Replicate Multi-GPU logo

Replicate Multi-GPU

Functions 4.6/10

Per-second multi-GPU hardware: 2x A100 80GB $10.08/hr, 4x A100 $20.16/hr, 8x A100 $40.32/hr, 2x H100 $10.98/hr, 8x H100 $43.92/hr, L40S multi up to 8x $28.08/hr.

Replicate Multi-GPU scores 4.6/10 for free value in Functions. The main constraint: Paid — 8x H100 $0.012200/s ($43.92/hr); 2x A100 $0.0028/s. Per-second multi-GPU hardware: 2x A100 80GB $10.08/hr, 4x A100 $20.16/hr, 8x A100 $40.32/hr, 2x H100 $10.98/hr, 8x H100 $43.92/hr, L40S multi up to 8x $28.08/hr.

Category Functions Kind Paid plan

Free value score

4.6 / 10
Show the math
Overall 4.6/10
value = 4.6/100
weighted_components_v1
Invocation value 1/10
value = 0.01 ×
fixed_log_v1
Runtime fit 8.2/10
value = 8.2/100
policy_score_v1
Deploy friction 8/10
value = 8/100
policy_score_v1

Key metrics

8xH100 $44h
8xA100 $40h
$12.2m/s

What you get

Paid — 8x H100 $0.012200/s ($43.92/hr); 2x A100 $0.0028/s. Per-second multi-GPU hardware: 2x A100 80GB $10.08/hr, 4x A100 $20.16/hr, 8x A100 $40.32/hr, 2x H100 $10.98/hr, 8x H100 $43.92/hr, L40S multi up to 8x $28.08/hr.

Linear per-GPU scaling (e.g. $5.49/hr per H100) with per-second billing; good for running big models on-demand without negotiating clusters.

Sources

paid value serverless-functions