Hardware Accelerators: What are they?

CPUs are general-purpose. They handle all processing pretty well. But technology has come a long way and some hardware is far superior for specific computations.

FPGAs and GPUs are programmable hardware that can provide compute efficiency. FPGAs in particular provide compelling advantages for data platforms like Apache Spark.

Hyperacceleration = Optimized processing + Zero code change

Hyperacceleration improves performance of data platforms like Apache Spark by automatically optimizing them with compute hardware. The flexible approach gives access to efficient hardware like Xilinx Alveo FPGAs with no impact on user Spark coding.


  • FPGAs and GPUs are two field-programmable hardware acceleration approaches.
  • FPGAs are typically much less expensive than GPUs or CPUs.
  • FPGAs generally use much less power than GPUs.
  • A large number of data analytics operations gain from FPGAs (e.g. scan, shuffle, filter, write).

Bigstream Hyperacceleration has flexible approach, with emphasis on FPGAs

Performance per $ Comparison Read the Blog

Performance per Watt Comparison Read the Blog

Flexible Deployment - On prem or Cloud

Bigstream delivers Hyperacceleration with FPGAs whether you run Spark clusters on premises or in the cloud.

Zero Code Change

The value of Bigstream is that it automatically provides the communication and translation between Spark code and the computing environment.

While other acceleration approaches force you to alter your code or switch platforms entirely, with Bigstream your Spark code remains 100% unchanged.

Deliver Results

Fast Insights

Jobs run up to 10X faster without changing your Apache Spark code, increasing data science productivity and the ability to meet tight SLAs

Enriched Analytics

Incorporate fuller data sets and more data sources by eliminating the bottlenecks forcing you to make compromises

Lower TCO

Instead of scale-out and scale-up, optimize your existing Spark environment with Bigstream, lowering hardware and operating expenses