NVIDIA L4 24GB vs NVIDIA T4 16GB

The L4 is the newer part: Ada Lovelace, launched in 2023, against the T4's Turing from 2019. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Dense throughput for the L4 against the T4: FP32 30 vs 8.1 TFLOPS, FP16 121 vs 65 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 24 GB against 16 GB, fed at 300 GB/s versus 300 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 70 W for the T4. At FP32 that works out to 0.42 against 0.12 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.