Self-host Meta's Llama models on an NVIDIA Tesla P40 (24 GB VRAM, Pascal architecture). Suited to 8B models at full precision and larger models when quantized. GPU servers are not orderable yet; contact us to be notified when they are.
$ curl -fsSL https://ollama.com/install.sh | sh && ollama run llama3
# Example install command
Llama is Meta's family of openly available large language models. Running Llama on your own GPU server keeps prompts and data private.
Runs 8B models on 24 GB of VRAM. 70B models do not fit on a single 24 GB card at common quantization levels.
Your prompts and data stay on your server, with no third-party API calls.
Use GGUF with llama.cpp or Ollama, or native PyTorch weights.
Serve models with Ollama or the llama.cpp server for an OpenAI-compatible endpoint.
LLaMA can use an NVIDIA GPU to speed up its work. VPS.org GPU servers are not available to order yet; contact us to hear when they are. LLaMA is a general GPU-accelerated workload. Make sure your software has CUDA support and that your driver / runtime versions match the workload requirements for LLaMA.
GPU servers are not available to order yet. Today you can run LLaMA on a regular CPU plan, where it works far more slowly than on a GPU: SSH in and run curl -fsSL https://ollama.com/install.sh | sh && ollama run llama3.
VRAM needs for LLaMA depend on the model and batch size; check the project documentation for current figures.
GPU pricing will be published when GPU servers can be ordered. Contact us if you want to hear when that happens.
Yes. Every VPS.org server comes with full root access, so you can run other tools alongside LLaMA within the server's RAM and storage.
Every VPS.org server comes with full root SSH access, so you can install drivers, libraries and LLaMA however you need.
No. No server comes with a CUDA runtime or NVIDIA driver preinstalled, and GPU servers are not available to order yet. You install the versions LLaMA needs yourself.
Yes. VPS.org servers are persistent virtual machines, not ephemeral instances: LLaMA models, configs and data stay on the server's SSD storage until you delete them.
Keep working data on the server's SSD for fast access during LLaMA runs, and copy finished artifacts (weights, generations, embeddings) to storage outside VPS.org; snapshots and backups stay on the same storage cluster.
You can upgrade any VPS.org server to a larger plan from the control panel; the resize restarts it and your LLaMA install carries over. GPU servers are not available yet.
Yes. Backups are a paid add-on (daily +30% of the plan price, kept 7 days; weekly +20%, kept 28 days) and snapshots are free, which is useful before long LLaMA runs. Both stay on the same storage cluster as the server.
Contact support within 30 days of your first purchase to request a refund. GPU servers are not sold yet, so no GPU order is involved.
GPU servers are not orderable yet. Tell us about your workload and we will let you know when they are.