Suggested Hardware/GPU Memory for training

[flexwend]

Can anybody suggest hardware for training? I got a dataset of about 1000 files, and would like to train with batch size 32.
I tried it with aws instance g4dn.xlarge (1 GPU, 4 vCPUs, 16 GiB of memory, 125 NVMe SSD, up to 25 Gbps network performance) and it seems like that its not enough memory and therefore reduces the performance significantly.

I would appriciate any tips

[marutichintan]

you can try this. i have tried performance is good.

https://e2enetworks.com
NVIDIA - RTX8000 | 48 GB | Ubuntu - 20.04 | GDC.RTX-16.115GB | CPU 16 | 115 GB | 900 GB

[flexwend]

@marutichintan Thanks for the answer and tip

Can you say how large your dataset was and how long training took for a field?

[test2a]

@flexwend did you use something? i just found this project and i am basically swamped. i feel overwhelmed. did you get to use this? what server did you use and what is the datasize-to-time ratio with regards to listed configs here?

don't we have to generate a training data just once?