In my project we’re planning to process around 500 mil symbols a day. Right now I am testing a 20$ droplet with 2 CPU’s and 4Gb RAM on digital ocean.
Estimate performance is around 100-120 symbols/second which adds up to 200-250 hours for 1 mil symbols.
Does it scales lineary when adding CPU’s?
What are your estimates for 24 core droplet?
CTranslate2 published benchmarks you can look at.
I’m not sure exactly what hardware you need but CTranslate2 is generally able to use more parallel cores and GPUs well.
Note that of you’re using a GPU you need to set the ARGOS_DEVICE_TYPE=cuda environment variable for Argos Translate.
I would look at adding a load balancer and use multiple droplets if you need to process a lot of requests, rather than adding more CPUs to a single droplet.