NVIDIA A100 SXM4 80GB vs NVIDIA B200 SXM 180GB

The B200 delivers 7.2x the FP16 throughput of the A100 (2,250 vs 312 TFLOPS dense). The B200 also carries 100 GB more memory (180 GB vs 80 GB).

A100: Ampere, 2020 B200: Blackwell, 2024

FP16 dense lead

7.2x

B200: 2,250 vs 312 TFLOPS

Memory

80 vs 180 GB

100 GB more for the B200

Bandwidth

2.0 TB/s vs 7.7 TB/s

B200 moves data faster

TDP

400 W vs 1.0 kW

A100 draws 600 W less

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

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 B200. 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 H100 SXM5 80GB, A100 PCIe 40GB and H200 SXM 141GB, across every published precision, memory and power. A100 vs H100 SXM5 80GB →

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 →

NVIDIA B200 SXM 180GB

Leads on FP16 Full specs →

What is the NVIDIA B200?

The B200 is a GPU from NVIDIA on the Blackwell 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. B200 full specifications →

Can I rent the B200 in the cloud?

Yes. Cloud providers tracked by Flopper rent the B200 by the hour, and the pricing page compares their current rates and the price history side by side. B200 rental prices →

How does the B200 compare with other accelerators?

This page sets the B200 against the A100. The B200 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 B100 SXM 192GB and B300 SXM 262GB, across every published precision, memory and power. B200 vs B100 SXM 192GB →

Where do the B200 figures come from?

Every B200 figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

Specifications

A100

B200

Architecture

AmpereBlackwell

Launch Year

20202024

Form Factor

SXMSXM

Memory

80 GB180 GB

Memory Bandwidth

2.0 TB/s7.7 TB/s

TDP

400 W1.0 kW

Process Node

7nm4nm

Spec

A100

B200

Architecture AmpereBlackwell

Launch Year 20202024

Form Factor SXMSXM

Memory 80 GB180 GB

Memory Bandwidth 2.0 TB/s7.7 TB/s

TDP 400 W1.0 kW

Process Node 7nm4nm

Performance (TFLOPS)

A100

B200

9.7 TFLOPS 37 TFLOPS

20 TFLOPS 75 TFLOPS

156 TFLOPS

312 TFLOPS sparse

1,125 TFLOPS

2,250 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

2,250 TFLOPS

4,500 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

2,250 TFLOPS

4,500 TFLOPS sparse

4,500 TFLOPS

9,000 TFLOPS sparse

FP6

4,500 TFLOPS

9,000 TFLOPS sparse

9,000 TFLOPS

18,000 TFLOPS sparse

624 TOPS

1,248 TOPS sparse

4,500 TOPS

9,000 TOPS sparse

Precision

A100

B200

9.7 TFLOPS 37 TFLOPS

20 TFLOPS 75 TFLOPS

156 TFLOPS

312 TFLOPS sparse

1,125 TFLOPS

2,250 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

2,250 TFLOPS

4,500 TFLOPS sparse

312 TFLOPS

624 TFLOPS sparse

2,250 TFLOPS

4,500 TFLOPS sparse

4,500 TFLOPS

9,000 TFLOPS sparse

FP6 4,500 TFLOPS

9,000 TFLOPS sparse

9,000 TFLOPS

18,000 TFLOPS sparse

624 TOPS

1,248 TOPS sparse

4,500 TOPS

9,000 TOPS sparse

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

Skipping Hopper makes the A100 to B200 move a decision per workload, not per cluster. Training and prefill are compute-bound, and the B200 delivers about seven times the dense BF16 compute for roughly four times the floor rental price: $3.06 against $0.75 per GPU-hour. FP8 and FP4 widen the gap, but Ampere has no path to either, so that gain needs requantised weights.

Decode is the trap. Token generation is bound by memory bandwidth, and 7.7 TB/s is under four times the A100’s, so a model that fits in 80 GB buys about the same bytes per dollar on either part.

Retire A100s from training first and keep them serving sub-80 GB models, BF16 fine-tunes and MIG-sliced jobs. Move serving once models outgrow 80 GB; the B200 holds 180 GB (announced at 192 GB). Compare live rates on the B200 pricing page or the one-generation step in H100 vs A100.

New to the terms? The glossary explains FP64, FP32, TF32, BF16, FP16, FP8, FP4, INT8, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

How much faster is the B200 than the A100?

On dense BF16, about seven times: 2,250 TFLOPS against 312 TFLOPS. Memory bandwidth grows less, 7,700 GB/s against 2,039 GB/s, so bandwidth-bound decode gains under four times. FP8 and FP4 add more on the B200, but Ampere has no path to either, so those rows have no A100 counterpart.

Should I replace my A100s with B200s?

Move training and prefill-heavy work first, where the compute gap outruns the price gap. Keep A100s for serving models that fit in 80 GB, BF16 fine-tunes and MIG-partitioned jobs. Plan new servers too: the B200 is a 1,000 W SXM part, not a drop-in for a 400 W A100 board.

Does the B200 have 192 GB or 180 GB of memory?

The shipping B200 carries 180 GB of HBM3e. NVIDIA announced it at GTC 2024 with 192 GB, and many spec tables still carry that figure. NVIDIA's DGX B200 datasheet lists 1,440 GB across eight GPUs, which is 180 GB each, against 80 GB on the A100 SXM4.

How much does it cost to rent a B200 compared with an A100?

Flopper tracks 37 B200 listings across 25 providers, from $3.06 to $14.24 per GPU-hour, and 56 A100 SXM4 80GB listings across 26 providers, from $0.75 to $5.04. Both floors are spot or community capacity, so compare on-demand rates before planning around them.

Where to Rent

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.

NVIDIA B200 SXM 180GB

Rent from

$3.59 / GPU-hr

Lowest of 35 live listings across 25 providers · median $6.92

See all B200 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 → A100 full specs B200 full specs Estimate cluster cost in the calculator →

NVIDIA A100 SXM4 80GB vs NVIDIA B200 SXM 180GB

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?

NVIDIA B200 SXM 180GB

What is the NVIDIA B200?

Can I rent the B200 in the cloud?

How does the B200 compare with other accelerators?

Where do the B200 figures come from?

Specifications

Performance (TFLOPS)

FLOPS by Precision

What actually differs

Frequently asked questions

How much faster is the B200 than the A100?

Should I replace my A100s with B200s?

Does the B200 have 192 GB or 180 GB of memory?

How much does it cost to rent a B200 compared with an A100?

Where to Rent

NVIDIA A100 SXM4 80GB

NVIDIA B200 SXM 180GB

Related comparisons

Get Comparison Updates

Stay Updated on GPU Specs