The A100 is the 80 GB GPU to rent when memory matters more than the newest number formats: 80 GB of HBM2e, up to 2,039 GB/s of bandwidth on the SXM variant, and NVLink between GPUs on the same board, usually at a lower hourly rate than an H100. Every offer on this page is a VM you can launch now with SSH access, billed by the hour and settled to the second when you stop.
- Offers now
- 16
- From
- $1.35/GPU-hr
- GPUs per VM
- 1x, 2x, 8x
- Providers
- 4
A100 offers you can launch now
The table lists every A100 80 GB VM in stock, cheapest first, and refreshes every 15 minutes.
| GPU | GPUs | Provider | Region | vCPU / RAM / Disk | Interconnect | Per GPU-hr | Per hour | Launch |
|---|---|---|---|---|---|---|---|---|
| A100 PCIe 80GB 80 GB | 1x | Hyperstack | CA | 28 / 120 GB / 850 GB | - | $1.35 | $1.35 | Launch |
| A100 SXM4 80GB 80 GB | 1x | Massed Compute | Virginia, US | 16 / 120 GB / 1000 GB | - | $1.38 | $1.38 | Launch |
| A100 SXM4 80GB 80 GB | 1x | Massed Compute | Iowa, US | 16 / 120 GB / 1000 GB | - | $1.38 | $1.38 | Launch |
| A100 SXM4 80GB 80 GB | 1x | Massed Compute | Iowa, US | 16 / 120 GB / 1000 GB | - | $1.38 | $1.38 | Launch |
| A100 SXM4 80GB 80 GB | 1x | Massed Compute | Iowa, US | 14 / 100 GB / 625 GB | - | $1.38 | $1.38 | Launch |
| A100 SXM4 80GB 80 GB | 1x | QuantaCloud | Midwest, US | 16 / 120 GB / 1000 GB | - | $1.50 | $1.50 | Launch |
| A100 SXM4 80GB 80 GB | 1x | QuantaCloud | Midwest, US | 14 / 100 GB / 625 GB | - | $1.50 | $1.50 | Launch |
| A100 SXM4 80GB 80 GB | 1x | QuantaCloud | Virginia, US | 16 / 120 GB / 1000 GB | - | $1.50 | $1.50 | Launch |
| A100 SXM4 80GB 80 GB | 1x | QuantaCloud | Midwest, US | 16 / 120 GB / 1000 GB | - | $1.50 | $1.50 | Launch |
| A100 SXM4 80GB 80 GB | 1x | Verda | Helsinki, FI | 22 / 120 GB / - | - | $2.00 | $2.00 | Launch |
| A100 PCIe 80GB 80 GB | 2x | Massed Compute | Iowa, US | 30 / 150 GB / 1024 GB | PCIe | $1.35 | $2.70 | Launch |
| A100 PCIe 80GB 80 GB | 2x | QuantaCloud | Midwest, US | 30 / 150 GB / 1536 GB | PCIe | $1.48 | $2.95 | Launch |
For A100 prices across the whole market, compare all providers on GpuPerHour. For a reference price over time, see the A100 rental index (GPUIndexes).
SXM or PCIe
| Variant | Memory | Memory bandwidth | GPU-to-GPU link | Where it makes sense |
|---|---|---|---|---|
| A100 SXM | 80 GB HBM2e | 2,039 GB/s | NVLink, 600 GB/s | Multi-GPU training and fine-tuning that syncs gradients often |
| A100 PCIe | 80 GB HBM2e | 1,935 GB/s | PCIe, or an NVLink bridge for 2 GPUs | Single-GPU inference, fine-tuning and notebooks |
Specifications from NVIDIA's A100 datasheet. NVIDIA rates the PCIe card at 300 W and the SXM card at 400 W, with some HGX A100 boards running a custom thermal design rated up to 500 W. The bandwidth gap between the two variants is small next to the jump to an H100, so the deciding factor is usually NVLink: pick SXM when a job is split across several GPUs, and PCIe when it is not.
No FP8 or FP4 tensor cores
This is the fact that separates the A100 from every newer card on this site. NVIDIA's A100 datasheet lists no FP8 Tensor Core throughput, the format the H100, H200 and B200 all support natively. For BF16 and FP16 training, fine-tuning and most inference, that gap rarely matters. It matters most for checkpoints published in FP8 or MXFP4, such as gpt-oss: the weights still take the same space in memory, but the A100 cannot run that format at native speed the way an H100 or newer card can, so check your serving engine's documentation for how it handles the combination before you commit to one.
What fits in 80 GB
These are weights-only sizes of the published checkpoints, calculated rather than measured. The space left over is what vLLM has for the KV cache after its default 92% memory limit.
| Model | Format | Weights | On one A100 80 GB |
|---|---|---|---|
| Qwen3.8-27B | BF16 | 55.6 GB | Fits, about 18 GB left for the KV cache |
| Gemma 4 31B | BF16 | 62.5 GB | Fits, about 11 GB left |
| Llama 3.3 70B | BF16 | 141.1 GB | Does not fit. Use 4x A100 in BF16, so the KV cache has room |
| gpt-oss-120b | MXFP4 | 65.2 GB | Fits in memory, about 8 GB left, subject to the FP8/FP4 caveat above |
| gpt-oss-20b | MXFP4 | 13.8 GB | Fits easily, but a 48 GB card costs less |
BF16 checkpoints such as Qwen3.8-27B and Gemma 4 31B behave on an A100 exactly as the numbers suggest. The two gpt-oss rows are the ones to plan around: they fit in memory, but the native FP4 math they were built for is not available on this card.
How long until it is ready
A VM usually reaches Running in 1 to 3 minutes. App templates take about 6 to 8 minutes, because the platform waits until the app itself answers before it marks the deployment ready. On some providers the machine reports Running a little before SSH accepts connections, so a Connection refused in the first minute usually clears on its own. The Connection refused fix covers what to check if it does not.
Paid time starts when provisioning starts, but only a launch that reaches Running is charged. A launch that fails refunds the reserved hour automatically.
One GPU or eight
A100 VMs come in fixed sizes of 1, 2, 4 or 8 GPUs, and the GPUs column shows what is in stock right now. The Interconnect column says whether a multi-GPU VM has NVLink between its cards. One VM is the largest unit DeployGPU launches: there is no self-serve multi-node cluster. For more than 8 A100s, or a fixed term at a fixed price, request a cluster quote and a person replies by email. The multi-GPU page goes deeper on how GPUs inside one node talk to each other.
Software on the machine
In the dashboard, every A100 VM boots an Ubuntu image with NVIDIA drivers and Docker installed by the provider, and you connect over SSH as the user shown on the deployment. Three app templates set up a ready app on a private HTTPS link instead: JupyterLab with PyTorch, ComfyUI and Open WebUI with Ollama. For an OpenAI-compatible server on your own A100, follow the vLLM guide.
What you pay
The price on each row is the hourly rate you are billed, including any platform fee. Launching reserves the first hour from your credit balance, and when you stop, the unused seconds of that hour go back to the balance. Credits are bought by card with a 5% + $0.30 processing fee on each top-up. The pricing page has the full breakdown.
Stopping an A100 deletes its disk, so copy results off first. The guide to stopping a GPU without losing work shows how.
Frequently asked questions
How much does an A100 cost per hour on DeployGPU?
The table on this page lists the current hourly price of every A100 80 GB VM you can launch, cheapest first, refreshed every 15 minutes. The price shown is the rate you are billed, including any platform fee.
Does the A100 support FP8?
No, NVIDIA's A100 datasheet lists no FP8 Tensor Core throughput, which arrived with the H100. The A100 still runs BF16 and FP16 workloads well, and it can hold FP8 or MXFP4 checkpoints in memory, just not at native speed.
Should I pick A100 SXM or A100 PCIe?
Pick SXM when a job spans several GPUs and needs the 600 GB/s NVLink between them. Pick PCIe for single-GPU work, where the two variants perform close enough that price is the better tiebreaker.
Can I rent a single A100?
Yes, whenever a 1x offer is in stock. The GPUs column shows how many A100s come with each VM, and you pay only for the time the VM runs.
How long until I can SSH into an A100?
Most VMs reach Running in 1 to 3 minutes. On some providers SSH starts accepting connections a little after the machine reports Running, so retry for a minute before assuming something is wrong.
Is an A100 or an H100 the better choice?
An A100 costs less and handles BF16 training, fine-tuning and inference well. An H100 is the better choice once a workload needs FP8, more memory bandwidth, or the faster 900 GB/s NVLink of an H100 SXM node.
Do I need a contract to rent an A100?
No. Every A100 here is on demand with no minimum term. You prepay credits, launch, and stop whenever you want.