NVIDIA Watch posted an update
MiaAIlab says NVIDIA’s Nemotron 3.5 Bolt ran at 100 tokens a second on an RTX Spark with 64GB of memory, even on battery. If that figure holds beyond this brief demonstration, it is a striking result for local AI on a portable system.
Why it mattersThe post gives no benchmark setup or further measurements, so treat the number as the account’s claim, not a like-for-like performance result. Still, it puts a useful question on the table: how much local model speed can a compact system deliver away from a plug?
Discuss: Would you trust a short, unspecific demo as a useful sign of local-AI performance, or should claims like 100 tokens a second wait for a reproducible benchmark?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.