A task here is transcription (extraction of words in several languages) of several terabytes (around 400 TB) archived speaches.
What (Windows) PC configuration would you recommend for SL up to 5.000,- €?
I wouldn’t do such tasks with SpectraLayers. I think it’s more reasonable to use something like OpenAI Whisper and write a script to automate that and maybe run that on a cloud service like Runpod.
1 Like
It has to run locally. So we need a good PC for this.
Which graphics card is recommendable for SL?
As mentioned, the price tag for the whole pc is around 5.000,-€
To get 400 TB into a cloud with 1 GB connection takes far too long, besides other tasks that run on the line. The storage is also not so cheap. Ok, this could be done in chunks, sure. So not that much space is needed. But it has to be local.
We do have the data already lokally on a NAS with 10 GB connection locally.
You generally want as much VRAM as possible. So, aim at least for 24 GB of VRAM. This means you should get a RTX 3090, 4090 or 5090. That will be definitely the most expensive part of your PC built.
Whisper runs also fine with more than 10 GB of VRAM locally.
Running something locally vs. in the cloud should be considered if running it locally will just take too much time. There is a reason why server GPUs cost $30k and upwards.
1 Like
Thx.
Well, yes, there are reasons to run it locally, also.
With that amount of data, I wouldn´t even consider any gaming GPU´s..
Of course these ones costs a lot…
Thx .