A100 80GB PCIe
from $0.89/hr.
The data-centre workhorse. 80 GB of HBM2e and a mature software stack: if a repo says it was tested on a GPU, it was probably this one.
Billed from your wallet · stop any time
What the A100 80GB
is built for.
Tensor-core precisions: BF16, FP16, TF32, INT8.
Best at
- Fine-tuning 7B–13B models in BF16
- Serving quantized 70B models
- Reproducing published research
- Batch inference and embeddings
Launch from a terminal
Same card, same price as the console.
Will my model fit
on 80 GB?
Rules of thumb for weights plus a modest KV cache or optimizer state. Longer contexts and bigger batches need more.
| Workload | Needs | On a A100 80GB |
|---|---|---|
| Llama 3.1 8B · FP16 inference | ~18 GB | Fits |
| Qwen 32B · FP8 inference | ~36 GB | Fits |
| Llama 70B · 4-bit inference | ~42 GB | Fits |
| Llama 70B · FP8 inference | ~78 GB | Tight |
| Llama 8B · LoRA fine-tune (BF16) | ~24 GB | Fits |
| 32B · QLoRA fine-tune | ~40 GB | Fits |
| Flux.1 [dev] · BF16 | ~33 GB | Fits |
| SDXL · FP16 | ~12 GB | Fits |
Guides for
the A100 80GB.
Benchmarks and sizing guides from our blog.
Rent A100 80GB in the Cloud – Deploy in Minutes on packet.ai
Six years on, the A100 is still the card half the world's models were trained on. 80 GB of HBM2e, 2 TB/s, every framework tuned for it.
Fine-Tuning LLMs on GPU Cloud: A100 vs H100, QLoRA Memory Math and Cost Per Run
QLoRA fine-tuning a 70B model on a single A100 80GB takes 24–36 hours. Here is the full cost math by method and model size.
NVIDIA A100 vs H100 in 2026: Price, Performance and Which GPU Fits Your Workload
The H100 delivers 2.88x more LLM inference throughput at high concurrency. For QLoRA fine-tuning, total run cost is within 12.6% at 70B. Here is the full benchmark and cost breakdown.
Not the right fit?
Compare the fleet.
Every card on sale is billed hourly, with no contract.
Launch a A100 80GB
from $0.89/hr.
Sign up, top up, launch. You'll be on a GPU in minutes.
