Taalas, a 2.5-year-old chip startup, is transforming LLM deployment by converting AI models into custom ASICs and showcased HC1, a TSMC 6nm chip embedding Meta's Llama 3.1 8B. The company states its "Hardcore Models" deliver approximately 10× token-per-second throughput and 20× lower production costs versus high-end software infrastructure, and it demonstrated a 30-chip cluster reaching 12,000 TPS per user.
Key Points
- 1Maps LLMs into custom ASICs, producing HC1 chip embedding Meta's Llama 3.1 8B model
- 2Achieves claimed 10× TPS and 20× lower production costs by merging storage and computation on silicon
- 3Requires per-model hardware; scales via clusters (30-chip demo reached 12,000 TPS/user), limiting weight updates
Scoring Rationale
Strong performance claims and demonstrable cluster scaling, limited by single-source company claims and narrow model scale.
Sources
Public references used for this report.
Practice with real Logistics & Shipping data
90 SQL & Python problems · 15 industry datasets
Used by DS/ML engineers at top companies
High-Value Overnight OrdersEasyDelivered International ShipmentsMediumOn-Time Delivery Rate by CarrierHard
250 free problems · No credit card
See all Logistics & Shipping problems
