Can I run Llama-3.1-8B-Instruct on a NVIDIA RTX 4080 Super?
Yes.Runs
You can run this with Q8 quantization. On a NVIDIA RTX 4080 Super (16 GB VRAM), Llama-3.1-8B-Instruct needs about 11 GB (Q8) and should generate at roughly ~57 tok/s. Recommended: Run on your GPU with Q8.
meta-llama/Llama-3.1-8B-InstructWhy
- Full precision needs ~20 GB which exceeds your accelerator, but 8-bit (INT8 / Q8) needs only ~11 GB and fits.
How to run Llama-3.1-8B-Instruct on a RTX 4080 Super
Run with Ollama (easiest)
ollama run llama3.1:8bOr with llama.cpp
llama-cli -hf meta-llama/Llama-3.1-8B-Instruct -p "Hello"Serve with vLLM (OpenAI-compatible API)
# Fast production serving on http://localhost:8000/v1
vllm serve meta-llama/Llama-3.1-8B-InstructPython (Transformers)
from transformers import pipeline, BitsAndBytesConfig
# pip install transformers accelerate bitsandbytes
quant = BitsAndBytesConfig(load_in_8bit=True)
pipe = pipeline("text-generation", model="meta-llama/Llama-3.1-8B-Instruct", device_map="auto",
model_kwargs={"quantization_config": quant})
print(pipe("Hello", max_new_tokens=50))Want it to run comfortably at higher precision? The smallest GPU that runs Llama-3.1-8B-Instruct well is the AMD RX 7900 XT (20 GB).