In production one would like to use a server for this. The problem is that servers usually don't have graphic cards to do the learning. Virtual Windows servers can be rented for 25-50 euro a month, and I one would like to find an AI solution for an amount in the same range.
So what can one do?
- A top of the class NVIDIA DGX-1 with Tesla V100 costs 150.000 euro, and offers a 1000 TFlops performance.
- Windows Azure seems to offer a reasonable solution for about 100 euro/month. The machine-learning-studio S1-tier includes 100,000 API transactions and 25 API compute hours.
- LeaderGPU offers servers with GPUs that starts with 365 euro/pm (2x1080GTX)
Nvidia changed the EULA: Geforce cards are no longer allowed in datacenters.
Geen opmerkingen:
Een reactie posten