From the source
Replicate announces the addition of
NVIDIA L40S GPUs to its supported hardware types, which are approximately 40% faster than A40 GPUs.
The company will phase out A40 GPUs over several weeks, migrating all models and deployments to L40S GPUs.
Pricing for private models and deployments remains the same, while public models and training will incur a new rate of $3.51 per hour.
Users are advised to adjust deployment minimum instances to avoid higher charges.
