Dell leads local AI laptop tests as shared memory expands model capacity

The Dell Pro Max 18 Plus delivered the fastest local AI performance in StorageReview lab testing, reaching 185.1 tokens per second on Phi with its RTX PRO 5000 Blackwell GPU. Its drawbacks include weight and measured battery life of 3 hours 39 minutes. The Dell Pro Max 16 Plus offered similar throughput with longer battery life, making it a more portable alternative.

For larger models, the HP ZBook Ultra G1a 14 prioritized capacity over speed. Its unified memory allowed DeepSeek-R1 70B to run locally, while Gemma 3 27B generated at 8.96 tokens per second. Shared-memory designs can accommodate workloads that exceed discrete GPU memory, but generally sacrifice generation speed.

The Lenovo ThinkPad P14s Gen 7 stood out for portability with a discrete GPU, while the Dell Pro Precision 5 14s Intel handled larger workloads through shared memory without dedicated graphics. The Snapdragon-powered HP EliteBook 6 G1q suited smaller-model chat, summarization and offline assistants. GPU-accessible memory remains more important than headline NPU ratings for most local LLM workloads.

Among RTX Spark laptops, the HP OmniBook Ultra 16 led the tested group for coding-agent work, while the Lenovo Yoga 9n 2-in-1 16 ran coolest. These systems used a separate benchmark suite, preventing direct comparisons with the Procyon-ranked laptops. Their battery results remain pending, and the other laptops’ productivity battery measurements do not represent sustained inference runtime.