NVIDIA H100 SXM5 80GB vs NVIDIA A100 SXM4 80GB

The H100 delivers 3.2x the FP16 throughput of the A100 (990 vs 312 TFLOPS dense).

H100: Hopper, 2022 A100: Ampere, 2020

FP16 dense lead

3.2x

H100: 990 vs 312 TFLOPS

Memory

80 vs 80 GB

Same capacity

Bandwidth

3.4 TB/s vs 2.0 TB/s

H100 moves data faster

TDP

700 W vs 400 W

A100 draws 300 W less

See rental pricing from $1.01/hr Download comparison sheet Printable version →

NVIDIA H100 SXM5 80GB

Leads on FP16 Full specs →

What is the NVIDIA H100?

The H100 is a GPU from NVIDIA on the Hopper architecture, an accelerator for AI training, inference and high-performance computing. It is an SXM module, the mezzanine format used in dense multi-GPU servers. H100 full specifications →

Can I rent the H100 in the cloud?

Yes. Cloud providers tracked by Flopper rent the H100 by the hour, and the pricing page compares their current rates and the price history side by side. H100 rental prices →

How does the H100 compare with other accelerators?

This page sets the H100 against the A100. The H100 is the newer generation of the two. Both are NVIDIA parts, and both ship as an SXM module, so they fit the same class of system. Flopper also compares it head to head with the H200 SXM 141GB, Instinct MI300X 192GB and B300 SXM 262GB, across every published precision, memory and power. H100 vs H200 SXM 141GB →

Where do the H100 figures come from?

Every H100 figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

NVIDIA A100 SXM4 80GB

Full specs →

What is the NVIDIA A100?

The A100 is a GPU from NVIDIA on the Ampere architecture, an accelerator for AI training, inference and high-performance computing. It is an SXM module, the mezzanine format used in dense multi-GPU servers. A100 full specifications →

Can I rent the A100 in the cloud?

Yes. Cloud providers tracked by Flopper rent the A100 by the hour, and the pricing page compares their current rates and the price history side by side. A100 rental prices →

How does the A100 compare with other accelerators?

This page sets the A100 against the H100. The A100 is the older generation of the two. Both are NVIDIA parts, and both ship as an SXM module, so they fit the same class of system. Flopper also compares it head to head with the B200 SXM 180GB and A100 PCIe 40GB, across every published precision, memory and power. A100 vs B200 SXM 180GB →

Where do the A100 figures come from?

Every A100 figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

Specifications

H100

A100

Architecture

HopperAmpere

Launch Year

20222020

Form Factor

SXMSXM

Memory

80 GB80 GB

Memory Bandwidth

3.4 TB/s2.0 TB/s

TDP

700 W400 W

Process Node

4nm7nm

Spec

H100

A100

Architecture HopperAmpere

Launch Year 20222020

Form Factor SXMSXM

Memory 80 GB80 GB

Memory Bandwidth 3.4 TB/s2.0 TB/s

TDP 700 W400 W

Process Node 4nm7nm

Performance (TFLOPS)

H100

A100

34 TFLOPS 9.7 TFLOPS

67 TFLOPS 20 TFLOPS

494 TFLOPS

989 TFLOPS sparse

156 TFLOPS

312 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

1,979 TFLOPS

3,958 TFLOPS sparse

FP6

1,979 TOPS

3,958 TOPS sparse

624 TOPS

1,248 TOPS sparse

Tensor Core

67 TFLOPS

Precision

H100

A100

34 TFLOPS 9.7 TFLOPS

67 TFLOPS 20 TFLOPS

494 TFLOPS

989 TFLOPS sparse

156 TFLOPS

312 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

990 TFLOPS

1,979 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

1,979 TFLOPS

3,958 TFLOPS sparse

FP6

1,979 TOPS

3,958 TOPS sparse

624 TOPS

1,248 TOPS sparse

Tensor Core

67 TFLOPS

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

Divide the price ratio by the bottleneck before choosing. At the floor the H100 costs $1.41 per GPU-hour against $0.75 for the A100: under twice the rate for three times the dense compute, 990 TFLOPS BF16 against 312 TFLOPS. Training, fine-tuning and prefill finish cheaper on the H100.

Decode is where the A100 holds on. Token generation is limited by memory bandwidth, and 3.35 TB/s against 2.0 TB/s is about 1.6 times, no more than the price ratio. Serving models that fit in 80 GB, BF16 code and spot-heavy batch jobs stay cheaper per token on the A100.

FP8 is the H100’s real lever. Hopper’s Transformer Engine runs it at 1,979 TFLOPS dense, and FP8 weights halve the bytes moved per token, which tilts decode too; Ampere has no FP8 path. Compare live rates on the H100 pricing page or step up to H200 vs H100.

New to the terms? The glossary explains FP64, FP32, TF32, BF16, FP16, FP8, INT8, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

How much faster is the H100 than the A100?

About three times on dense BF16 or FP16: 990 TFLOPS against 312 TFLOPS. Memory bandwidth grows less, 3,350 GB/s against 2,039 GB/s, so bandwidth-bound token generation gains about 1.6 times. FP8 adds more, at 1,979 TFLOPS dense, but Ampere has no FP8 path.

Should I rent H100s or stay on A100s?

Move training, fine-tuning and prefill-heavy serving to the H100, where three times the compute outruns a rate under twice as high. Stay on A100s for decode-heavy serving of models that fit in 80 GB, BF16 code with no FP8 path, and spot-friendly batch jobs that can wait.

Do the H100 and A100 have the same memory?

Both carry 80 GB, so neither holds a larger model than the other. The difference is speed: the H100's HBM3 moves 3,350 GB/s against 2,039 GB/s of HBM2e on the A100. Both split into up to seven MIG instances of 10 GB each.

How much does it cost to rent an H100 compared with an A100?

Flopper lists 67 H100 SXM5 80GB listings across 34 providers, from $1.41 to $12.29 per GPU-hour, and 56 A100 SXM4 80GB listings across 26 providers, from $0.75 to $5.04. Both floors are community capacity; on-demand starts at $1.79 for the H100 and $1.07 for the A100.

Where to Rent

NVIDIA H100 SXM5 80GB

Rent from

$1.73 / GPU-hr

Lowest of 66 live listings across 34 providers · median $3.85

See all H100 rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

NVIDIA A100 SXM4 80GB

Rent from

$1.01 / GPU-hr

Lowest of 58 live listings across 27 providers · median $1.79

See all A100 rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → H100 full specs A100 full specs Estimate cluster cost in the calculator →

NVIDIA H100 SXM5 80GB vs NVIDIA A100 SXM4 80GB

NVIDIA H100 SXM5 80GB

What is the NVIDIA H100?

Can I rent the H100 in the cloud?

How does the H100 compare with other accelerators?

Where do the H100 figures come from?

NVIDIA A100 SXM4 80GB

What is the NVIDIA A100?

Can I rent the A100 in the cloud?

How does the A100 compare with other accelerators?

Where do the A100 figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

How much faster is the H100 than the A100?

Should I rent H100s or stay on A100s?

Do the H100 and A100 have the same memory?

How much does it cost to rent an H100 compared with an A100?

Where to Rent

NVIDIA H100 SXM5 80GB

NVIDIA A100 SXM4 80GB

Related comparisons

Get Comparison Updates

Stay Updated on GPU Specs