#local-llm

Hardware security researcher Kate has published work showing that NVIDIA's CMP 170HX mining card uses the same chip as the data center A100. She used voltage fault injection to bypass the fuse limits and unlock most cards to 80GB of memory and more compute units, with half-precision speed above the retail A100. The method is very hard to do and can destroy the card.

Strata has to support NVIDIA, AMD, Intel, Apple and many other kinds of hardware, so it can only take a general-purpose approach and can't squeeze out every bit of performance for one particular GPU generation. We think a version aimed specifically at the 40 and 30 series may come later and could be faster; you can already use Strata on 30 and 40 series cards today.

Generate 48kHz voiceovers for free on your own computer, design a voice with one sentence, or clone a voice from your own recording.

Use when the user wants voiceover, text-to-speech, a new custom voice, or to clone their own voice. Uses the open-source VoxCPM2 to generate 48kHz speech locally, supports voice design from a one-sentence description and cloning from a short recording, Apache-2.0 licensed and free for commercial use. Requires an NVIDIA GPU; only clone your own voice or voices whose owners have consented. Use case: Generate voiceovers locally for free: type text to speech, design a voice from one sentence, or clone a voice from the user's own recording.

An experience piece (members only): for occasional use (chat, copywriting, code edits), an API is cheaper; for high-frequency batch work (daily transcription, classification, processing large volumes of material, fixed workflows), running locally may pay off. The core dividing line is monthly token consumption, plus whether the task needs top-tier model quality. The article works out the numbers and the hidden costs on both sides for you.