kaya" was cute playing soccer! Will it appear in RoboCup someday?

"Carter" was carrying the sweets that the picking robot put on.
While grasping its own position with the sensor, it seemed to detect what was reflected by the camera using Superpixel.

OmniSci's booth was also popular.

This was also featured extensively in Keynote as a server-side big data analysis tool.
(In Keynote, he used Wifi access point hotspot analysis as an example to visualize the confusion matrix, data pipeline, heatmap, histogram for each region, and real-time time series data!!)

At the exhibition booth, while displaying the difference between the reference (CPU) of the racing game and the player's operation parameters and laps (the model of the course was also displayed, but does it have a modeling function?), The information calculated from the collected data was also displayed in real time.

In addition, I was embarrassed to play a game while being watched by people, so I gave up on participating...

Musashi Seimitsu Industry Co., Ltd. exhibited the cooperative operation of the picking robot and visual inspection, and this was to detect the dents of the target object by visual inspection with a camera and reflect it in the control of the picking robot.
The Edge Computing Box made by Musashi Seimitsu Industry Co., Ltd. was also on display, and there was a crowd!

So, this time we mainly introduced Keynote and the exhibition booth!

Next time, I would like to take up the interesting things in the sessions and cases.

The Jetson Nano development kit introduced here can be purchased from the link below.
You can also get detailed information on other NVIDIA GPU products, so please take a look!