![AdobeStock_352079780-[Converted].png](https://static.wixstatic.com/media/aaa813_8f76e6b5f08e4c1f9deec5aeaffb72cf~mv2.png/v1/fill/w_160,h_67,al_c,q_85,usm_0.66_1.00_0.01,enc_avif,quality_auto/aaa813_8f76e6b5f08e4c1f9deec5aeaffb72cf~mv2.png)
meetup December 3-4
Meet us at HPE Discover Barcelona
SIGMOD 2026 Best Paper: Microsoft's CoddSpeed benchmark features the Speedata APU. Read more
One Chip Can't Do It All
AI runs on data. The data layer is where modern infrastructure breaks.
GPUs train and serve models. But the work that feeds them - Apache Spark, batch ETL, AI data preparation - still runs on CPUs that weren't built for the volumes today's enterprises now process. Pipelines stall, SLAs slip, and the GPUs sit idle waiting for data.
The data layer isn't shrinking - it's compounding. That's the Analytics Processing Unit: a new class of silicon, purpose-built for the workloads that prepare structured data for analytics and AI.
The C200 Analytics Processing Unit
The C200 is the first processor architected specifically for the data layer - Apache Spark, batch ETL, and the AI data preparation pipelines that feed model training and inference.
​
CPUs serialize. GPUs parallelize for matrix math. The Analytics Processing Unit pipelines for analytics: row processing mapped directly into hardware, hundreds of stages advancing on every clock cycle, throughput measured at over a billion rows per second.
​
It drops into existing Apache Spark environments with zero code changes. PCIe form factor. Standard server integration. The silicon your data layer was missing.
Why Speedata
Purpose-built silicon for analytics acceleration - measured in performance, infrastructure footprint, and architectural fit.
Performance
• Up to 100x faster than CPU and GPU on accelerated workloads
• Over a billion rows per second, processed in hardware
• Batch jobs that took hours now run in minute
TCO & Footprint
• Up to 90% lower TCO on accelerated workloads
• Process more data on dramatically fewer servers
• Less rack space, less power, less cooling
Purpose-Built Fit
• Engineered for Apache Spark, ETL, and AI data preparation
• Zero code changes - drops into existing Spark
• Frees GPUs to focus on training and inference
Up to 100x Faster. Up to 90% Lower TCO. Zero Code Changes.
Production workloads, real customers, measured results.
62.7x
· Acceleration on production Spark workloads.
· 37 servers consolidated to 3.
90 → 8
Hours of processing time reduced, same hardware budget.
​
52x
· More performance per dollar than a GPU on analytics queries.
· industry-standard benchmark.
Built for the Workloads That Matter Most
From batch ETL to agentic analytics, the APU accelerates the data layer your business runs on.
Get the APU Your Way
Three deployment paths. Same acceleration. Standard server form factors.
01
Pre-Configured from Speedata
A fully integrated 2U server with two C200 accelerator cards, ready to run Apache Spark workloads out of the box. The fastest path to acceleration.
​
02
OEM Partners
Source servers equipped with the C200 through OEM partners Dell and HPE. Customize the server model and configuration to match your existing data center standards.
03
Speedata Acceleration for AWS
Keep your data in S3. Sync deltas to a Speedata-hosted environment, accelerate ETL and AI data prep on the APU, and push processed data back to AWS. Built for AWS-dependent customers.