NVIDIA L4 vs NVIDIA Tesla V100 (32 GB)

The NVIDIA L4 is a workstation graphics card that launched in Q1 2023. It is built on the Ada Lovelace GPU microarchitecture (codename AD104) and is manufactured on a 5 nm process.

Memory

The L4 has 24 GB of GDDR6 memory, with a 1,563 MHz memory clock and a 192 bit interface. This gives it a memory bandwidth of 300.1 Gb/s, which affects how fast it can transfer data to and from memory. GPU memory stores temporary data that helps the GPU with complex math and graphics operations. More memory is generally better, as not having enough can cause performance bottlenecks.

Cores and Clock Speeds

The L4 includes 7,680 CUDA cores, the processing units for handling parallel computing tasks. The GPU operates at a core clock speed of 795 MHz and can dynamically boost its clock speed up to 2,040 MHz. Complementing the processing units are 240 texture mapping units (TMUs) for efficient texture filtering and 80 render output units (ROPs) for pixel processing. Additionally, the GPU features 240 tensor cores optimized for AI-accelerated workloads and 60 RT cores dedicated to real-time ray tracing calculations.

Compatibility & Power Consumption

The L4 is a compact, low-profile graphics card that fits into 1 PCIe expansion slot. NVIDIA recommends a power supply of at least 250 W to handle the GPU's thermal design power (TDP) of 72 W.