Vpipe vs Draw Things on M5 Pro: 24% faster at 1K, finishes 2K where Draw Things crashes
TgoAI · reddit · 2026-09-26
The Vpipe developer posted a side-by-side benchmark against Draw Things on the same M5 Pro 24GB machine with identical settings:
- 1024×1024: Vpipe 41s vs Draw Things 54s (24% faster)
- 2048×2048: Vpipe finished in 173s; Draw Things took 221s then crashed (22% faster, stable)
The author argues the bigger differentiator is memory management: Vpipe is built on a native C++/Metal runtime with explicit memory control over peak usage and tensor lifetimes, rather than relying on whether a model happens to fit in unified memory. That makes 16GB Macs a first-class target rather than an edge case. A video of the actual runs is included, and results from other M-series configs are invited.
More from Infra
- Polars 2.0 RC2 ships Map dtype, join-derived predicates and parquet scan optimizations — JeremyCMorgan · 2026-09-26
- PrismML runs a 2B vision model with 1-bit weights on 4GB smart glasses at 2x speed — lmoroney · 2026-09-26
- Goldman: AI drives nearly half of S&P 500 EPS growth in 2026, turns drag by 2028 — rohanpaul_ai · 2026-09-26
- Oracle accused of hiding cash flow issues via Enron-style prepayments amid $7.5B share sale — SumitGup · 2026-09-26
- NVIDIA B300 sandboxes now available on-demand and spot on Daytona — mattturck · 2026-09-26
- How Long Until Local ~30B A3B Models Match GLM 5.3 Flash Quality? — Aggravating-Push-207 · 2026-09-26