HP Z6 G5 A vs NVIDIA DGX Spark

HP's Z6 G5 A is the smallest Threadripper Pro OEM workstation on the market and the rational mid-tier pick below the flagship Z8 Fury G5. Reviewers at PCMag, AnandTech, StorageReview, Phoronix, and DEVELOP3D consistently praised the build quality, toolless serviceability, and 96-core CPU ceiling, with StorageReview giving it their 'highest recommendation for a high-end tower workstation.' For local-LLM use, configurations with 1-3 RTX 6000 Ada GPUs (48 GB VRAM each at ~960 GB/s) land in the 25-40 tokens/sec range on Llama-3-70B Q4 with a single GPU, and go substantially higher with multi-GPU tensor parallelism. One caveat: none of the published professional reviews ran formal Llama-3 70B Q4 benchmarks, so the LLM-specific numbers here come from single-GPU norms rather than measurements on the HP Z6 itself.

The NVIDIA DGX Spark is the productized version of Project DIGITS — a 150 x 150 x 50.5 mm box housing the GB10 Grace Blackwell Superchip, 128 GB unified LPDDR5x, and the full CUDA AI stack out of the box. Tom's Hardware called it 'a well-rounded toolkit for local AI'; ServeTheHome called it 'must-have for AI developers'; LMSYS published the most thorough independent benchmarks. The 128 GB unified-memory ceiling is the headline feature: it loads models that would otherwise need a $30K+ multi-GPU rig. The catch is bandwidth-limited decode — LMSYS measured Llama-3.1 70B FP8 at 2.7 tokens/sec single-batch, while GPT-OSS 120B (MoE, ~17B active) hits ~14.5 tokens/sec per ServeTheHome. Best understood as a CUDA-native development box for buyers who need to iterate on big-model code without renting cloud GPUs.