A100 SXM cloud rental prices
Every provider offering A100 SXM, normalised to hourly on-demand pricing with VRAM and region details.
About the A100 SXM
- Memory
- 80 GB
- Architecture
- Ampere
- Memory bandwidth
- 2.0 TB/s
- Dense FP16 throughput
- 312 TFLOPS*
* Vendor-published figures, with sparsity where the vendor states it. Verify on the official datasheet before you size a cluster.
30-day price history
Live GPU price comparison
Filter by provider, GPU or region — filters run locally in your browser.
| Provider | GPU model | VRAM | GPUs | vCPU | System RAM | Region | On-demand /hr | Spot /hr | Actions |
|---|---|---|---|---|---|---|---|---|---|
| Runpod | A100 SXM | 40 GB | 1 | — | — | Global (secure + community cloud) | $1.0000 | — | Visit provider |
| Lambda Labs | A100 SXM | 80 GB | 1 | — | — | Reference listing | $1.2900 | — | Visit provider |
| Runpod | A100 SXM | 80 GB | 1 | — | — | Global (secure + community cloud) | $1.3900 | — | Visit provider |
Frequently asked questions
Which GPUs are best for LLM inference?
For models that fit in memory, memory bandwidth matters more than raw FLOPS, so cards like the L40S, RTX 4090 or A100 often give better tokens per dollar than an H100. Once a model needs multi-GPU tensor parallelism, NVLink-equipped H100 or H200 instances usually win on throughput.
How much VRAM do I need to fine-tune a model?
Full fine-tuning needs roughly the parameter count times the bytes per parameter for weights, plus optimiser state which can be several times larger again. LoRA and QLoRA reduce the optimiser footprint dramatically. Use the LLM VRAM calculator to size a specific configuration.
Is it cheaper to rent or to buy a GPU?
Renting usually wins unless you can keep a device busy for a large fraction of the year. Cloud rental also shifts the cost of hardware refreshes onto the provider, which matters when new generations arrive every twelve to eighteen months.
Why do the same GPU models have different prices?
Configuration differs: number of GPUs per instance, system RAM and vCPU allocation, local disk, region, and whether the capacity is interruptible. We normalise to a single GPU so you can see through those differences.