Tenstorrent and Smallest.ai bring on-prem voice AI to Galaxy servers
Tenstorrent and Smallest.ai launched an on-premises voice AI stack that runs Smallest.ai's Lightning V2 text-to-speech model natively on Tenstorrent Galaxy Blackhole servers. The companies claim 4x lower inference cost than NVIDIA L40S with comparable audio quality.
- Lightning V2 is available on Tenstorrent hardware with usage-based pricing
- Galaxy server: 32 Blackhole ASICs, 23 PFLOPS Block FP8, list price $160,000
- Over 95% of Lightning V2 layers run at LoFi computational fidelity
- DNSMOS 3.801 on P150 vs 3.872 on L40S, within minor perceptual variation
Read next
Security