The resurrection of cloud computing power—in the name of AI
Fashion is a cycle, and so is Web 3.
Near "re" became an AI public chain, and its founder Transformer's status as one of the authors allowed him to attend the NVIDIA GTC conference and talk about generative AI with the leather-clad Lao Huang. In the future, Solana will successfully transform into an AI concept chain as a gathering place for io.net, Bittensor, and Render Network. In addition, there will be emerging players involved in GPU computing such as Akash, GAIMIN, and Gensyn.
If we raise our sights, while the currency price is rising, we can find several interesting facts:
The battle for GPU computing power comes to the decentralized platform. The more computing power, the stronger the computing effect. CPU, Storage and GPU are paired with each other;
In the transition of computing paradigm from cloud to decentralization, Behind the scenes is the shift in demand from AI training to reasoning, and on-chain models are no longer empty talk;
< strong>The underlying software and hardware composition and operating logic of the Internet architecture have not fundamentally changed, and the decentralized computing power layer plays more of a role in stimulating networking.
Let’s first make a conceptual distinction. The cloud computing power in the Web3 world was born in the era of cloud mining, which refers to the mining industry. The computing power of the machine is sold in packages, eliminating the huge expenditure for users to purchase mining machines. However, computing power manufacturers often "oversell", for example, mixing the computing power of 100 mining machines and selling them to 105 people in order to obtain excess profits, which ultimately makes the company Words are tantamount to deception.
The cloud computing power in this article specifically refers to the computing power resources of GPU-based cloud vendors. The question here is whether the decentralized computing power platform belongs to the cloud vendors. The front-end puppet is still the next version update.
The integration between traditional cloud vendors and blockchain is deeper than we imagined. For example, public chain nodes, development and daily storage basically revolve around AWS, Alibaba Cloud and Huawei Cloud are deployed to eliminate the expensive investment in purchasing physical hardware. However, the problems caused cannot be ignored. In extreme cases, unplugging the network cable will cause the public chain to go down, which seriously violates the spirit of decentralization.
On the other hand, decentralized computing power platforms either directly build a "computer room" to maintain network stability, or directly build an incentive network, such as IO.NET The airdrop strategy of promoting the number of GPUs is just like Filecoin’s storage of FIL tokens. The starting point is not to meet usage needs, but to empower tokens. One evidence is that few major manufacturers, individuals or academic institutions will actually use them for ML. Work such as training, inference or graphics rendering causes serious waste of resources.
But in the face of rising currency prices and FOMO sentiment, all accusations that decentralized computing power is a cloud computing power scam have disappeared.
Do two types of cloud computing power have the same name and same destiny?
Inference and FLOPS, quantifying GPU computing power
AI The computing power requirements of the model are evolving from training to inference.
Let’s take OpenAI’s Sora as an example. Although it is also manufactured based on Transformer technology, its number of parameters is compared to GPT-4’s tens of thousands. At the billion level, academic circles speculate that it is below the hundreds of billions level. Yang Likun even said that it is only 3 billion, which means that the training cost is low, which is also very understandable. The computing resources required for small parameters are also attenuated in proportion.
But in turn, Sora may need stronger "reasoning" capabilities. Reasoning can be understood as the ability to generate specific videos based on instructions. Videos have long been regarded as creative Content, therefore requires AI to have stronger understanding capabilities, and training is relatively simple. It can be understood as summarizing rules based on existing content, stacking computing power without brains, and working hard to create miracles.
In the past, AI computing power was mainly used for training, and a small part was used for reasoning capabilities, and was basically covered by various NVIDIA products. However, in Groq After the advent of LPU (Language Processing Unit), things began to change. Better reasoning capabilities, slimming down and improving accuracy of superimposed large models, and having a brain to talk logic are slowly becoming mainstream.
In addition, I would like to add the classification of GPU. It is often seen that it is the bad game players that save AI. The reason for this is that the game market has a strong demand for high performance. The strong demand for GPUs covers research and development costs. For example, 4090 graphics cards can be used by those who play games and AI alchemy. However, it should be noted that game cards and computing power cards will gradually be decoupled. This process is similar to Bitcoin. Mining machines have evolved from personal computers to dedicated mining machines, and the chips used in them also follow the order from CPU, GPU, FPGA and ASIC.
LLM special card is under development...
With the maturity and progress of AI technology, especially the LLM route, more There will be more and more similar attempts at multi-TPU, DPU and LPU. Of course, the current main product is still NVIDIA's GPU. All the discussions below are also based on GPU. LPU, etc. are more of a supplement to GPU, and it still needs to be completely replaced. time.
The decentralized computing power competition does not compete for GPU acquisition channels, but attempts to establish new profit models.
At this point, NVIDIA has almost become the protagonist. Basically, NVIDIA occupies 80% of the graphics card market. The battle between N card and A card It only exists in theory. In reality, everyone is dissatisfied with their integrity.
The absolute monopoly position has created a fierce competition for GPUs, from the consumer-level RTX 4090 to the enterprise-level A100/H100. Home cloud manufacturers are the main force in stocking up. However, AI-related companies such as Google, Meta, Tesla and OpenAI all have actions or plans to produce self-made chips, and domestic companies have turned to domestic manufacturers such as Huawei, and the GPU track is still extremely crowded.
For traditional cloud vendors, what they sell is actually computing power and storage space, so whether to use their own chips is not as urgent as AI companies, but for going As for centralized computing power projects, they are currently in the first half, competing with traditional cloud vendors for computing power business, focusing on cheap and easy-to-obtain computing power. But in the future, like Bitcoin mining, there is little chance of Web3 AI chips appearing.
An additional comment, since Ethereum switched to PoS, there have been fewer and fewer dedicated hardware in the currency circle. The market size of Saga mobile phones, ZK hardware acceleration and DePIN etc. Too small. We hope that decentralized computing power can explore a unique path for Web3 for dedicated AI computing power cards.
Decentralized computing power is the next step or supplement of the cloud.
The computing power of GPU is usually compared in the industry with FLOPS (Floating Point Operations Per Second), which is the most commonly used The calculation speed index, regardless of the GPU specifications or optimization measures such as application parallelism, is ultimately judged by FLOPS.
It has taken about half a century from local computing to moving to the cloud, and the concept of distribution has existed since the birth of computers. Driven by LLM, it has The combination of centralization and computing power is no longer as vague as it used to be. I will summarize as many existing decentralized computing power projects as possible, and there are only two dimensions of investigation:
The number of hardware such as GPUs, that is, examining their computing speed. According to Moore's Law, the newer the GPU, the stronger its computing power. The greater the number under the same specifications, the stronger the computing power;
Incentive layer organization method, which belongs to The industry features of Web3, such as dual tokens, added governance functions, airdrop incentives, etc., make it easier to understand the long-term value of each project, instead of focusing too much on short-term currency prices and only looking at how many GPUs you can own or schedule in the long term.
From this perspective, decentralized computing power is still based on the DePIN route of "existing hardware + incentive network" , or in other words, the Internet architecture is still the bottom layer, and the decentralized computing power layer is the monetization after "hardware virtualization". The focus is on access without permission. Real networking still requires the cooperation of hardware.
Computing power should be decentralized and GPU should be centralized
With the help of blockchain three In the difficult dilemma framework, the security of decentralized computing power does not need to be specially considered. The main issues are decentralization and scalability. The latter is the purpose of GPU networking, and it is currently in a state of unparalleled success in AI.
Starting from a paradox, if the decentralized computing power project is to be completed, the number of GPUs on the network must be as large as possible. The reason Without him, the parameter volume of large models such as GPT explodes, and GPUs of a certain scale cannot achieve training or inference effects.
Of course, compared to the absolute control of cloud vendors, at the current stage, decentralized computing power projects can at least set up mechanisms such as no access and free migration of GPU resources. However, due to the improvement of capital efficiency, it is possible that products similar to mining pools will be formed in the future.
In terms of scalability, GPU can not only be used for AI, but cloud computing and rendering are also feasible paths. For example, Render Network focuses on rendering work, while Bittensor etc. Focusing on providing model training, from a more straightforward perspective, scalability is equivalent to usage scenarios and purposes.
So two additional parameters can be added to the GPU and incentive network, namely decentralization and scalability, to form a comparison index from four angles. Please Please note that this method is different from technical comparison, it is purely for fun.
< /p>
In the above projects, Render Network is actually very special. It is essentially a distributed rendering network, and its relationship with AI is not direct. In AI training and inference , all links are intertwined. Whether it is SGD (Stochastic Gradient Descent) or backpropagation and other algorithms, they are required to be consistent, but rendering and other work does not have to be so. Videos and pictures are often cut. points facilitate task distribution.
Its AI training capabilities are mainly integrated with io.net and exist as a plug-in of io.net. The GPU is working anyway, so why? Instead of doing it, what is more forward-looking is its defection to Solana when Solana was underestimated. Later it was proved that Solana is more suitable for the high-performance requirements of rendering and other networks.
The second is io.net's scale development route of violent GPU replacement. Currently, the official website lists a full 180,000 GPUs in the decentralized computing power project. Being in the first gear, there is an order of magnitude difference from other opponents, and in terms of scalability, io.net focuses on AI reasoning, and AI training is a hands-on way of working.
Strictly speaking, AI training is not suitable for distributed deployment. Even for lightweight LLMs, the absolute number of parameters will not be much less. , the centralized computing method is more cost-effective in terms of economic cost. The combination of Web 3 and AI in training is more about data privacy and encryption operations, such as technologies such as ZK and FHE, and AI reasoning Web 3 has great potential. On the one hand , which has relatively low requirements on GPU computing performance and can tolerate a certain degree of loss. On the other hand, AI reasoning is closer to the application side, and incentives from the user's perspective are more substantial.
Filecoin, another company that mines for tokens, has also reached a GPU utilization agreement with io.net. Filecoin will use its 1,000 GPUs in conjunction with io.net. It can be regarded as a collaboration between the seniors and juniors. I wish you both good luck.
Once again, Gensyn is not yet online. We also come to the cloud to evaluate it. Because it is still in the early stages of network construction, the number of GPUs has not been announced, but its main use The scenario is AI training. Personally, I feel that the number of high-performance GPUs is not small, at least it must exceed the level of Render Network. Compared with AI inference, AI training and cloud vendors are in a direct competitive relationship, and the specific mechanism design is also will be more complicated.
Specifically, Gensyn needs to ensure the effectiveness of model training. At the same time, in order to improve training efficiency, it uses off-chain computing paradigms on a large scale, so model verification The anti-cheating system requires multi-party role play:
Submitters: task initiator, and ultimately training Cost payable.
Solvers: train the model and provide proof of effectiveness.
Verifiers: Verify the validity of the model.
Whistleblowers: Check validator work.
Overall, the operation method is similar to PoW mining + optimistic proof mechanism. The architecture is very complex, and the calculation may be transferred to the chain. It can save costs, but the complexity of the architecture will bring additional operating costs. At the current juncture where decentralized computing power is mainly focused on AI reasoning, I wish Gensyn good luck here.
The last one is Akash, who basically started with Render Network. Akash focuses on the decentralization of CPU, and Render Network first focused on the decentralization of GPU. Unexpectedly, after the outbreak of AI, both parties moved into the field of GPU + AI computing. The difference is that Akash pays more attention to reasoning.
The key to Akash’s rejuvenation is to take a fancy to the mining problems after the upgrade of Ethereum. The idle GPU can not only be used as a second-hand use by female college students, but also Now we can work on AI together. Anyway, we are all contributing to human civilization.
However, one advantage of Akash is that the tokens are basically fully circulated. After all, it is a very old project, and it also actively adopts the pledge system commonly used in PoS, but how? From the looks of it, the team is more Buddhist, not as youthful as io.net.
In addition, there are THETA for edge cloud computing, Phoenix for providing niche solutions for AI computing power, and legacy computing companies such as Bittensor and Ritual. Due to space limitations, I cannot list them all. The main reason is that some parameters such as the number of GPUs cannot be found.
Conclusion
Throughout the history of computer development, various computing paradigms can be constructed. The only regret about the centralized version is that they have no impact on mainstream applications. The current Web3 computing project is mainly self-promoting within the industry. The founder of Near went to the GTC conference because of the authorship of Transformer, not the foundership of Near.
What is even more pessimistic is that the current scale and players of the cloud computing market are too powerful. Can io.net replace AWS? If the number of GPUs is large enough, it is true. It's possible, after all, AWS has long used open source Redis as the underlying component.
In a sense, the power of open source and decentralization are not equal. Decentralized projects are overly concentrated in financial fields such as DeFi, while AI may be a key path to entering the mainstream market.
Preview
Gain a broader understanding of the crypto industry through informative reports, and engage in in-depth discussions with other like-minded authors and readers. You are welcome to join us in our growing Coinlive community:https://t.me/CoinliveHQ