Skip to content

Vectorizationthe technology behind real-time analytics at scale

Kinetica was designed from the ground up to leverage parallel compute capabilities of GPUs and modern 'vectorized' CPUs. This introduces a new level of brute-force compute power that breaks open the door to faster and more flexible querying across large and streaming datasets.

What is Vectorization?

The secret sauce behind breakthrough analytics lies in Kinetica's ability to utilize modern vectorized CPUs and GPUs

Most databases have evolved with the CPU

The CPU has been the core of the computer for decades. Database systems have evolved alongside using sequential processing to perform calculations.

Take this example of an array of numbers. To add five to each number and place them into a new array, a CPU will rapidly work through the list.

But this sequential process has its limits.


What if you could do 1000 instructions at once?

GPUs which typically have thousands of cores were designed to speed up drawing of graphics on a screen. Instead of rendering a pixel at a time, a GPU could render a whole screen in one go – a technique known as single-instruction, multiple data (SIMD).

It turns out this same capability is well suited to performing repeated similar instructions on data in parallel. With Intel's Advanced Vector Extensions (AVX) making it into CPUs in the data center, the path is now wide open to leverage vectorized compute in the cloud for analytics workloads.


How does Kinetica harness vectorization?

Kinetica was designed from the ground up to leverage the vectorization capabilities of GPUs and modern CPUs. Analytical functions in Kinetica have all been written from scratch to take advantage of vectorization.

Vectorization unleashes significant performance improvements – particularly on spatial and temporal queries at scale. Aggregations, predicate joins, windowing functions, graph solvers all operate far more efficiently.


Industry Crushing Analytics Performance

Kinetica's vectorized queries deliver remarkable performance, without tuning, on standardized decision-support benchmark tests.
6x Faster
than Apache Spark 3
6x Faster
than ClickHouse 21
8x Faster
than Databricks 9.1 LTS (Photon)
10x Faster
than Snowflake
All using identical CPU-only hardware against TPC-DS SF1000

Vectorization Gives You Freedom

With so much raw compute power, you won't need to worry about indexing, partitioning or downsampling!.

Simpler Data Structures

Brute force vectorized compute means there is less need to think through schemas before data can be explored.

Low Latency

Simpler data structures means less to index. Combined with Kinetica's lockless, distributed architecture, data is available for query immediately after it lands.

Linear Scale Out

With less to index, the database scales in proportion to the size of the data. This leads to a smaller and more predictable scale-out footprin.t

Less Engineering

Spend less time engineering schemas, and more time using your data. Business analysts have more flexibility and freedom for ad-hoc data discovery projects.

Try it Now: Kinetica Developer Edition is Free Get Started »

Better Performance with Less Infrastructure.

With vectorized algorithms and reduced need for supporting data structures, Kinetica enables you to do more with fewer resources and less work than comparable systems.

Large US Financial Institution

700-node spark cluster running queries in hours took seconds on 16 nodes of Kinetica

Top US Retailer

Consolidated 100 nodes of Cassandra(NoSQL) and Spark into 8 Kinetica nodes

Large Pharma

Identical performance between a 88-node Impala cluster and a 6-node Kinetica Intel cluster in Azure
White Paper

Vectorization: The New Era of Big Data Parallelism

Every five to 10 years, an engineering breakthrough emerges that disrupts database software for the better. Vectorization is the newest breakthrough gaining momentum towards widespread adoption. Early adopters are using fully vectorized databases to foster new applications and reap lower costs.

Learn more about vectorization in this white paper.

Download the White Paper
Vectorization White Paper