MCDMA Enables Direct RDMA Between NVIDIA Spark and Apple Silicon
StephanSturges · x · 2026-08-19
MCDMA (Metal CUDA Direct Memory Access) enables a direct RDMA path between NVIDIA Spark GPUs and Apple Silicon Macs via USB-C, allowing direct access between CUDA memory and Metal unified memory. The author's setup uses dual Sparks for prompt processing, connected concurrently to a Mac Studio for decoding via two independent USB-C links. Benchmarks show 939 MB/s on a single link, 1.80 GB/s Mac to dual Sparks, and 1.25 GB/s reverse, providing an efficient data transfer solution for heterogeneous computing architectures.
More from Infra
- Inference Era Competition: Cloud Decisions Driven by Revenue per Watt — BenBajarin · 2026-08-19
- Code.Storage Opens Signups: Unlimited Git Infrastructure for AI Agents — dsp_ · 2026-08-19
- Lava lamps help secure internet data — dreamwieber · 2026-08-19
- Why cloud providers stopped auctioning compute: The $999 spot instance incident — lauriewired · 2026-08-19
- Engy launches permissionless TEE worker onboarding for confidential GPU compute — markjeffrey · 2026-08-19
- Linux Desktop Share Surges Past 10%, Driven by AI Developers — DavidLinthicum · 2026-08-19