Same response time in 1 core and 8 core aws gpu

[JRMeyer]

>>> halder.nayan35
[March 10, 2020, 8:14am]

Hi all, slash
I am running mozilla TTS on 8 core AWS GPU and the response time is same
as 1 core.

i first run the 1) export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 slash
2) source tts-venv/bin/activate slash
3) python -m TTS.server.server

Am i missing any configuration setting.

[This is an archived TTS discussion thread from discourse.mozilla.org/t/same-response-time-in-1-core-and-8-core-aws-gpu]

[JRMeyer]

>>> dkreutz
[March 10, 2020, 9:19pm]

No, to my understanding inference/synthesizer can not use multiple GPUs
as there is no batch processing possible like when training a model.

[Archived Post]

[JRMeyer]

>>> nmstoker
[March 10, 2020, 11:01pm]

Yes, using one GPU is consistent with Eren's response here:
mozilla/TTS#363

[Archived Post]

[JRMeyer]

>>> halder.nayan35
[March 12, 2020, 5:03pm]

Thanks for the reply. I understood 8 core gpu will not improve response
time. Is there a way out to improve TTS response time within 1 core AWS
gpu. My current TTS response time 4-5 sec. I want to bring down to this
below 1-2 sec. Is it doable in 1 core gpu by some configuration changes
or settings.

[Archived Post]

[JRMeyer]

>>> erogol
[March 12, 2020, 11:37pm]

number of GPUs does not do any good for a single utterance response
time. You can only increase the number of utterances processed in
parallel.

[Archived Post]

[chikiuso]

Hi @JRMeyer , may I ask how to increase the number of utterances processed ? thanks.