NVIDIA A100 Server: Dedicated 1x to 8x A100 GPU Nodes. 8x A100 Server from $12.00/hr
A dedicated A100 80GB server with 1, 2, 4 or 8 GPUs, rented by the minute. The A100 gives you 80GB per GPU on the Ampere generation, and an 8x node gives you 640GB in one box. The table below shows the cheapest live node at each size.
TL;DR: NVIDIA A100 server
- 8x node: the cheapest 8x A100 server on Spheron is $12.00/hr (A100 PCIe 80GB, on-demand), $1.50/hr per GPU, as of Oct 20, 2026.
- 1x server: a single-GPU A100 server on Spheron is $1.20/hr (A100 PCIe 80GB, spot), as of Oct 20, 2026.
- Node pricing: each size is its own offer, not the 1x rate times 8: an 8x A100 SXM4 80GB on-demand node works out to $1.62/hr per GPU against $1.80/hr for 1x, as of Oct 20, 2026.
- Form factors: A100 SXM4 80GB (80GB HBM2e), A100 PCIe 80GB (80GB HBM2e). See the interconnect section for NVLink and PCIe.
- Access: dedicated GPUs, root SSH, per-minute billing. Bare-metal clusters on reserved orders. Single A100 GPU rental
A100 server configurations and node prices
| Node | Node / hr | Per GPU / hr | Host |
|---|---|---|---|
| 1x A100 SXM4 80GBOn-demand | $1.80/hr | $1.80/hr | 24 vCPU · 116GB RAM · 2.1TB disk |
| 8x A100 SXM4 80GBOn-demand | $12.96/hr | $1.62/hr | 240 vCPU · 1,280GB RAM · 9.6TB disk |
| 1x A100 PCIe 80GBOn-demand | $1.43/hr | $1.43/hr | 28 vCPU · 120GB RAM · 850GB disk |
| 1x A100 PCIe 80GBSpot | $1.20/hr | $1.20/hr | 28 vCPU · 120GB RAM · 850GB disk |
| 2x A100 PCIe 80GBOn-demand | $2.86/hr | $1.43/hr | 60 vCPU · 240GB RAM · 1.6TB disk |
| 2x A100 PCIe 80GBSpot | $2.38/hr | $1.19/hr | 60 vCPU · 240GB RAM · 1.6TB disk |
| 4x A100 PCIe 80GBOn-demand | $5.90/hr | $1.48/hr | 60 vCPU · 300GB RAM · 3.1TB disk |
| 4x A100 PCIe 80GBSpot | $4.76/hr | $1.19/hr | 124 vCPU · 480GB RAM · 3.3TB disk |
| 8x A100 PCIe 80GBOn-demand | $12.00/hr | $1.50/hr | 120 vCPU · 1,800GB RAM · 9TB disk |
Each row is the cheapest live offer at that size, form factor and tier when this page was built. A size missing from the table had no offer at that moment. Spot nodes can be reclaimed without notice. Need one GPU, not a node? See A100 GPU rental or current GPU pricing.
A100 form factors and interconnect
- Memory
- 80GB HBM2e, 2.0 TB/s
- GPU to GPU
- NVLink 3, 600 GB/s per GPU, all GPUs in the node
The HGX A100 board. Pick it for multi-GPU training on a budget. 8x node from $12.96/hr.
- Memory
- 80GB HBM2e, 1.9 TB/s
- GPU to GPU
- PCIe Gen4, 64 GB/s per GPU. These offers list no NVLink bridge
80GB per GPU on PCIe cards. Good for single-GPU inference and data-parallel fine-tuning. 8x node from $12.00/hr.
Which A100 node size to pick
- 1x A100
- Serving 7B to 34B models (34B at 8-bit), or QLoRA fine-tuning up to 70B.
- 2x A100
- Llama 3 70B at 8-bit for serving, or full fine-tuning of 7B models.
- 4x A100
- 70B at FP16 for batch inference, or full fine-tuning of 13B models with FSDP.
- 8x A100
- 640GB of GPU memory. Full fine-tuning of 30B-class models, LoRA on 70B, and training runs where cost per hour matters more than time to finish.
Dedicated A100 server or bare metal
Every A100 node above, on-demand or spot, is a single-tenant VM: the GPUs are passed through whole, nobody else is scheduled on them, and you get root SSH into the node. That is different from a GPU VPS, where a host slices one card between tenants or caps your share of it.
Bare-metal A100 clusters, where you control the host OS and multi-node networking, are sold on reserved terms. Reserved GPU commitments cover capacity, term and pricing.
A100 server FAQ
What an A100 server costs, which form factor to pick, and what you get on Spheron.
On Spheron, an 8x A100 server starts at $12.00/hr for the whole node (A100 PCIe 80GB, on-demand), $1.50/hr per GPU. By form factor: A100 SXM4 80GB from $12.96/hr (on-demand), A100 PCIe 80GB from $12.00/hr (on-demand). Rates are live as of Oct 20, 2026 and move with marketplace supply. Spot nodes can be reclaimed without notice; on-demand nodes run until you stop them.
An A100 server is a machine with one to eight NVIDIA A100 GPUs. The 8-GPU HGX A100 build puts eight SXM4 GPUs on NVLink 3 at 600 GB/s each. PCIe A100 servers use cards on PCIe Gen4, which is fine when GPUs rarely talk to each other.
For fine-tuning and inference on models up to about 70B, yes. It has the same 80GB as an H100 at a lower hourly rate. It has no FP8 and less compute, so it is slower than an H100 on transformer training, and on long runs the extra hours can cancel the saving. Compare both rates on the tables before you pick.
Dedicated. Every GPU in the node is passed through to your VM, with root SSH. Bare-metal A100 clusters are available on reserved orders.