3 ms·
> I think a better comparison would have been some of the smaller discrete GPU wielding laptop. I agree, but from what I have seen they also perform poorly aga
by teruakohatu 6y ago
> I think a better comparison would have been some of the smaller discrete GPU wielding laptop.
I agree, but from what I have seen they also perform poorly against collab. A 1660 Ti 4gb RAM laptop gpu, runs about as fast as Collab CPU does. [1]
I have not been able to find how the RTX 2080 Super (Laptop) and upcoming RTX 3080 (Laptop) compares to Collab. The 3080 has over 3x as many cores as the 1660 Ti.
[1] https://towardsdatascience.com/google-colab-how-does-it-compare-to-a-gpu-enabled-laptop-851c1e0a2ca9\ https://towardsdatascience.com/google-colab-how-does-it-comp...
- joshocar 6y agoAre the Quadro cards any good for it?
- teruakohatu 6y agoThey are, other than being priced more expensively compared with consumer cards. When doing things like RNN's, RAM becomes an issue regardless of how powerful the card is in other ways.
- apohn 6y ago>A 1660 Ti 4gb RAM laptop gpu, runs about as fast as Collab CPU does According to the chart on that page you linked to the Lenovo Legion (with the 1660ti) is about 30% faster than the Colab GPU. The Lenovo 480s (CPU only, no GPU) is about the same speed as the Colab CPU. So if you can train your model on the on 4GB of VRAM, a laptop with a discrete GPU (if you exclude the really basic business laptop GPUs) may be useful.