Kuzco is a specialized LLM large language model computational power mining network. This year, it was selected for the Crypto Startup Accelerator (CSX) Fall Accelerator Program launched by a16z in New York. The projects selected for this program will receive at least $500,000 in investment from a16z within 8 weeks, and will receive guidance and support from the a16z operations team.
Kuzco is still in its early stages and belongs to the GPU computing network field, similar to io.net. However, io.net has gained more market exposure as it has completed user airdrops earlier and successfully launched on mainstream cryptocurrency exchanges such as Binance, Coinbase, and Bybit, thanks to its strong financial support and background.
I have personally participated in mining on io.net and, to be honest, the whole process was like making money effortlessly with graphics cards. You just need to hang the graphics cards and they will automatically generate income for you, without even spending much on electricity. Each mining machine consumes only 1.2 kWh of electricity per day, totaling only $10.8 in three months. With the depreciation cost of used graphics cards and the miscellaneous expenses of internet tools, the cost of $100 to $200 can easily be covered. In the end, I earned $4,000 in revenue, which seemed like a “castle in the air” that kept rising.
The subsequent decline in token prices has verified this point. In addition, I have compiled an article titled “70% Price Crash: How Did the AI Computing Power Rental Bubble Burst?” which further reflects the bubble in the AI computing power market. Both the rental market prices and the decline in project tokens are the market’s correction of the true value of these projects. This price correction is in line with market laws and indicates that the future may develop in a more rational direction.
Kuzco has significant differences in its mining mechanism compared to the previous popular project, io.net. io.net is like a “castle in the air,” easily earning tens of times the profit by hanging graphics cards, but the actual value behind it is not obvious. On the other hand, Kuzco is more down-to-earth and has received support from a16z. I have spent several weeks testing mining machines with different configurations, including single cards, multiple cards, and different system setups. Through these practical operations, I will share these operational experiences with everyone, allowing you to understand a mining project that provides practical value to users, rather than a false prosperity supported by a bubble.
Specialized LLM Large Language Model Computational Power Mining Network
Kuzco is a decentralized GPU network in the Solana ecosystem. It aims to utilize the idle GPU resources of individual users to provide efficient and affordable computational power services for large language models such as Llama3, Mistral, and Phi3. Users can use these models through OpenAI-compatible APIs. Kuzco’s distributed architecture can integrate idle computational resources to run large language models, while incentivizing miners who provide computational power through a reward mechanism.
Project Operation Status
As of October 21st, the Kuzco project has 2,000 online GPU miners, reaching a peak of 6,000 at one point. The most commonly used GPU models include 3090 and 3060. Miners currently receive KZO Point rewards, but these points cannot be cashed out yet, and the project has not announced its token economic model. I speculate that with the advancement of the a16z accelerator program, the project may have more new developments and updates in the future.
Deployment
Official Hardware Requirements
The Kuzco project can run on Mac, Windows, and Linux operating systems, supporting various hardware configurations. The minimum system requirements are 16GB of RAM, 30GB of available disk space, and at least 10MB/s of network bandwidth. Kuzco supports NVIDIA (N card) and AMD (A card) graphics cards with at least 8GB of VRAM. The minimum requirement for ordinary users’ N cards is GTX750, while most professional mining graphics cards are supported.
However, I do not recommend using A cards for mining as AMD graphics cards have poor compatibility, especially in AI tasks. If you must use A cards, you need to check the official compatibility support list. In the field of AI, I must say: AMD, NO!
My Five-Card Platform, ASUS Z490 Motherboard
I have tested the mining efficiency of several hardware configurations, but the efficiency performance may fluctuate due to network fluctuations.
GTX1070: 20 tok/s
RTX2060: 30 tok/s
RTX2070S: 40 tok/s
RTX3080: 80 tok/s
RTX4060Ti: 50 tok/s
RTX4070S: 70 tok/s
M2: 20 tok/s
M3: 30 tok/s
Unit: Avager Tokens/Second (average calculation power completed per second)
Real-Time Monitoring of Mining Machine Operation
Deployment Method
Kuzco provides a client application that users can download and use to start mining. However, this method is sometimes unstable and may experience disconnections without automatic restart. I recommend a more stable approach, which is to use a Linux system or the Windows WSL (Windows Subsystem for Linux) environment to start mining through the command line (CLI) or Docker containers. If multiple GPUs are used, multiple GPUs can be specified for multi-card mining in Linux through Docker containers. For example, to specify GPU0: “docker run –rm –runtime=nvidia –gpus ‘”device=0″‘ -d kuzcoxyz/worker:latest –worker ", multiple GPUs can be started.
When mining with multiple cards, please pay attention to the following hardware aspects:
Power Supply: The power supply is crucial and should not be compromised. It is recommended to purchase a power supply based on the standard of "1 RMB = 1 watt" and use a gold-rated power supply if possible. When mining, choose a 1500W to 2000W power supply according to the number of graphics cards, or use multiple power supplies. However, multiple power supplies require additional startup lines to connect to the motherboard for normal power supply. Otherwise, the system will not start properly.
Cables: In high-power environments, power cables are prone to damage. Therefore, I recommend using higher-quality power cables. In addition, different brands of modular power supplies use different cable interfaces, and they are not interchangeable. Using cables from different brands may cause equipment damage or burnout. Therefore, the cables must match the brand of the power supply to ensure compatibility.
Motherboard: Each channel of the motherboard (regardless of x1, x8, x16) can only support one graphics card. For example, if there are several channels, they can support several graphics cards. B85 motherboards, which were popular during the Ethereum mining period, are a good choice.
CPU: The more threads the CPU has, the better, as it needs to handle multiple tasks simultaneously. When starting mining with Docker, it will initially consume a large amount of CPU resources. If multiple graphics cards are used, they must be started one by one in order, otherwise the system may crash or freeze.
Graphics Cards: One Docker process occupies approximately 6GB of VRAM (official documentation states 8GB, but 6GB is sufficient in practice). If a graphics card has 12GB of VRAM, you can run two Docker processes on one graphics card. During mining, the workload of the graphics card will occupy 50% to 90%, and the temperature of the graphics card should be kept below 85 degrees Celsius for it to be reasonable and safe.
PCIE Extension Cables: It is recommended to use x1 to x16 PCIE extension cables, with x1 plugged into the motherboard and x16 plugged into the graphics card. If using 40 series graphics cards, an x16 extension cable is required.
Network Connection Quality: The network connection quality has a significant impact on mining efficiency. Through testing, I found that tasks received from Singapore network nodes are more frequent than those from Hong Kong nodes, which means that choosing better network nodes can improve mining efficiency.
If any malfunctions occur during mining, first check the running status of the motherboard and software. If the problem is at the software level, you can determine the cause by checking the terminal error messages. It may be a memory error that requires a computer restart, or it may be due to the failure to timely update the local mining machine's code caused by updates from the official files. The solution is to change nodes or update the code.
At the hardware level, if the mining machine cannot boot, check the fault lights on the motherboard. Using ASUS motherboards as an example, the most common issue is the VGA light turning on, indicating a problem with the graphics card power supply. In such cases, try unplugging and reconnecting the PCIE and graphics card power cables. However, sometimes the white light may still be on while the machine continues to run normally.
In conclusion, although I referred to io.net as a "castle in the air," indicating that its market value was significantly overestimated, it did manage to secure $40 million in financing with a valuation of $1 billion. However, after the launch of io.net, numerous imitation projects emerged, and the products and backers behind these imitations could not withstand rigorous scrutiny, indicating that the success of io.net cannot be easily replicated by every project.
Based on this observation, I have been searching for a mining project that has more practical value. Eventually, I discovered Kuzco. Kuzco has gained incubation and support from a16z, which increases its credibility and potential. Moreover, Kuzco's mining mechanism truly provides computational power services through GPUs.
Furthermore, from a macro perspective, large language models (LLM) in the field where Kuzco operates are the most widely used AI products by the public. Countless people around the world use these models every day, and these models require significant computational power support. I believe it is meaningful to provide computational power for such a massive demand, and it has practical commercial value. Therefore, Kuzco is a project worth paying attention to.
Additionally, participating in the GPU computing network has low costs, especially with the stable prices of 40 series graphics cards in the second-hand market and their low depreciation costs. However, I would like to remind you not to choose to rent graphics cards, as renting cards costs much more than buying second-hand graphics cards. At the same time, Kuzco's airdrop incentives are not clear, so renting cards on a large scale may carry higher risks.
Moreover, mining machines themselves have high practical value. They can not only be used to mine Kuzco but also provide more stable and reliable returns compared to high-risk altcoins. Due to the scalability of mining machines, they can be used to mine other GPU projects or be validators for blockchain nodes, running scripts and services to generate more income. This means that even if you stop mining Kuzco, the mining machine can still generate value.
Finally, many people ask how much profit can be earned in a day from mining. There is no definitive answer to this question, and no one can accurately predict the final profit except for the project team. Mining returns have a high degree of uncertainty and can be either high or lower than expected. Therefore, it is impossible to determine in advance how much money can be earned.
