ok so something i hadnt thoUght about with cometbft here the validators verifying proofs dont need gpus at all just commodity hardware while the actual inference work needing gpus happens on a completely separate set of noDes.
that split is kiNd of the whole point honestly. you dont want validation to require the same expensive hardware as the work being validated otherwise you have just recreated the same accessibility problem on a differrent layer.
so technically anyOne with a normal computer could run a full node and help with consensus the actual ai compute stays gated behind whoever can afford gpus. decentralized validation sitting on top of a less decentralized execuTion layer underneath it.
dont know if anyones acTually broken down what percentage of the network is full nodes versus inference nodes that ratio would tell you a lot more than just the network is decentralized as a blanKet claim.
@OpenGradient $OPG #OPG chat.opengradient.ai
that split is kiNd of the whole point honestly. you dont want validation to require the same expensive hardware as the work being validated otherwise you have just recreated the same accessibility problem on a differrent layer.
so technically anyOne with a normal computer could run a full node and help with consensus the actual ai compute stays gated behind whoever can afford gpus. decentralized validation sitting on top of a less decentralized execuTion layer underneath it.
dont know if anyones acTually broken down what percentage of the network is full nodes versus inference nodes that ratio would tell you a lot more than just the network is decentralized as a blanKet claim.
@OpenGradient $OPG #OPG chat.opengradient.ai