Your real-time inference bill runs about 60% higher than the GPU pricing page shows egress, idle time, the noisy-neighbor tax, and cold starts. Here’s how each is priced on one workload, and how DigitalOcean’s Dedicated Inference designs three of the four out.