原文作者:@rargulati,MartinShkreli
Original text: twitter
Compilation: Vernacular blockchain
@ionet is a decentralized computing network built on Solana, belonging to the Depin and AI sectors, and has received funding from Mult1C0in Capital and Moonhill Capital for an undisclosed amount.
io.net is a Solana-based decentralized cloud platform for machine learning training on GPUs, providing instant, permissionless access to a global network of GPUs and CPUs. With 25,000 nodes and revolutionary technology to cluster GPU clouds together, the platform saves up to 90% on compute costs for large-scale AI startups.
It is currently built on Solana, which belongs to the current hot Depin and AI sectors, and today let's take a look at the two on X to analyze their GPUs and existing problems:
@ionet拥有多少个GPU（Graphics Processing Unit，图形处理器)是用于图形处理的芯片）？
X @MartinShkreli analyzed the four answers:
1) 7648 (when attempted on deployment)
2) 11107 (manually calculated from their explorer)
3) 69415 (unexplained number, unchanged?)
4) 564306 (There is no support, transparency, or substantive information here.) Not even CoreWeave or AWS so much)
Think the real answer is actually 320.
Why 320?
Take a look at the explorer page with me. All GPUs are "free", but you still can't rent one. If they're free, why can't they be rented? People want to get paid, right?
There are only 320 of them that you can actually rent.
If you can't rent them, then they don't really exist. Even if you can rent, it will increase...
@rargulati said Martin was perfectly right to question the matter. Decentralized AI protocols have the following problems:
1) There is no cost-effective and time-efficient way to perform useful online training on a highly distributed, general-purpose hardware architecture. This requires a major breakthrough that I am not currently aware of. That's why FANG spends more money than all the liquidity of cryptocurrencies for expensive hardware, network connections, data center maintenance, and more.
2) Inferring on general-purpose hardware sounds like a good use case, but the hardware and software side is evolving so rapidly that the general-purpose decentralized approach doesn't perform well on most key use cases. You can refer to the latest OpenAI latency and the growth of Groq.
3) Extrapolate from properly routed requests, GPU clusters that coexist tightly with requests, and leverage decentralized cryptocurrencies to drive down the cost of capital to compete with AWS and incentivize enthusiast participation. Sounds like a good idea, but with so many vendors and fragmented liquidity in the GPU spot market, no one has put together enough supply to provide to the people running the real business.
4) The software routing algorithm must be very good, otherwise there are many problems in the operation of the general hardware of the consumer operator. Forget about network breaches and congestion controls, if someone decides to play a game or use anything that uses WebGL, you may experience an outage from one of the carriers. Unpredictable supply can cause confusion for operations and uncertainty for demand-side requesters.
These are tricky questions that take a long, long time to solve. All bids are just memes.