NVIDIA A100 SXM4 80GB vs NVIDIA A100 PCIe 40GB

The A100 SXM4 80GB and the A100 PCIe 40GB deliver near-identical FP16 throughput (312 vs 312 TFLOPS dense). The A100 SXM4 80GB also carries 40 GB more memory (80 GB vs 40 GB).

A100 SXM4 80GB: Ampere, 2020 A100 PCIe 40GB: Ampere, 2020
VRAM
80 vs 40 GB
40 GB more for the A100 SXM4 80GB
Bandwidth
2.0 TB/s vs 1.6 TB/s
A100 SXM4 80GB moves data faster
TDP
400 W vs 250 W
A100 PCIe 40GB draws 150 W less

Specifications

A100 SXM4 80GB
A100 PCIe 40GB
Architecture
AmpereAmpere
Launch Year
20202020
Form Factor
SXMPCIe
VRAM
80 GB40 GB
Memory Bandwidth
2.0 TB/s1.6 TB/s
TDP
400 W250 W
Process Node
7nm7nm

Performance (TFLOPS)

A100 SXM4 80GB
A100 PCIe 40GB
FP64
9.7 TFlops 9.7 TFlops
FP32
19.5 TFlops 19.5 TFlops
FP16
312.0 TFlops
624.0 TFLOPS sparse
312.0 TFlops
624.0 TFLOPS sparse
BF16
312.0 TFlops
624.0 TFLOPS sparse
312.0 TFlops
624.0 TFLOPS sparse
INT8
624.0 TFlops
1248.0 TOPS sparse
624.0 TFlops
1248.0 TOPS sparse

FLOPS by Precision

What actually differs

Both GPUs launched in 2020: the A100 SXM4 80GB on Ampere and the A100 PCIe 40GB on Ampere.

Dense throughput for the A100 SXM4 80GB against the A100 PCIe 40GB: FP64 9.7 vs 9.7 TFLOPS, FP32 20 vs 20 TFLOPS, FP16 312 vs 312 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 80 GB against 40 GB, fed at 2.0 TB/s versus 1.6 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

Power budgets are 400 W for the A100 SXM4 80GB and 250 W for the A100 PCIe 40GB. At FP32 that works out to 0.05 against 0.08 TFLOPS per watt.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.

Frequently asked questions

Is the A100 SXM4 80GB faster than the A100 PCIe 40GB?

They are close on paper: both deliver about 312 TFLOPS of dense FP16 throughput. Memory, bandwidth, and power are the deciding differences.

Which has more memory, the A100 SXM4 80GB or the A100 PCIe 40GB?

The A100 SXM4 80GB carries 80 GB of VRAM versus 40 GB for the A100 PCIe 40GB. Memory bandwidth is 2.0 TB/s for the A100 SXM4 80GB and 1.6 TB/s for the A100 PCIe 40GB.

How much power do the A100 SXM4 80GB and the A100 PCIe 40GB draw?

The A100 SXM4 80GB is rated at 400 W TDP and the A100 PCIe 40GB at 250 W. On FP32 throughput per watt, the A100 PCIe 40GB is the more efficient part.

Can I rent the A100 SXM4 80GB or the A100 PCIe 40GB in the cloud?

Yes. Live cloud listings tracked by Flopper start at $0.90 per GPU hour. The rental pricing section on this page lists current providers and rates for both GPUs.

Where to Rent

NVIDIA A100 SXM4 80GB

ProviderConfigurationPrice/GPU-hrChecked
DataCrunch
1× A100 SXM4 Spot $0.90 5h ago View →
CoreWeave
8× A100 SXM4 Spot $1.21 5h ago View →
RunPod
1× A100 SXM4 Community $1.39 5h ago View →
Denvr Dataworks
8× A100 SXM4 $1.40 5h ago View →
Denvr Dataworks
1× A100 SXM4 $1.40 5h ago View →
RunPod
1× A100 SXM4 $1.59 5h ago View →
Hyperstack
1× A100 SXM4 $1.60 5h ago View →
DataCrunch
1× A100 SXM4 $1.79 5h ago View →
AceCloud
1× A100 SXM4 $1.81 5h ago View →
Lambda Labs
1× A100 SXM4 $1.99 5h ago View →
Crusoe
1× A100 SXM4 $2.30 5h ago View →
CoreWeave
8× A100 SXM4 $2.70 5h ago View →
Paperspace
1× A100 SXM4 $3.19 5h ago View →
Paperspace
1× A100 SXM4 $3.28 5h ago View →
Microsoft Azure
8× A100 SXM4 $4.10 11h ago View →
Microsoft Azure
8× A100 SXM4 $4.10 11h ago View →
All A100 SXM4 80GB listings and price history →

Get Comparison Updates

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