Interactively Explore 20 Billion Records of Ship AIS Data on a Single GPU: HEAVY.AI on the new Nvidia Grace Hopper Superchip

Introduction

NVIDIA’s new superchip architecture is dramatically changing  price- performance for interactive analytics using massive amounts of data.  HEAVY.AI’s customers have long appreciated the power of a GPU native platform that supports interactive analytics for large-scale data.   With the Grace Hopper superchip, the economics  have fundamentally changed.   Now, even larger  data sets can be explored, exposing deeper  insights.

‍

HEAVY.AI recently announced official support for the NVIDIA ARM-based superchips, and created a  demonstration of this capability in partnership with Vultr.  Vultr provides a very attractive price performance option with the Grace Hopper based instances, delivering  unmatched performance.  HEAVY.AI created a demo to highlight the performance this new architecture is capable of.

Why Grace Hopper is a game changer

HEAVY.AI has always exploited the native memory bandwidth that GPU offers to deliver analytics at speeds far beyond the reach of CPU based options.  By bringing data into the GPU memory, HEAVY.AI can process the data very quickly, but in cases where the amount of data required to service a query exceeds the capacity of GPU memory, HEAVY.AI relies on the CPU to process the data.  These CPU based operations traditionally have been significantly slower than operations that

‍

‍

With the introduction of the Grace Hopper superchip and its ultra-fast NVLink interconnect between CPU and GPU, HEAVY.AI can now much more efficiently leverage the full combination of CPU and GPU memory to process large amounts of data by streaming data across the fast bus to GPU.  This dramatically improves price / performance calculations for a HEAVY.AI deployment.  These improvements allow more data to be analyzed without the need to add GPU cards in order to keep all the data in GPU memory.  We have seen massive performance gains that allow the ability to analyze much more data on a single GH200 instance compared to existing implementations that require larger and more expensive A100 class servers.

‍

The charts below show the relative performance of the standard TPC-H benchmark tests comparing an 8xA100 instance and a GH200 instance with the GH200 delivering a 3X speedup at 80% lower Total Cost of Ownership (TCO).

‍

‍

Where is this heading ?

With the HEAVY.AI v8.3 release, we now support the Grace Hopper ARM based architecture and are implementing additional optimizations to maximize its capabilities.   We believe the upcoming Grace Blackwell architecture will drive additional price-performance gains for our customers by making large scale data analysis more accessible, and allowing our customers to solve more compelling use cases. When more data can be analyzed at once, more unique insight can be gained.

‍

See This All in Action

To showcase the game-changing price-performance gains of HEAVY.AI on Grace Hopper, we created a public demo, accessible on our website (https://demo-ships.heavy.ai/ais/dashboard/22?tab=-OAxMGO4IIU3XU6sQ18x). The centerpiece of the demo is the extremely large, publicly available, Automatic Identification System (AIS) ship traffic dataset, from Marine Cadastre, a US government organization. AIS continuously transmits a vessel’s position, identity, speed, course, and other relevant data to nearby AIS-equipped ships. We ingested about 7 years of historical ship tracking data, totaling more than 20 billion rows of data. We also loaded roughly half a million records of ship metadata, live-joining it to the AIS data for additional context. This dashboard demonstrates the ability for users to access this massive amount of data and easily interact with it in ways not possible without the power of HEAVY.AI.

‍

If you have never used HEAVY.AI, you will likely be quite shocked to see how it seamlessly visualizes 20 billion rows of data.  This performance which can’t be achieved by traditional CPU based analytics tools allows you to filter, zoom and interact with the data in real-time through your web browser.

For example, in the timeline view of all these records, you may notice a spike in the blue line that represents the number of records recorded.  “Why such a dramatic increase in the number of ship records?”  In the timeline view, you can select the specific range of time to investigate with a click and drag.  Or maybe the important thing to look at is commercial traffic, now you can select from Vessel type and narrow down to ‘Cargo’ and ‘Tanker’ traffic.

‍

Imagine that you want to look at a specific US port that is often busy, for example Long Beach, CA. You can enter this value in the map view search bar and zoom in to see what was happening in Long Beach during this time period.

‍

‍

You can see there are a number of ships  anchored outside the harbor for long periods of time.  This happens to be a time period when Long Beach experienced a massive uptick in shipping due to COVID-19 related supply chain issues leading up to the holiday season.

This ability to rapidly pinpoint key correlations and anomalies hidden within massive data sets at interactive speed, demonstrates HEAVY.AI’s unique  value - It is the only  platform that can process this much data at interactive speed.

‍

In the coming weeks, we will add more data to further demonstrate the capabilities of our platform.  We plan to  layer additional related maritime data (weather, global wave height, synthetic aperture satellite, undersea data cable locations and global bathymetry data) over the top of the AIS  data.  By combining these datasets, you can discover things like illicit shipping behavior or other types of maritime movement concerns.

‍

Summary

The Grace Hopper superchip, combined with HEAVY.AI’s interactive analytics, is reshaping big data analysis, making it faster, more cost-effective, and scalable for the largest possible datasets. Our partners at Vultr have GH200 instances available for $2.99 / hour, dramatically less than typical A100 and H100 instances that would often be used for these types of workloads.  The incredible price-performance of HEAVY.AI on Grace Hopper enables our customers to interrogate larger than ever datasets and tackle new workloads that were previously not possible .

‍

‍Let us know if you’d like a personalized demo of how HEAVY.AI on the new Grace Hopper superchip can accelerate your business, we’d love to chat.

‍