Unsloth adds native ARM64 Windows CUDA support for RTX Spark, 2.7x faster training
danielhanchen · x · 2026-10-17
Unsloth now runs natively on NVIDIA RTX Spark with ARM64 + Windows + CUDA support, running and training diffusion, multimodal, language and decision models natively — optimized to 2.7x faster. With up to 128GB unified memory, you can run and train 120B+ parameter models locally on a laptop. Install via winget or the Unsloth Desktop download.
More from Infra
- ExecuTorch is now an official Zephyr module, bringing PyTorch models to MCUs via west — PyTorch · 2026-10-17
- AMD shows OCP OAC accelerator cards with up to 5000W liquid-cooled designs — wkmyrhang · 2026-10-17
- Fixing Omarchy install on ASUS laptops when UEFI NVRAM is full, with Codex's help — GuyHachmon · 2026-10-17
- fx v0.0.14 ships durable agents: same API survives crashes on laptop, cloud via Vercel — cramforce · 2026-10-17
- VM filesystem sharing still painful: virtiofs costs too much with live migration — davidcrawshaw · 2026-10-17
- MicroVM shared-fs prototyping: NFS cuts throughput in half, IOPS to 20% — zeeg · 2026-10-17