GPUs

English Version pytorch
1

https://d2l.ai/chapter_builders-guide/use-gpu.html

2

I didn’t use nvidia.So I didn’t install GPU version and cuda.
torch.cuda.device('cuda')
AssertionError : Torch not compiled with CUDA enabled
torch.cuda.device_count()
0

Why didn’t AssertionError happen when I run the following code?
torch.cuda.device('cuda:1')
<torch.cuda.device at 0x16349fb2488>

3

5.6.5. Exercises

  1. you should see almost linear scaling? I’m confused what is linear scaling.
4

‘Linear scaling’ is that the computation speed is proportional to the number of GPUs you use.
e.g., With one GPU, two tasks take 2 sec, and with two GPUs, two tasks (each on one GPU) takes only 1 sec.

1 Like
5

awsblog – 玩转GPU实例之系统工具 – NVIDIA 篇

7

I am confused about most of the questions here

Exercises and my silly answers

  1. Try a larger computation task, such as the multiplication of large matrices, and see the difference in speed between the CPU and GPU. What about a task with a small amount of calculations?
  1. How should we read and write model parameters on the GPU?
  • by using net.to(device=torch.device('cuda'))
  1. Measure the time it takes to compute 1000 matrix-matrix multiplications of 100 × 100 matrices and log the Frobenius norm of the output matrix one result at a time vs. keeping a log on

the GPU and transferring only the final result.

  • I did not get it but the difference between time in cpu and gpu is 40ms.
  1. Measure how much time it takes to perform two matrix-matrix multiplications on two GPUs

at the same time vs. in sequence on one GPU. Hint: you should see almost linear scaling.

  • I have one GPu