Fact · Speech recognition
2926MB of VRAM
In the faster-whisper README's GPU benchmark, faster-whisper in int8 precision used 2926 MB of GPU memory with the large-v2 model.
Checked against its source on October 17, 2026
In the source's own words
“| faster-whisper | int8 | 5 | 59s | 2926MB |”
faster-whisper: README (GitHub, SYSTRAN), SYSTRAN, GitHub (SYSTRAN/faster-whisper), 2026, Large-v2 model on GPU (benchmark table).
Open the original- What it measures
- The benchmark ran on an NVIDIA RTX 3070 Ti with 8 GB of memory, CUDA 12.4 and beam size 5. The same row shows the 59 second run time.
- What it does not show
- One GPU and one recording. The README gives no figure for other GPUs, so treat the number as a guide to the scale of memory needed.