Still using GPU mining? You are losing a lot.

Recently, a machine learning shared computing power platform called "Vectordash" appeared on the other side of the ocean. In the words of the developers themselves, it is equivalent to Airbnb for GPUs.

Do you have a GPU but think mining is too unprofitable? You can share your computing resources on this platform and help others with AI.

The income is 3 to 4 times that of mining!

AI developers rent shared GPUs on the platform at a price that is only 1/5 of AWS.

Poor student

This welfare-like platform is a side project of poor student Sharif Shameem and his two friends.

Shameem posted on Reddit that they were inspired by their own experiences to build this platform.

As a poor student studying artificial intelligence and deep learning, Shameem always has to use GPU to train neural networks, but AWS and Google Cloud are so expensive, and the student discount is not enough.

"Teacher, can I hand in this semester's final assignment later? I really don't have the money to rent a GPU, and the model is still being trained slowly using the CPU..."

This reason for submitting homework late seems really shabby, but poor students have their own ways.

He found a friend who used a 1080 Ti to mine Ethereum, paid him twice the price of the mining income, and happily trained his neural network using the miner's computer. It was still much cheaper than AWS.

But sharing happiness is worse than enjoying it alone. Such a good model is worth uniting miners all over the world, and also uniting people without GPUs all over the world.

So Vectordash was born.

Come, remember the names of these three poor students, just in case they become the next Bill Gates. They are: Sharif Shameem, Arbaz Khatib from the University of Maryland, and Abhishek Bhargava from Carnegie Mellon University.

The owner gets more money for the machine, and the buyer gets less money

The official website provides a clear price list for various GPUs: the old GPU 1050, which earns the least, is $1.92 per day; the Titan V earns the most, at $10.08 per day.

GPU providers earn more than mining, but the GPU instances that developers rent are not expensive, starting at $0.17 per hour, which is only 1/5 of AWS. Renting the latest Titan V costs only $0.85 per hour.

Each GPU instance is also pre-installed with Ubuntu 16.04, CUDA, cuDNN, TensorFlow, PyTorch and other essential machine learning environments.

Comparing the two lists, the owner of the machine gets more money and the buyer spends less. If you calculate the price difference, it seems that the platform also makes a lot of money. Everyone is truly happy.

How to use it?

For developers, using Vectordash is similar to using other cloud services: you rent a GPU instance and get an IP address and SSH key for connection.

Let’s take a look at this video demonstration:

Unfortunately, Quantum位 was not able to try it out for everyone in person because all the machines are currently rented out.

△ Sorry, all machines are currently in use. Are you interested in becoming a machine owner?

I didn’t rent a GPU, but I was asked a heart-touching question by the Vectordash website: Are you interested in sharing your GPU and making some money as a machine owner?

If QuantumBit had a 1080 Ti, how could it miss this opportunity to become a cloud service provider in a minute, benefit AI researchers, and make $7.68 a day? I haven't bought the GPU yet, so it's good to help everyone explore the way first.

It is not difficult to become a machine owner. You need to fill in a form first:

△ https://goo.gl/forms/ghFqpayk0fuaXqL92

Then, you will receive a guide email from Vectordash within a day, instructing the owner to install the hypervisor and share the GPU.

The owner can choose the time period for providing GPU in the management program. In addition, this management program also supports automatic switching function. When no one occupies the GPU on Vectordash, it can be used for mining to maximize profits.

In order to receive payment, the owner of the machine also needs to bind a virtual currency wallet. After the user uses the GPU, the owner will receive the money within 5 seconds. You can choose Bitcoin, Ethereum or Litecoin, and you can withdraw it at any time.

New opportunities

Oh my god, this model makes everyone excited.

"Drop out of school, you'll be the next Bill Gates."

On the Reddit forum, more than one person came up with this idea. They all urged Sharif Shameem and his partners to stop doing research and focus on this project to become Amazon's rival.

Currently, the GPU cloud computing services provided by Amazon use dedicated products such as K80, while cheaper consumer-grade GPUs cannot be used in commercial scenarios such as cloud computing.

However, this sharing model seems to bypass Nvidia's restrictions.

The founders of Vectordash said that they trained neural networks on a 1080Ti five times faster than using a Tesla K80 on the Amazon cloud.

People have expressed their willingness to rent out their GPUs. Shameem said on Twitter that within 17 hours of the beta version being online, 143 GPUs were received.

Some people are unwilling to ask: Are AMD graphics cards okay?

Of course not.

Vectordash currently does not support AMD graphics cards, because OpenCL has not yet been widely supported by machine learning libraries. Of course, Vectordash partners also hope that AMD, Intel, and Google can catch up as soon as possible. They don't want to live in a world dominated by Nvidia alone.

This business model inevitably reminds people of Golem.

According to the introduction, Golem is a decentralized global computing power market. Golem changes the way computing power tasks are organized and executed. Below is a video introduction without subtitles.

Hidden worries

"Decentralized machine learning is really a good idea," said Tianqi Chen (author of XGBoost, MXNet, and TVM), "but there may be security risks."

+1, Keras author François Chollet seconded and further pointed out that using random people's computers sounds like a privacy and security disaster. This is especially true for projects where the training data is sensitive, so a more secure way is needed.

On the Reddit forum, a netizen asked: I am interested in providing GPUs, but how can I ensure that my personal information will not be leaked? How can I ensure that I will not become part of a botnet?

The founder of Vectordash replied that they used a Hypervisor based on LXC (the core of virtualization technology, which allows multiple operating systems and applications to share a set of basic physical hardware).

But just these, it seems that people are not completely reassured.

“I have a perfect solution!”

Now David Ha from the Google Brain team said that he plans to use blockchain to solve trust and security issues, and plans to issue TensorFlowCoin, and then ICO...

Being the first to issue coins, he is worthy of being a big shot in the field of investment banking and AI.

Of course he was joking.

Anyway, everything is just beginning. Maybe one day in the future, it will be possible to realize in the field of artificial intelligence: just one click and training will start immediately.

"Are you at home? Come online and play PUBG!"

"How about pesticides?"

"What's wrong?"

"My GPU is working and earning money, I don't want to disturb it."