Startup demos upcoming decentralized GPU infrastructure network to OpenAI, Uber

What began as an institutional-grade cryptocurrency and stock quantitative trading system has transformed into a decentralized network procuring GPU computing power to meet the growing demand for artificial intelligence (AI) and machine learning (ML) services.

io.net has developed a test network that sources GPU computing power from various data centers, cryptocurrency miners, and decentralized storage providers. Aggregating GPU computing power is believed to significantly reduce the cost of renting these resources, which has become increasingly expensive as artificial intelligence and machine learning advance.

In an exclusive interview with Cointelegraph, CEO and co-founder Ahmad Shadid revealed details of the network, which aims to provide a decentralized platform for renting computing power while At a fraction of the cost of existing centralized alternatives.

related: The future of payments: Visa to invest $100 million in artificial intelligence

Shadid explained the idea for the project during the Solana Hackathon in late 2022. Io.net was developing a quantitative trading platform that relied on GPU computing power for high-frequency operations, but was in trouble due to the high cost of renting GPU computing power.

The io.net platform will allow GPU computing providers to provide resources to clusters to meet the needs of artificial intelligence and machine learning. Source: io.net

The team solved the problem of renting high-performance GPU hardware cores document, the price to rent a single Nvidia A100 averages about $80 per card per day. If it takes more than 50 cards per month to run 25 days, the cost will be over $100,000.

The solution was found in the discovery of Ray.io, the open source library used by OpenAI to distribute ChatGPT training on over 300,000 CPUs and GPUs. The library streamlined the project’s infrastructure, with its backend developed in just two months.

Shadid Demonstrative Io.net’s working testnet at the AI-focused Ray Summit in September 2023, highlighting how the project aggregates computing power and makes it available as a cluster to GPU consumers to meet specific AI needs or machine learning use cases.

“This model not only allows Io.net to provide GPU computing that is 90% cheaper than existing vendors, but also allows for virtually unlimited computing power.”

The decentralized network will leverage Solana’s blockchain to provide SOL (SOL) and USD Coin (USDC) payments to machine learning engineers and miners who rent or provide computing power.

“When ML engineers pay for the cluster, those funds go directly to the miners using GPU services in the cluster, while a small network fee is allocated to the IO.net protocol.”

The project’s roadmap will include the launch of a dual native token system with IO and IOSD capabilities. The token model will reward miners for executing machine learning workloads and maintaining network uptime, while taking into account the dollar cost of electricity consumption.

“The IO coin will be freely traded on the crypto market and will be a gateway to access computing power, while the IOSD token will serve as a stable credit token, algorithmically pegged to $1.”

Shadid told Cointelegraph that Io.net is fundamentally different from centralized cloud services such as Amazon Web Services (AWS):

“For example, they are United Airlines and we are Kayak; they own the planes and we help people book flights.”

The founder added that any business that needs AI computing typically uses third-party vendors because they lack the GPUs to handle it all in-house. Hadid said that GPU demand is expected to grow 10 times every 18 months, but capacity is often insufficient to meet demand, resulting in long wait times and high prices.

The situation is compounded by what he describes as poor utilization of data centers that are not optimized for the rapidly growing types of artificial intelligence and machine learning work:

“There are thousands of independent data centers in the United States alone, with average utilization rates of 12%–18%. As a result, bottlenecks continue to emerge, which has a ripple effect and drives up the price of GPU computing.”

The upside is that ordinary cryptocurrency miners can make a profit by renting out their hardware to compete with companies like AWS. Hadid said miners using the 40GB A100 earned an average of $0.52 per day, while AWS sells the same card for AI computing for $59.78 per day.

“Part of the value proposition of Io.net is, first of all, we allow participants to access the AI ​​computing market and resell their GPUs at a much cheaper price than AWS for ML engineers.”

Data shared by Cointelegraph estimates that miners with GPU resources can earn up to 1,500% more than mining various cryptocurrencies.

Magazine: Blockchain Detective: Mt. Gox’s Collapse Witnesses the Birth of Chainaanalysis