Technical requirements for Adaptable Language Pairs
The technical requirements for obtaining Adaptable Language Pairs refer to: minimum system requirements, PUs, GPU and OP.
Minimum system requirements for SDL Machine Translation Edge and SDL Machine Translation Edge Trainer
The minimum system requirements to install and run SDL Machine Translation Edge Trainer are the same as for Generic Language Pairs, but a GPU is mandatory for the customization of Adaptable Language Pairs.
Minimum hardware requirements for SDL Machine Translation Edge application:
- 1 CPU core
- 1 GB of RAM
- 4 GB HD
Minimum hardware requirements to train an Adaptable Language Pairs:
- 1 CPU core
- 2 GB of RAM
- 30 GB of HD
- 12 GB of GPU RAM
Minimum hardware requirements to run an Adaptable Language Pair (in CPU optimized for Quality mode):
- 4 CPU cores
- 2 GB of RAM
- 4 GB of HD
The requirements mentioned above are additive, based on the number of language pairs that need to run in parallel.
Processing Units for Adaptable Language Pairs
Similarly to Generic Language Pairs, 1 Processing Unit (PU) is required to start 1 Translation Engine which supports 1 Adaptable Language Pair which translates with a throughput of 2000 wpm.
To run multiple different Adaptable Language Pairs in parallel, you need as many PUs as Adaptable Language Pairs.
For example, if you create 2 customized Translation Engines for the English → French Adaptable Language Pair and want to run them in parallel, you will need 2 PUs (1 for each Translation Engine).
GPU for Adaptable Language Pairs
1 GPU is required to perform the training of 1 Adaptable Language Pair.
Once the Adaptable Language Pair has been created with SDL Machine Translation Edge Trainer, it can be operated in the following hardware optimization modes: GPU or CPU Optimized for Quality.
During the training, 1 GPU is used exclusively for the training of the Adaptable Language Pair. It is not recommended to have any started Translation Engines using the GPU on the host where the training takes place.
GPUs must have at least 12 GB of GPU RAM.
SDL Machine Translation Edge supports the following GPUs:
- Nvidia Tesla P100
- Nvidia Tesla V100
- Nvidia Tesla T4
OS for Adaptable Language Pairs
The first SDL Machine Translation Edge Trainer is supported only on Linux.