I hit the common RuntimeError: CUDA out of memory even if I put batchsize=2 in the IMDB text classification task (01_Intro.ipynb). So, I concluded that a NVIDIA RTX 2060 will cause to much struggles since it has only 6 GB of VRAM (and I don’t think that gc.collect could solve the problem since I tried to run only a particular cell).
I ran into similar problems with the free instance of Paperspace (it works at batchsize=16, but it froze after about 2 long hours of training).
So, I would like to buy a new laptop/PC to avoid struggles. Therefore, what GPU should I buy instead? I found interesting benchmarks here! Also, I still don’t know if I should go for a laptop/PC and for Linux/Windows. What should I choose?
not the answer you wer hoping for… but why don’t you try to setup a server on AWS or the other vendors? You could probably start today to continue your NLP quest, you’ll have the possibility to setup servers with the newest GPUs, and you’ll probably save some money.
@jjohnson I currently have the Lenovo X1 Extreme Gen 3. I love this computer. Truly a workstation. This is the most recent in a line of excellent Lenovo machines: t61p, t500, w510, t440p.
The most recent Lenovo X1 Extreme Gen 4 supports as an option the RTX 3080 mobile card, with 16gb VRAM. I may very well opt to upgrade around Black Friday, which typically has lowest prices AFAIK.
At price point, build quality (ie unlike others who do not properly ventilate the machine so eventually components fail), and warranty ease/length/price, I would say this is an ideal laptop going forward.
In addition, it has the little red mouse which is great for programming on laptops when you’re on the go.
For Linux, many use Archlinux on these machines. I attempted a multi boot scenario on one HD, but it would be simpler to add a separate SSD for your secondary OS. Ubuntu is very simple to install on these computers typically and Lenovos tend to be well documented for Ubuntu.