From Upgrade to Purchase: ORIN SUPER Board Performance Enhancements and Buying Advice

Hello everyone,As we all know. Everyone is already aware of the news that Nvidia has upgraded its ORIN board to ORIN SUPER through software upgrade

Today we will test the actual performance improvement

LLM performance improvement on Jetson Orin Nano SUPER

VLM performance improvement on Jetson Orin Nano SUPER

Visual Transformer performance improvement on Jetson Orin Nano SUPER

First, Let’s look at the performance improvement.

According to the official blog, wecan see that the Generative AI performance of Jetson Orin Nano official kit has increased by 1.7 times. Generative AI performance of Jetson Orin modules has increased by 2 times.

Click here to check NVIDIA official blog.

According to the official computing power table, as shown below. 

We can know that

Jetson Orin Nano 4GB increased from 20 TOPS to 34 TOPS, increased by about 70%;

Jetson Orin Nano 8GB increased from 40 TOPS to 67 TOPS, increased by about 67.5%;

Jetson Orin NX 8GB increased from 70 TOPS to 117 TOPS, increased by about 67%;

Jetson Orin NX 16GB increased from 100 TOPS to 157 TOPS, increased by about 57%.

Second, we compared the four boards(Jetson ORIN NANO 4GB, Jetson ORIN NANO 8GB, Jetson ORIN NX 8GB,  Jetson ORIN NX 16GB) before and after the upgrade.

We ran the same program before and after the upgrade and recorded the specific changes. The test content covered large language models, visual language models and visual models.

The results show that using the super board to run a large language model, can bring performance gains.

Similarly, using the super board to run the visual language model. The performance gain of some boards can be as high as 2.5 times.

We also run some common visual models, can also see different performance improvements.

Overall, the four boards have performance gains and performance improvements that are consistent with official publicity.

Next. We will provide some purchase suggestions based on the price

It should be noted that NVIDIA official price only provides a rough reference.

When actually purchasing, you also need to consider tariffs, freight. And price fluctuations caused by changes in market supply and demand.

Before ORIN NANO is upgraded to SUPER, we can see the official prices of NVIDIA's different development kits and core modules as shown in the table below.

But now, with performance improvements and unchanged hardware. The market guide price of the Jetson Orin Nano Super official kit has dropped from $499 to $249. 

The prices of other core modules remain unchanged.

Except for the Jetson Orin Nano SUPER official kit, other core modules need to bear the cost of purchasing an additional carrier board.

Everyone would think that the Jetson Orin Nano SUPER 8GB official kit has the best cost performance. But due to the insufficient inventory of the Jetson Orin Nano SUPER 8GB official kit.

Actual purchases may require a 30%~50% premium to be able to buy. And insufficient inventory and unstable purchase prices will affect the implementation of subsequent projects. So the advantages of the official kit are not really great.

If you can buy the Jetson Orin Nano 8GB official kit at a lower premium. It is a good choice. 

If the price of the official kit increases by more than 40%. Users can consider the SUB version developer kit with official core modules.

Got Yahboom ORIN NANO 4GB/8GB SUPER Developer Kit

About Jetson Orin NX, they provide more than twice the performance improvement of Jetson Orin Nano, Filling the gap of high-level computing power, suitable for tasks that require high computing power for educational training/lightweight robots.

Such as basic tasks such as visual image processing, choosing Jetson Orin Nano 4GB can already meet the needs. If you want to perform depth autopilot, deploy large language models, visual language models, and multimodal models. It is recommended to use Jetson Orin Nano 8GB and Jetson Orin NX.

Got Yahboom ORIN NX 8GB/16GB Developer Kit

The higher the computing power, the faster the frame rate and processing speed.

Therefore, when making a selection. You can consider that according to your specific needs and budget, ensure that the progress of the project will not be restricted due to insufficient computing power, also avoid unnecessary computing power vacancy and resource waste.