#gpu

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.

Google AI Ultra includes monthly Colab compute that, at A100 rates, runs for over 200 hours, which is a good deal for running open-source models. Compares the official Colab CLI and the remote Colab MCP: the CLI suits complex deployments, while MCP lets you turn GPUs on and off from within a chat at any time. Includes a quota table, hands-on steps, cost-saving principles and FAQs.

Use when a user wants an agent to start a GPU or TPU on Google Colab, run code, deploy or fine-tune open-source models, fetch results, and release the machine. Combines the official google-colab-cli (Apache-2.0) and the Colab remote MCP into one workflow: the CLI handles file transfer, SSH, Drive, and one-off jobs, while MCP handles lightweight start/stop from a chat client. Enforces releasing runtimes after use, asking before expensive GPUs, and never paying on the user's behalf. Use case: Let an agent start a GPU on Colab for you, run open-source models, fetch the results, and shut down automatically. Use the CLI from a local terminal, or the remote MCP from a chat client.

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.