Custom Functions and Models
GPU Acceleration
Accelerate your Sieve functions with GPU
Sieve let’s you run code on GPUs. To do so, all you need to do is add a gpu
tag to your function with the GPU you want. Today, Sieve offers a couple machine configurations listed below.
Name | GPU | Memory (GB) | vCPUs | Tag |
---|---|---|---|---|
T4 (default) | T4 | 16 | 4 | t4 |
A100-40GB | A100-40GB | 85 | 12 | a100 |
A100-20GB | A100-20GB | 42.5 | 6 | a100-20gb |
V100 | V100 | 16 | 4 | v100 |
L4 | L4 | 32 | 8 | l4 |
You can specify one of these in code as follows.
@sieve.function(
name="gpu_function",
gpu="t4" # Tag
)
def runs_on_gpu(video: sieve.Video) -> sieve.Video:
...
Read more about the way this field works in SDK reference.