NVIDIAAI GPU · PCIe (low profile)
L4
NVIDIA L4 is a Ada Lovelace data-center accelerator with 24 GB GDDR6 and 0.3 TB/s of memory bandwidth.
Specifications
Dense throughput, no sparsity| Architecture | Ada Lovelace |
|---|---|
| Form factor | PCIe (low profile) |
| Process | TSMC 4N |
| FP16 / BF16 tensor (dense) | 121 TFLOPS |
| FP8 tensor (dense) | 242.5 TFLOPS |
| FP4 tensor (dense) | — |
| Memory | 24 GB GDDR6 |
| Memory bandwidth | 0.3 TB/s |
| Power | 72 W |
| Launch | Mar 2023 |
AI index
Tensor × bandwidth100 × √((121 ÷ 989.4) × (0.3 ÷ 3.35)) = 10.5
The AI index is the geometric mean of dense 16-bit tensor throughput and memory bandwidth, each relative to the H100 SXM. It is a spec-sheet comparison, not a measured benchmark.