3 ms·

I am just curious (and a total machine learning novice). If you were to experiment with ML, what are the benefits to getting a fast consumer card like this or u

by kayoone 10y ago

I am just curious (and a total machine learning novice). If you were to experiment with ML, what are the benefits to getting a fast consumer card like this or use something like AWS GPU instances (or some other cloud provider) ? Or phrased differently: When does it make sense to move from a local GPU to the Cloud and vice versa ?

marmaduke 10y ago
AWS would make sense if you don't yet know if your ML algo scales well on a GPU: you can try the hardware without losing much money. Consumer cards are a good deal if you know your algo works well on GPUs in single (32-bit) floating point and you don't mind dicking around with hardware configurations to save money.
nightski 10y ago
I'm a relative novice also, but we have run an EC2 8 way GPU instance. Basically the GPUs are a bit slower and they only have 4 GB of ram. Also getting everything set up and uploading a data set is tedious. Costs weren't too bad. I could see AWS being used for training smaller problems and learning. If you have a serious project it probably won't fit the bill.
frozenport 10y ago
Image analysis is cumbersome on the cloud, copying data back and forth gets tedious really quick. I would use the cloud only when I was confident as what I was doing.
mtw 10y ago
The AWS Linux GPU instances only have 4GB of memory which means processing can be slow (memory swaps) if your datasets is large. Generally, it makes sense to have a local GPU to try out models, and move to the cloud for large scale computing, once you have a better idea of the model.
verbify 10y ago
It's about how much machine learning you're currently doing and how much you expect to spend. E.g. if you play around for a little bit, you could end up spending $50 on the cloud. By contrast, a card will set you back $350+. If you're doing lots of machine learning, the cloud will quickly rack up. There's other factors. E.g. the cloud's performance can vary depending on the time of day (and whose using the application). Don't buy a graphics card unless you're certain you're going to do lots of ML. Or buy one if you think that'll motivate you to do justify the purchase by doing lots of ML.

by kayoone 10y ago

marmaduke 10y ago

AWS would make sense if you don't yet know if your ML algo scales well on a GPU: you can try the hardware without losing much money. Consumer cards are a good deal if you know your algo works well on GPUs in single (32-bit) floating point and you don't mind dicking around with hardware configurations to save money.

nightski 10y ago

I'm a relative novice also, but we have run an EC2 8 way GPU instance. Basically the GPUs are a bit slower and they only have 4 GB of ram. Also getting everything set up and uploading a data set is tedious. Costs weren't too bad. I could see AWS being used for training smaller problems and learning. If you have a serious project it probably won't fit the bill.

frozenport 10y ago

Image analysis is cumbersome on the cloud, copying data back and forth gets tedious really quick. I would use the cloud only when I was confident as what I was doing.

mtw 10y ago

The AWS Linux GPU instances only have 4GB of memory which means processing can be slow (memory swaps) if your datasets is large. Generally, it makes sense to have a local GPU to try out models, and move to the cloud for large scale computing, once you have a better idea of the model.

verbify 10y ago

It's about how much machine learning you're currently doing and how much you expect to spend. E.g. if you play around for a little bit, you could end up spending $50 on the cloud. By contrast, a card will set you back $350+. If you're doing lots of machine learning, the cloud will quickly rack up. There's other factors. E.g. the cloud's performance can vary depending on the time of day (and whose using the application). Don't buy a graphics card unless you're certain you're going to do lots of ML. Or buy one if you think that'll motivate you to do justify the purchase by doing lots of ML.