Full Development of Kaggle Deep Learning Project
Budget: $10 – $30 USD
This assignment is designed to make yourself familiar with the Kaggle environment, as well as the available computational devices on Kaggle including CPU, single GPU (P100), two GPUs (2*T4), and cloud TPU. We'll benchmark the performance of these computational devices.
The benchmarking code can be created using AI tools. For example, I have the following ChatGPT conversation to create PyTorch code to benchmark the performance of my Apple Laptop (M3 chip), which has CPU and GPU (called "MPS" in PyTorch) devices:
https://chatgpt.com/share/30fec905-5837-43c2-b37b-39208b222016
As you see, in my benchmarking code I used matrix multiplication to test the performance of my GPU. I also tried multiplying matrics of various sizes such as 10k*10k, 20k*20k and 30k*30k (not shown in the ChatGPT conversation).
(Links to an external sA few tips
Deliverables:
A report in PDF format, which includes links to your notebooks on Kaggle.
Rubric:
1 point: your code shows CPU performance.
1 point: your code shows single GPU (P100) performance.
1 point: your code shows cloud TPU performance. Note that the TPU is very popular on Kaggle, and you might have to wait in line for some time to be allocated one TPU instance.
2 points total: your code shows the performance of two GPUs (T4). Details marking scheme as follows:
1 point if you are only use the two GPUs sequentially, that is, you first performed some computations on one GPU, and then you performance some computations on the other GPU.
2 points if you manage to use the two GPUs simultaneously, that is, your code makes both GPUs busy at the same time.
The benchmarking code can be created using AI tools. For example, I have the following ChatGPT conversation to create PyTorch code to benchmark the performance of my Apple Laptop (M3 chip), which has CPU and GPU (called "MPS" in PyTorch) devices:
https://chatgpt.com/share/30fec905-5837-43c2-b37b-39208b222016
As you see, in my benchmarking code I used matrix multiplication to test the performance of my GPU. I also tried multiplying matrics of various sizes such as 10k*10k, 20k*20k and 30k*30k (not shown in the ChatGPT conversation).
(Links to an external sA few tips
Deliverables:
A report in PDF format, which includes links to your notebooks on Kaggle.
Rubric:
1 point: your code shows CPU performance.
1 point: your code shows single GPU (P100) performance.
1 point: your code shows cloud TPU performance. Note that the TPU is very popular on Kaggle, and you might have to wait in line for some time to be allocated one TPU instance.
2 points total: your code shows the performance of two GPUs (T4). Details marking scheme as follows:
1 point if you are only use the two GPUs sequentially, that is, you first performed some computations on one GPU, and then you performance some computations on the other GPU.
2 points if you manage to use the two GPUs simultaneously, that is, your code makes both GPUs busy at the same time.