Unified memory vs discrete GPU: HP ZBook Ultra G1a vs Razer Blade 16

Should an AI laptop rely on a big unified-memory APU or a discrete NVIDIA GPU? We compare the strongest example of each for running models locally.

Published ·Last updated

Short answer

Buy the Razer Blade 16 for the fastest local inference we measure (110 tok/s) and CUDA tooling. Buy the HP ZBook Ultra G1a for 128GB memory that fits larger models, nearly double the battery (11h vs 6h), 1kg less weight and $1,000 lower price.

  • HP ZBook Ultra G1a (Ryzen AI Max+ 395): AMD Ryzen AI Max+ 395 (Strix Halo), 50 NPU TOPS, 128GB RAM, 84 tok/s on Llama 3 13B Q4, 11h battery, 1.5kg, $3,499.
  • Razer Blade 16 (2025, RTX 5090): AMD Ryzen AI 9 HX 370 + NVIDIA RTX 5090, 50 NPU TOPS, 64GB RAM, 110 tok/s on Llama 3 13B Q4, 6h battery, 2.45kg, $4,499.
HP ZBook Ultra G1a (Ryzen AI Max+ 395)
Option A

HP ZBook Ultra G1a (Ryzen AI Max+ 395)

Workstation-class — 128GB unified memory for 70B local

$3,499

Check price →
Razer Blade 16 (2025, RTX 5090)
Option B

Razer Blade 16 (2025, RTX 5090)

Top GPU for 70B local inference

$4,499

Check price →

Spec-by-spec

SpecHP ZBook Ultra G1aRazer Blade 16
Price$3,499$4,499
ChipAMD Ryzen AI Max+ 395 (Strix Halo)AMD Ryzen AI 9 HX 370 + NVIDIA RTX 5090
NPU TOPS50 TOPS50 TOPS
RAM128 GB64 GB
LLM tokens/sec (13B Q4)84 tok/s110 tok/s
Sustained thermal score90/10096/100
Battery life11 h6 h
Weight1.5 kg2.45 kg
AIPC workload-fit9.89.5

Best for

Raw tokens/sec (CUDA)Razer
Largest model size (128GB)HP
Battery + portabilityHP
PriceHP
NPU TOPS (50)Tie
AIPC verdict

Buy the Razer Blade 16 for the fastest local inference we measure (110 tok/s) and CUDA tooling. Buy the HP ZBook Ultra G1a for 128GB memory that fits larger models, nearly double the battery (11h vs 6h), 1kg less weight and $1,000 lower price.

Open chip-level breakdown on AIPC →

See full rankings

More comparisons