Benchmarking Sugoi Translator (Patreon)

I did a little experiment on how Translator++ load balancing multiple instance of Sugoi Translator server. 

As you probably already know, Translator++ can use multiple instances of the Sugoi Translator server. In the latest version of Translator++ we can even manage these servers easily. We can even quickly switch from CPU to GPU processing.

Spec:

Core i7 7700 16 GB RAM

NVIDIA GTX 1050 2GB RAM 

Setup:

121 row of data

5 Batch per request (This is the stable number of batch that my potato GPU can take)

Result:

1 CPU:

64238ms

1 GPU:

23204ms

1CPU & 1GPU:

22534ms

2CPU:

49028ms

Verdict:

We can get extra speed by running another instance of the server. In this case we can cut down 23.6% execution time by running two instances of the server compared to one CPU based server.

But unexpected results appear when I run the server CPU and GPU together. There is an increase of speed but very insignificant. Of course we can expect more performance if the servers are run on different machines.