- Vitalik Buterin tested DeepSeek v4 and shared local AI performance results.
- Buterin found that DeepSeek v4 runs faster on Apple hardware than on AMD hardware.
- Buterin linked CROPS AI and Ethereum tools to the growth of local-first AI infrastructure.
Ethereum co-founder Vitalik Buterin says the recently released DeepSeek Version 4 supports his local AI setup on the Ethereum network. In his latest post on X, Buterin highlighted a new 2-bit quantized version that runs about 90 GB VRAM, with the capacity to handle larger workloads on a single high-end machine.
DeepSeek v4 Hardware Compatibility
Following an independent test of the protocol, Buterin noted that the results showed about 35 tokens per second on Apple hardware and about 7 tokens per second on AMD hardware. According to him, this wide performance gap highlights why local AI will pay closer attention to hardware diversity in its next phase.
In the same post, Buterin noted progress on his messaging-daemon project, which now has alpha Telegram support, despite a complex setup procedure which he described as “quite janky.” He continued by mentioning Lucebox Hub as a promising way to run “dense” models such as Qwen 27B more efficiently.
CROPS AI, Hardware Choice, and Ethereum
Buterin drew his followers’ attention to “CROPS AI”, differentiating it from basic “decentralized AI.” Buterin thinks that supporting more than one hardware manufacturer is central to the difference between the two solutions’ categories. However, he hopes the ecosystem can improve on this, as more local-first tools launch.
In the meantime, Buterin noted that projects like VoxTerm, which provide local AI recording without third-party servers, reflect a broader shift toward user-controlled AI. According to him, there are several intersections between ‘CROPS Ethereum access layer’ and ‘CROPS AI.’ He thinks a zero-knowledge system for paid remote LLM calls could also be “just as useful” for private RPC reads on Ethereum.
Meanwhile, the blockchain expert highlighted application-specific fine-tuned models, calling them “another key piece” in the ecosystem. He cited Mistral’s Leanstral, which returns approximately 38 tokens per second on AMD using under 70 GB of VRAM, as a typical example. He explained that such tools are a huge boon for writing more source code.
How to Stay Competitive
Responding to concerns about Europe’s technological dependence on the United States and China, Vitalik Buterin said the solution is to rally around open source. According to Buterin, open-source collaboration is the only way for non-superpower regions to compete against dominant technology network effects while bringing more countries onto the same ecosystem.
Related: Algorithm vs. Altcoin: DeepSeek AI Sets Intriguing XRP Price Targets – April
Disclaimer: The information presented in this article is for informational and educational purposes only. The article does not constitute financial advice or advice of any kind. Coin Edition is not responsible for any losses incurred as a result of the utilization of content, products, or services mentioned. Readers are advised to exercise caution before taking any action related to the company.
