Speedata logoSP

Speedata

About

Speedata designs the C200 Analytics Processing Unit (APU), a custom silicon accelerator purpose-built to execute analytics operations directly in hardware. The APU offloads Apache Spark SQL operations - including joins, aggregations, and transformations - from general-purpose compute, targeting the latency and throughput bottlenecks that emerge in data-center analytics pipelines.

Analytics workloads on conventional hardware face a fundamental trade-off: scaling compute increases total cost without proportional gains in query performance when operations like shuffles and aggregations become memory-bound or network-bound. Hardware acceleration at the data-processing layer attempts to shift this trade-off by embedding specialized execution into silicon, reducing the gap between theoretical and realized throughput while lowering per-query resource consumption.

The C200 integrates hardware and software into a unified platform. The company claims the design addresses operational pain points: underutilized compute resources, SLA misses, and project delays caused by data bottlenecks. Evaluation would require measurement against specific workload characteristics - query complexity, data volume, data locality, and network topology - since acceleration gains vary across different cluster configurations and query patterns.

Similar companies

Graphcore logoGR

Graphcore

Graphcore is a British semiconductor company that develops Intelligence Processing Units (IPUs) and complete AI compute systems to accelerate machine learning and AI applications.

198 jobs
Tenstorrent logoTE

Tenstorrent

Tenstorrent is a next-generation computing company that builds computers for AI, developing AI Graph Processors, high-performance RISC-V CPUs, and configurable chiplets with an open-source software stack.

116 jobs
Cerebras logoCE

Cerebras

Cerebras Systems builds the world's fastest AI infrastructure with industry-leading speed, scale, and quality through wafer-scale AI chips.

87 jobs
d-Matrix logoD-

d-Matrix

d-Matrix builds purpose-built AI inference computing platforms to make generative AI commercially viable, efficient, and sustainable through digital in-memory compute technology.

43 jobs
Etched logoET

Etched

Etched builds specialized AI inference chips, including Sohu, the world's first transformer-specific ASIC that delivers 10x faster performance than GPUs.

25 jobs
SambaNova logoSA

SambaNova

SambaNova Systems is a full-stack AI infrastructure company delivering the fastest and most energy-efficient AI inference platform through custom RDU chips and software, enabling enterprises to deploy sovereign AI with complete data control.

3 jobs