NVIDIA H20 vs NVIDIA L40S
Side-by-side comparison of specifications, performance, and efficiency.
NVIDIA H20
Architecture
Hopper
Year
2024
VRAM
96 GB
TDP
400 W
NVIDIA L40S
Architecture
Ada Lovelace
Year
2023
VRAM
48 GB
TDP
350 W
Specifications
NVIDIA H20 NVIDIA L40S
Architecture
Hopper Ada Lovelace
Launch Year
2024 2023
Form Factor
SXM PCIe
VRAM
96 GB48 GB
Memory Bandwidth
4.0 TB/s864 GB/s
TDP
400 W350 W
Process Node
4nm 4nm
| Spec | NVIDIA H20 | NVIDIA L40S |
|---|---|---|
| Architecture | Hopper | Ada Lovelace |
| Launch Year | 2024 | 2023 |
| Form Factor | SXM | PCIe |
| VRAM | 96 GB | 48 GB |
| Memory Bandwidth | 4.0 TB/s | 864 GB/s |
| TDP | 400 W | 350 W |
| Process Node | 4nm | 4nm |
Performance (TFLOPS)
NVIDIA H20 NVIDIA L40S
FP64
1.0 TFlops1.4 TFlops
FP32
44.0 TFlops91.6 TFlops
FP16
148.0 TFlops733.0 TFlops
BF16
148.0 TFlops733.0 TFlops
FP8
—733.0 TFlops
INT8
296.0 TFlops733.0 TFlops
| Precision | NVIDIA H20 | NVIDIA L40S |
|---|---|---|
| FP64 | 1.0 TFlops | 1.4 TFlops |
| FP32 | 44.0 TFlops | 91.6 TFlops |
| FP16 | 148.0 TFlops | 733.0 TFlops |
| BF16 | 148.0 TFlops | 733.0 TFlops |
| FP8 | — | 733.0 TFlops |
| INT8 | 296.0 TFlops | 733.0 TFlops |
FLOPS by Precision
Efficiency — TDP vs FP32 Performance
Analysis
Most Efficient
NVIDIA
L40S
0.26 TFLOPS/W
Get Comparison Updates
New GPUs added weekly. Be the first to see how they compare.