Webinar Recap: The New Computing Paradigm for Advanced Analytics & AI
- Daniela Sztulwark
- Dec 10, 2025
- 3 min read
Updated: 4 days ago
Beyond CPU. Beyond GPU. The APU Era Has Arrived.
Recently, Speedata CEO Adi Gelvan, VP EMEA Bam Gobets and VP R&D and Co-founder Dani Voitchekov hosted a live session exploring how purpose-built silicon is transforming large-scale analytics and AI data processing. If you missed it, or want a refresher, here's what we covered:
- How the APU delivers 100x faster Apache Spark performance
- Where purpose-built silicon excels vs general-purpose compute
- Real-world results from pharma and adtech deployments
- Getting started with Speedata's Workload Analyzer - our no-cost APU simulator, run in your own environment.
The Problem - CPUs and GPUs Weren't Built for This
As data volumes explode and AI workloads become more demanding, organizations are hitting the limits of traditional compute. CPUs offer general-purpose flexibility but limited parallelism. GPUs excel at matrix math for ML training and inference, but neither was architected for the specific demands of query processing, data transformations, and analytics at scale.
The result? Sprawling clusters, mounting infrastructure costs, and batch jobs that take hours (or days) when insights are needed in minutes.
Enter the APU Purpose-Built for Analytics
The Analytics Processing Unit (APU) represents a new category of processor designed from the ground up for data-intensive workloads. Speedata's C200 APU, powered by our Callisto ASIC, delivers massively parallel operator- and query-level execution, not generic vector math.
How does APU performance compare?
CPU | GPU | APU | |
Optimized for | General purpose computing | Graphics, ML training, AI inference | Data transformations, ETL, accelerated analytics, AI data processing |
Execution type | Serial with limited parallelism | Massively parallel vector/matrix math | Massively parallel operator & query instructions |
Analytics performance/$ | 1x (baseline) | ~2-3x | 10x-300x |
The C200 APU accelerates the entire data pipeline in hardware: decompression, decoding, filtering, columnar processing, joins, aggregations, and shuffle preparation, all without touching main memory the way CPUs must.
Native Apache Spark Integration means Zero Code Changes
One of the biggest questions we hear: "How disruptive is adoption?"
The answer: not at all. Speedata integrates natively with Apache Spark via a plugin architecture. Your existing SQL queries and DataFrames run unchanged. The APU code generator sits alongside the standard CPU path, and the scheduler intelligently routes work to APU-equipped nodes.
No rewrites. No framework changes. Step-by-step integration into your existing environment.
Real-World Results - Pharma & Adtech Case Studies
Pharmaceutical: From 90 Hours to 8 Hours
A pharmaceutical company developing new medicines needed to dramatically improve time-to-insight on repeated large batch jobs processing multiple petabytes across a 100-node cluster.
The result:
Before: 100 single-CPU servers, 90 hours
After: 1 Speedata server with 4 APUs, 8 hours
Speedup: 11x faster end-to-end or 275x faster when comparing CPU count to APU count
Global Adtech Leader: From Minutes to Seconds
A global adtech company processing trillions of ad impressions monthly across 1,200 nodes needed low-latency, high-throughput analytics for real-time bidding.
The result:
CPU: 8 minutes 44 seconds
APU: 48 seconds
Speedup: ~11x acceleration, >90% runtime reduction
Key Use Cases for APU Acceleration
During the session, we walked through three primary deployment patterns:
Traditional Batch ETL – Accelerate data cleaning, transformations, and enrichments in your existing Spark pipelines
AI Data Preparation – Speed up the heavy lifting before model training: de-duplication, normalization, tokenization, language detection, and creating curated training sets
Table Augmented Generation (TAG) – Enable real-time SQL execution against structured data for LLM-powered analytics with explainable, auditable, hallucination-free results
Get Started - Free Workload Analyzer
Want to see what APU acceleration could mean for your specific workloads? Our Workload Analyzer is a free APU simulator you run in your own environment—your data never leaves your infrastructure.
How it works:
Download the Workload Analyzer executable (free)
Attach your Apache Spark logs - in your own environment.
Visualize your current Spark jobs
Add virtual APUs to predict job completion time reductions
In testing, we've seen predictions like 39.6x speedup, from 1 hour 15 minutes down to under 2 minutes.
Try our Workload Analyzer
Watch the Full Session
Ready to dive deeper? Watch the complete webinar recording above, or book a demo to see the APU in action with your team.
