A100 price per hour: rent an NVIDIA A100

An NVIDIA A100 is rented per GPU-hour, and the rate depends on memory size (40 or 80 GB), form factor (SXM or PCIe) and region. The table below shows the best live offer for each A100 variant on EverXCloud, excluding tax.

The A100 is an Ampere GPU rated at 312 TFLOPS in BF16. The SXM 80GB version has 2,039 GB/s of HBM2e bandwidth at 400 W; the PCIe 80GB version has 1,935 GB/s at 300 W. The 40 GB SXM version uses HBM2 at 1,555 GB/s.

SXM boards connect through NVLink 3 at 600 GB/s for multi-GPU training; PCIe boards fit standard servers. For a single-GPU job that fits in memory, the difference is mostly bandwidth and power.

The A100 remains a common choice for fine-tuning, mid-size training and inference of models that fit in 40 or 80 GB, and for scientific codes that need double precision.

Newer GPUs offer more throughput per GPU; compare cost per completed job rather than hourly rate alone.

On an exchange, the rate moves with supply and demand: more offers lower the best price, more demand raises it. Region, GPU count per machine, storage and network included in the configuration also change the price of an offer.

Compare the same variant, memory size and form factor, in the same region. A headline hourly rate rarely describes the full session cost, so add preparation time, storage and any separately billed resources.

If the best offer is above your budget, place a bid at the price you are willing to pay per GPU-hour. The bid is matched when an offer meets it.

During pre-launch, prices shown on EverXCloud are indicative and become rentable on launch day. The table above is read from the live exchange each time the page is generated.