NVIDIA L4 24GB vs NVIDIA L40 48GB

Both GPUs launched in 2023: the L4 on Ada Lovelace and the L40 on Ada Lovelace.

Dense throughput for the L4 against the L40: FP32 30 vs 91 TFLOPS, FP16 121 vs 181 TFLOPS, FP8 243 vs 362 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 24 GB against 48 GB, fed at 300 GB/s versus 864 GB/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 72 W for the L4 and 300 W for the L40. At FP32 that works out to 0.42 against 0.30 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.