CPUs getting swamped?

Teratide reduces the workload before it even gets to your CPUs, powered by accelerators

Teratide reduces the workload before it even gets to your CPUs, powered by accelerators

Teratide provides high-performance analytics solutions for computationally intensive and time-sensitive big data applications. We are specialized in the design, implementation and optimization of accelerated systems by enabling cross-platform in-memory computation and efficient high-performance hardware utilization.

With the increasing computational demands on big data analytics, there is a growing need for high-speed implementations of scalable algorithms. Our company has expertise in optimizing big data solutions on existing cluster infrastructures using in-memory technologies such as the Apache Arrow data framework. In addition, we provide effective methods to integrate high-performance accelerators (such as GPUs and FPGAs) into these clusters. This allows us to combine cutting-edge technologies from the fields of big data analytics and high-performance computing into a unique offering.

Process data in-flight, from any source

Whether your data is coming from IoT sensors, handheld devices, cloud, databases, or storage, our technology directly connects to any hardware interface. There is no need to hit CPU memory before we can start to (pre-) process your data.

By directly connecting to network, storage, or other sources of fast-moving data, our accelerators can filter and pre-process data to prevent any downstream CPUs and memory from being overburdened.

Process data in-flight, from any source
High throughput, low latency, reduced cost per query

High throughput, low latency, reduced cost per query

We never back down on performance, but it should come at a reasonable price. Scaling out is not always the answer, and the CPU no longer reigns supreme. The future is heterogeneous. We are highly-skilled accelerator craftspeople, and do not shun using GPGPUs or FPGAs to meet your needs. With our strong background in computer engineering, we can help you achieve your performance targets, be it high throughput or low latency, or even both, within budget. Take a look at some applications we’ve accelerated using our technology.

Integrated, with the latest and greatest in open source

No matter how fast an accelerator is on its own, it has to deliver its performance while integrated in a larger system. We have extensive knowledge and experience integrating accelerators with modern big data analytics frameworks without a loss of performance through, for example, costly abstractions or (de) serialization. We rely on Apache Arrow and its open-source offspring; Fletcher for FPGA accelerators (of which we are main contributors), and RAPIDs for GPGPU accelerator implementation and integration.

Integrated, with the latest and greatest in open source

Applications

(De)compression

  • Snappy decompression
  • 7✕ higher throughput
  • 8 GB/s per FPGA kernel

Data pre-processing

  • JSON parsing
  • 4✕ higher throughput
  • 10✕ lower latency on Intel PAC

Regex Matching

  • 10✕ higher throughput on AWS EC2 F1
  • 5✕ reduced cost/query

File serdes

  • Parquet-to-Arrow conversion
  • 3✕ higher throughput

Examples of Teratide applications. Speedup nos. are compared to fully utilized multicore CPUs.

Platforms

Our technology can be used with a variety of systems, whether you prefer to accelerate your solutions on-premise in the cloud or at the edge.

Xilinx Alveo

Intel PAC

AlphaData 9H7

Nimbix

AWS EC2 F1

OpenCAPI

Integration technologies

Analytics frameworks

Icons made by monkik from www.flaticon.com