Can a Nvidia 4070 ti with 12gb run medium model?

[jeveloper]

Apologies for this silly question but as someone fairly new to ML. Would this just fail due to allocation of memory.

Can resizable bar help with the lack of memory on this GPU?

Thank you kindly.

[donaldanixon]

It's running on my 1070 with 8gb VRAM, so definitely

[jeveloper]

@ResidentChief Oh thank you, do you know why specifying a bigger model requires 16gb or more. Is there particular bottleneck that does occur? in README it seems to be pretty clear about memory on GPU.

Thanks

[donaldanixon]

I think they are referring to the large model which has over twice the parameters and thus twice the vram

On Tue, 20 Jun 2023 at 13:39, Serge Bornow @.***> wrote:

@ResidentChief https://github.com/ResidentChief Oh thank you, do you know why specifying a bigger model requires 16gb or more. Is there particular bottleneck that does occur? in README it seems to be pretty clear about memory on GPU.

Thanks

— Reply to this email directly, view it on GitHub https://github.com/facebookresearch/audiocraft/issues/117#issuecomment-1597998889, or unsubscribe https://github.com/notifications/unsubscribe-auth/AIRC5P6FSAOLLT27AW24B33XMD5LVANCNFSM6AAAAAAZMOYRO4 . You are receiving this because you were mentioned.Message ID: @.***>

[olamedia]

I was able to run large on 3060, may be because of the drivers.