Why is the A100 PCIE FP32 figure lower than some RTX cards?

The 15.6 TFLOPS number is plain FP32. Most deep learning runs in BF16 or FP16 mixed precision, and LLM inference is usually limited by memory bandwidth, where the A100 PCIE's 1313 GB/s is well ahead of consumer Ampere cards.

Which LLMs fit in 40 GB of VRAM?

As a rule of thumb, 16-bit weights take about 2 GB per billion parameters, so 13B models fit with room for KV cache. 30B–34B models fit when quantized to 4 or 8 bits. 70B models need an 80 GB card such as the A800 PCIE or H100.

What is the difference between the A100 PCIE and the A100 SXM4?

Both are listed with 40 GB, 15.6 TFLOPS FP32 and 1313 GB/s. The PCIE card runs at a 250 W power limit and the SXM4 at 400 W, which gives the SXM4 more thermal headroom under sustained load.

How do I start an A100 PCIE server on Cloudj?

Choose a template, pick a live A100 PCIE configuration (filter by country if location matters), select your SSH key and confirm the price. Then connect over SSH with the host and port shown on the instance page.

GPU Cloud · NVIDIA · Ampere

Rent an NVIDIA A100 PCIE GPU with 40 GB VRAM

Name: NVIDIA A100 PCIE cloud GPU
Brand: Cloudj
Availability: InStock

Rent A100 PCIE cloud GPU servers by the hour on Cloudj: 40 GB VRAM, Ampere architecture and about 1313 GB/s of memory bandwidth for LLM and training jobs. It is the practical datacenter step for ML engineers whose models no longer fit on a gaming card, with 15.6 TFLOPS FP32, a 250 W power limit and compute capability 8.0. Pick a live configuration, start from the Ubuntu 22.04 + CUDA or PyTorch + JupyterLab template and connect over SSH with your key.

Rent A100 PCIE All GPUs

A100 PCIE specifications

VRAM per GPU

40 GB

Architecture

NVIDIA Ampere

Compute capability

FP32 compute

≈ 15.6 TFLOPS

Memory bandwidth

≈ 1313 GB/s

Power limit

250 W

Values reported by the infrastructure provider for machines currently listed; real performance varies by machine.

Live A100 PCIE configurations and prices

From $0.50/h

incl. 32 GB disk · Live price, updated every 15 minutes

Deploy

What you can run on the A100 PCIE cloud GPU

Serve 13B–34B models with vLLM

40 GB holds a 13B model in 16-bit or a quantized 30B-class model, and 1313 GB/s keeps token generation moving under concurrent requests.

LoRA and QLoRA fine-tuning

Fine-tune 7B–13B language models on your own data with Unsloth or plain PyTorch in BF16, without squeezing batch size to fit.

Mixed-precision training runs

Train vision, speech or tabular models in BF16 on a stable datacenter card instead of renting a full multi-GPU node.

Embeddings and batch inference

Push large document sets through embedding or classification models overnight and delete the server when the batch is done.

Is the A100 PCIE the right GPU for your workload?

Choose the A100 PCIE 40 GB when your workloads exceed the memory limits of consumer GPUs and require high-bandwidth datacenter hardware. Opt for a smaller consumer GPU for lightweight inference, or step up to a higher-VRAM datacenter model if your unquantized models or batch sizes exceed 40 GB.

A100 PCIE vs similar cloud GPUs

GPU	VRAM per GPU	FP32 compute	Power limit
NVIDIA A100 PCIE	40 GB	≈ 15.6 TFLOPS	250 W
NVIDIA A100 SXM4	40 GB	≈ 15.6 TFLOPS	400 W	Details →
NVIDIA A40	48 GB	≈ 29.9 TFLOPS	300 W	Details →
NVIDIA L40S	48 GB	≈ 73.3 TFLOPS	350 W	Details →
NVIDIA Tesla P40	24 GB	≈ 9.4 TFLOPS	250 W	Details →

Guides for A100 PCIE

A100 cloud Europe guide: renting A100 GPUs in the EUA100 cloud Europe guide: PCIe vs SXM4 on the GPU list, what the 40 GB versions handle, searching EU codes and when an A100 beats an H100 on value.Rent A100 USA: PCIe and SXM4 options billed by the minuteRent A100 USA hosts for training and batch work: PCIe vs SXM4 in practice, multi-GPU listings, and a cost estimate that includes storage while stopped.A100 cloud in Asia: Singapore, Japan and Korea searchesPlan an A100 cloud Singapore Japan search: the SG, JP, KR and HK codes, the A100 variants listed, 48 GB alternatives in the region and moving data from Vietnam.L40S cloud Europe: an inference and rendering GPU in the EURent an L40S cloud Europe machine for mixed inference and rendering: when 48 GB helps, how to search EU codes and how it compares with A100 and RTX 6000 Ada.

All articles →

A100 PCIE cloud GPU: frequently asked questions

Rent the A100 PCIE cloud GPU by the hour

Pick a live configuration, choose a template and connect over SSH with your key. Billing stops when you destroy the instance.

Rent A100 PCIE All GPUs