Is it possible to do machine learning on an integrated Intel Iris 5200 graphics card, or is it better to compute in the clouds?

Question

data mining

Hello!
Began to be interested in machine learning, help me understand if Intel Iris Pro graphics 5200 is enough for learning ML and computing?
I plan to make a bias in E-commerce, it is also possible to work with medical data.
If this map is not enough, then how expensive will it be to process data in the clouds. I looked at Amazon, but I didn’t immediately figure out how much it costs to process a certain amount of information and what is needed for this.
I looked at external vidyuhi, but they are not cheap and you can’t take one like that anywhere.
What fork of the amount of information do you have to work out and how much does it cost in the clouds?
What would you advise to approach this matter?
Thanks for the help!

Answer 1

Start working with the CPU, then get a normal GPU if you need to. If you don’t get to deep learning, then you don’t need a GPU.

Answer 2

Look at the software from INTEL for these purposes, because. there is no software for Nvidia cards.
Decide on the tasks, perhaps the Iris 5200 will calculate in a reasonable time.

Answer 3

EMNIP, almost all frameworks use either CPU or CUDA with OpenCL. It turns out that Intel is still in flight in terms of video cards.
Z.Y. And yes - what tasks?

Answer 4

You don’t need to learn a video card at all, but if you dig into the same convolutional neurons, retrain them, then you need to take at least an Nvidia 1050Ti, and preferably 1060 6Gb.
Alternatively, you can rent servers from Selectel.

Answer 5

You can even on much older and much weaker equipment.
The only question is expediency and efficiency.
And in the expediency and effectiveness of alternatives.
What are your limits on efficiency, what are your limits on spending money, etc.
And the questions of expediency and effectiveness for your case - only you can decide.
What do you have to say?
In grams per kilometer, what will the weather be like on the moon?
The question sounds like this: I need to process some data. How much exactly will it cost.
Answer:
Try it and find out.
Compare in the clouds and on Iris your specific case. If it turns out to be too expensive, you can easily interrupt the operation (and payment) of the cloud at any time. And you will know exactly
the cost for your particular case .

1. I would start with what is already there.
   
2. If it wasn't enough, I would try clouds.
   
3. Then it would have become clear - is it worth buying your own hardware for it.