Speed Results for Yolov5x not consistent in Colab with official table

[rickymedrano]

❔Question

In the main readme, it shows the speed of yolov5x on a v100 is 6ms. I'm using the official Colab notebook and ran the #Reproduce section. The GPU of my instance is a T4 which has a compute capability of 7.5 while the V100 has 7.0
I'm getting 20ms for speed (you can tell it's yolov5x from the 218GLOPS).

I also tested the same call on my RTX 2070 and get 12.9ms

Both these cards have more compute capability than the V100 yet are running slower speed than the claimed 6.0ms in the table.
I don't have Colab Pro so can't specifcally check out a V100 to test this but wanted to see if others are experiencing slower than 6.0ms speed on cards with better compute capability than the V100?

output of git log -1:
commit 404749a (HEAD -> master)
...

I also tested yolov5s and get fairly close results to the speed table:
Colab w/ T4: 4.5ms
RTX 2070: 2.5ms

[github-actions]

👋 Hello @rickymedrano, thank you for your interest in 🚀 YOLOv5! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.

For business inquiries or professional support requests please visit https://www.ultralytics.com or email Glenn Jocher at [email protected].

Requirements

Python 3.8 or later with all requirements.txt dependencies installed, including torch>=1.7. To install run:

$ pip install -r requirements.txt

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training (train.py), testing (test.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@rickymedrano hey buddy. I think you have a misunderstanding of relative GPU speeds, a V100 is a much faster GPU than a T4, and is also slightly faster than the best 3090 consumer GPU. The benchmark speeds logically require similar hardware to reproduce, i.e. an n1-standard-8 instance on GCP or a p3 instance on AWS, which you can get started with via the environment quickstart guides below.

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Google Colab and Kaggle notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide