What level GPU can work with real time?

[EricKong1985]

On enterprise GPUs and PyTorch nightly, Bark can generate audio in roughly real-time. What level GPU can work with real time? I have use 3060Ti, It need no little seconds to generate audio files！！！

[aleksandermajos]

Exacly, I have RTX 4090 and 14 sec clip My GPU need like 30 sec...

[EricKong1985]

Oh... It looks two slow...

[tonyabracadabra]

Marking this thread for future improvements

[lc-spxl]

GeForce RTX 3060 12GB, 15sec to generate 2sec audio :cry:

[d4rkc0de]

Same, is there a specific GPU where I can run this faster ? Anything that can generate a 10-second clip in like 10 seconds ?

[Goddard]

If you could change it to a streaming audio buffer you could at least get the start of the audio faster.

[d4rkc0de]

If you could change it to a streaming audio buffer you could at least get the start of the audio faster.

Thanks, I agree 100%, but is there any idea how to do that ?

[snehccurry]

i have mx 130. it has taken me hours honestly.

[AI-General]

@Goddard Hello, do you have streaming version of Bark?

[mswoff]

I also haven't been able to get streaming or realtime playback working.