What to know:
- 2-bit DeepSeek V4 runs at ∼35 tokens/sec on Apple, 7 on AMD within 90 GB VRAM.
- Focus shifts from “decentralized AI” to multi-hardware support for broader adoption.
- CROPS AI enables ZK-based LLM calls, private RPC reads, and Ethereum-tuned models for contract security.
Ethereum co-founder Vitalik Buterin is supporting the idea of integrating local AI with blockchain infrastructure. He notes that a real “CROPS AI” should be able to work effortlessly on different hardware platforms.
In his recent message, he mentions both technical developments and a re-orientation of the focus from only “decentralized AI” to wider hardware compatibility and utilization of Ethereum-native features.
DeepSeek V4 Benchmarks on Various Hardware
Buterin released the data on the performance of a 2-bit quantized version of DeepSeek V4, which fits in about 90 GB of VRAM. On Apple hardware, the model processes about 35 tokens per second and on AMD systems, 7 tokens per second.
These results indicate that capable large language models can be run locally. With that, there is less dependence on centralized cloud providers and the practicality for throughput remains for development and research purposes.
Also Read: Ethereum Price Wobbles Near $2K as Justin Bons Blasts Vitalik Buterin’s ETH Vision
Redefining CROPS AI
Vitalik said if we want real CROPS AI, we should stop thinking of it as simply decentralizing AI, but rather supporting many different platforms. This way, people will be able to use their existing devices to run the models, which is what really matters for adoption.
Also, this concept of being hardware flexible is very much a part of the spirit of Ethereum of enabling innovation without permission, even though it does acknowledge the practical side of deployment constraints of different kinds of processors like CPU, GPU, and specialized chip architectures.
Also Read: SoFiUSD Stablecoin Launches in SoFi App With Ethereum and Solana
Ethereum as a Platform and Keeping Protocols Secure
Buterin saw CROPS AI as also a “CROPS Ethereum access layer, ” which means ZK-based paid remote LLM calls and private Ethereum RPC reads.
He also expressed the need for AI models that have been optimized for Ethereum to increase the efficiency of smart contract auditing and the overall security of protocol code.
Also Read: Bitmine Holdings Expands After Buying 111,942 ETH During Ethereum Price Dip
Be the first to comment