Home
Solution
Galvatron
Free, open and open-source
distributed AI training network
Issues & Solutions
issue

Insufficient chip manufacturing capacity[More]

  • The computility required for daily usage of just ChatGPT-4 alone would require 60,000 NVIDIA A100 GPUs, with each unit costing over $10,000. Currently, the demand for such resources continues to rise.
  • Even with computility rental, the costs remain prohibitively high. Many students, enthusiasts, researchers, and academic institutions lack the financial capacity to lease specialized equipment.
  • Indeed, hardware maintenance and upgrade costs can be quite high. In addition to the initial investment in hardware, there are ongoing expenses for maintenance, repairs, and potential upgrades to keep up with the evolving technology.
solution

AIForce is dedicated to creating a free, open, and open-source distributed AI training network, Galvatron. Anyone with AI training needs can complete their training tasks on the Galvatron network by paying an ultra-low cost.

issue

A large amount of idle computility[More]

  • By 2023, the global number of servers is estimated to exceed 60 million units. Based on data estimates, the idle rate is roughly around 30%. This would roughly estimate the global idle computility at around 130 EFLOPS (The average computility of servers is assumed to be equivalent to a Radeon HD 5970, with an idle rate of 25%).
  • Additionally, there are billions of personal computers and mobile devices worldwide. Assuming these devices have an average computility of only 10% of a GeForce GTX 660 and 50% idle rate, the idle computility of these devices may even surpass that of all running servers.
solution

Individuals and organizations with idle computational resources can join the Galvatron network at any time and from anywhere to earn additional income. Even individuals without any AI background can participate in the AI industry through the Galvatron network.

issue

The AI industry has a high barrier to get involved[More]

  • For investors: Investment channels are limited, mainly relying on purchasing graphics cards and stocks. Early-stage AI projects require substantial capital support.
  • For developers: There is a lack of development tools available, limited availability of data samples, and a significant time investment in data processing. Fine-tuning model frameworks requires substantial effort and experience.
solution

AIForce will also develop a set of comprehensive developer tools, including clients, official nodes, and SDKs, built on top of the Galvatron network. Additionally, AIForce plans to launch a Launchpad platform to facilitate fundraising for developers' AI products, providing investors with more choices and opportunities.

What is Galvatron

Galvatron is a P2P network of heterogeneous computility which can accept training tasks and working under unstable network environment.

By enhancing algorithms such as distributed deep learning algorithms, Galvatron enables the communication of network nodes distributed around the world.

Galvatron also designed a variety of mechanisms to ensure the completion of distributed training tasks in the environment that node frequently enters and exits the network.