NVIDIA RTX Spark Linux drops to 14 tok/s
MindStudio reports that native Linux on the RTX Spark laptop delivers just 13.9 tokens per second on GPT-OSS-120B, versus roughly 65 tokens per second on Windows and WSL. The slowdown appears when plugged in, with Linux GPU clocks falling to about 780 MHz.
This looks less like a Linux kernel failure than a serious platform-support gap: RTX Spark’s local-AI promise depends on Windows-only power-management logic.
- –MindStudio measured native Linux at roughly 21% of Windows performance, while WSL preserved near-native speed.
- –The reported cause is an encrypted power table handled by NVIDIA’s Platform Controllers and Framework; NVIDIA has not independently confirmed the explanation.
- –WSL is the practical workaround, but it forces Linux developers to retain Windows’ driver stack and manually adjust unified-memory allocation.
- –The contrast with Linux-based DGX Spark systems exposes an ecosystem split between NVIDIA’s desktop AI hardware and its Windows-focused laptop platform.
DISCOVERED
1h ago
2026-11-04
PUBLISHED
5h ago
2026-11-04
RELEVANCE
AUTHOR
novaRom