In this post address the question that’s been on everyone’s mind; Can you run a state-of-the-art Large Language Model on-prem? With *your* data and *your* hardware? At a reasonable cost?

In this post address the question that’s been on everyone’s mind; Can you run a state-of-the-art Large Language Model on-prem? With *your* data and *your* hardware? At a reasonable cost?
This is just a short post to announce a more usable version of the NVIDIA GPU powerlimit setup script that I released a few months ago. This update to version 0.2 uses an interactive mode to set GPU powerlimits and optionally setup a systemd unit file to set these limits on subsequent reboots.
This post presents testing data showing that power-limit reduction on NVIDIA GPUs have give significant benefits for both high wattage and lower wattage GPUs. Power-limit vs Performance data is presented for 1-4 A5000 and 1-4 RTX3090 GPUs.
In this post I am referencing a Bash shell script I recently put together for setting up automatic NVIDIA GPU power-limit lowering at system boot. This allows a reliable way to configure and maintain multi-GPU systems for stable operation under heavy load.
NVIDIA Enroot has a unique feature that will let you easily create an executable, self-contained, single-file package with a container image AND the runtime to start it up! This allows creation of a container package that will run itself on a system with or without Enroot installed on it! “Enroot Bundles”.
For computing tasks like Machine Learning and some Scientific computing the RTX3080TI is an alternative to the RTX3090 when the 12GB of GDDR6X is sufficient. (Compared to the 24GB available of the RTX3090). 12GB is in line with former NVIDIA GPUs that were “work horses” for ML/AI like the wonderful 2080Ti.
Enroot is a simple and modern way to run “docker” or OCI containers. It provides an unprivileged user “sandbox” that integrates easily with a “normal” end user workflow. I like it for running development environments and especially for running NVIDIA NGC containers. In this post I’ll go through steps for installing enroot and some simple usage examples including running NVIDIA NGC containers.
I recently wrote a post introducing Intel oneAPI that included a simple installation guide of the Base Toolkit. In that post I promised a follow-up about the the oneAPI AI Analytics Toolkit. This is it! I’ll describe what it is and give recommendations for doing an install setup of the AI toolkits using conda with Anaconda Python.
This is a follow up post to “Quad RTX3090 GPU Wattage Limited “MaxQ” TensorFlow Performance”. This post will show you a way to have GPU power limits set automatically at boot by using a simple script and a systemd service Unit file.
Can you run 4 RTX3090’s in a system under heavy compute load? Yes, by using nvidia-smi I was able to reduce the power limit on 4 GPUs from 350W to 280W and achieve over 95% of maximum performance. The total power load “at the wall” was reasonable for a single power supply and a modest US residential 110V, 15A power line.