Apple Silicon's memory economics: M5 Ultra offers 5.6x better $/GB/s than DGX Spark for LLM inference
SumitGup · x · 2026-10-19
LLM inference is heavily memory-bound, so $/GB/s of memory bandwidth is the key metric. M5 Ultra: $5499, 96GB @ 1.2TB/s = $4.5/GB/s — 5.6x better than DGX Spark ($25.5/GB/s) and 4.7x better than DGX Station ($21.1/GB/s). The author projects a 512GB Mac Studio at $15,899, with RDMA over Thunderbolt 5 offering sub-10μs latency and up to 400Gb/s interconnect across 5 cables.
More from Infra
- ReaDy Uses Recurrent States as Dynamic Context for Speculative Decoding in Hybrid Linear Attention LLMs — Bohan Sun · 2026-10-19
- Celesto Claims CI Runner With 96% Lower Per-Minute Cost Than GitHub Actions — aniketmaurya · 2026-10-19
- Huawei's Atlas 960E superpod links 4,096 NPUs with near-packaged optics — bookwormengr · 2026-10-19
- OCP attendees split on CPO/NPO light sources: external CW lasers (ELSFP) vs micro-VCSEL — bookwormengr · 2026-10-19
- Robot brain inference cost fell 84% in two months, from $2.36 to $0.38 per task — victor_explore · 2026-10-19
- Cross-encoder reranking loses to a stricter cosine threshold on 2 of 3 semantic cache benchmarks — Reasonable_Royal_621 · 2026-10-19