Start Building
In stock · Provisions in ~5 min

NVIDIA Ampere, 80 GB HBM2e, SXM

NVIDIA A100 80GBThe proven training workhorse.

The NVIDIA A100 80GB is the Ampere-generation flagship GPU with 80 GB of HBM2e memory and 2 TB/s of bandwidth. The industry standard for LLM training and fine-tuning. Runs 30B models at FP16 and supports 70B with tensor parallelism. Available on packet.ai from $1.43/GPU-hour.

From $1.43/GPU-hr~28% below H100 cost

Dedicated $1.43/hr · Monthly $940/mo

80GB
HBM2e memory
2TB/s
Memory bandwidth
312TFLOPS
TF32 compute
600GB/s
NVLink bandwidth
Architecture

Ampere HBM2e — the training standard.

The A100 combines 80 GB HBM2e with third-generation NVLink for multi-GPU training at scale.

3rd-gen Tensor Cores

TF32 training and FP16 inference with sparsity. Established as the industry baseline for mixed-precision training.

80 GB HBM2e at 2.0 TB/s

Enough memory for 7B to 30B models natively. The sweet spot for cost-effective LLM work.

NVLink 3.0 at 600 GB/s

Connect up to 8 GPUs in an HGX node with NVSwitch for distributed training at scale.

MIG: 7 instances per GPU

Partition a single A100 into up to 7 isolated MIG instances for multi-tenant inference and dev/test.

Technical specs

NVIDIA A100 80GB specifications.

SpecificationValueGreat for
GPU architecture
NVIDIA Ampere
Industry standard for mixed-precision training.
GPU memory
80 GB HBM2e
30B at FP16 and 70B at 4-bit.
Memory bandwidth
2 TB/s
Fast gradient movement in training loops.
TF32 compute
312 TFLOPS
Mixed-precision training baseline.
NVLink
3.0 · 600 GB/s
Multi-GPU tensor parallel training.
MIG
Up to 7 instances
Efficient multi-tenant inference serving.
Pricing

Three ways to run A100.

Dedicated or monthly — plus multi-node clusters.

DedicatedMonthly · Single-tenant
$940 /month
35% off hourly rate

Reserved A100 at a flat monthly rate. Full single-tenant isolation, predictable cost.

Deploy Monthly →
Multi-node Cluster
From 8 GPUs

Scale training across multiple A100 nodes with NVLink and InfiniBand interconnect.

  • 8–512 GPUs per cluster
  • NVLink + InfiniBand
  • Provisioned in <1 hr
Get a wholesale quote →
Use cases

What the A100 is built for.

LLM fine-tuning

80 GB handles 30B at FP16 natively. Industry-standard toolchain support.

  • 30B at FP16 native
  • 70B at 4-bit
  • MIG for smaller models

Distributed training

NVLink 3.0 and SXM form factor make A100 clusters the standard for large-scale training.

  • 600 GB/s NVLink
  • Standard toolchain
  • Multi-node clusters

Production inference

80 GB HBM2e and MIG support make A100 the flexible choice for serving multiple models.

  • Up to 7 MIG instances
  • 99.99% SLA
  • Monthly flat billing
FAQ

NVIDIA A100, answered.

For anything else, reach help@packet.ai.

What is the NVIDIA A100 80GB?

Ampere flagship — 80 GB HBM2e, 2 TB/s bandwidth, industry standard for LLM training and fine-tuning.

How much does the A100 cost?

$1.43/GPU-hour dedicated, or $940/month flat rate.

What models fit in A100 80GB?

30B at FP16 natively, 70B at 4-bit. For full FP16 70B, use H200 or B200.

How does A100 compare to H100?

H100 SXM is ~3× faster on transformer workloads and supports FP8. A100 is ~28% cheaper and has wider toolchain support.

Does A100 support MIG?

Yes — up to 7 isolated MIG instances, down to 10 GB each. Great for multi-model inference serving.

How fast can I deploy A100?

SSH-ready in under 5 minutes on Dedicated hourly. Multi-node clusters provision in under 1 hour.

Run the A100 80GB. From $1.43/hr.

The industry standard for LLM training. From $1.43/hr dedicated or $940/mo flat.

On-demand · hourly billing · EU region (Frankfurt)

NVIDIA A100 80GBfrom $1.43/GPU-hr In stock
Deploy A100 →