Month of data on repurposed mining hardware for AI been…
0 reactions
Month of data on repurposed mining hardware for AI
been loosely following this network (qubic) that routes mining hardware toward AI training. about a month of data now
what they've shown: existing mining hardware can run non-hashing workloads at decent scale. seems stable, good uptime, economics work for operators
what they haven't shown: whether the training output actually competes with datacenter compute quality-wise. still no independent verification
honestly if the AI part turns out to be real that's a genuinely interesting approach to the compute access problem. if it's not then it's just mining with extra steps. someone needs to actually benchmark the output against known baselines
#technology
been loosely following this network (qubic) that routes mining hardware toward AI training. about a month of data now
what they've shown: existing mining hardware can run non-hashing workloads at decent scale. seems stable, good uptime, economics work for operators
what they haven't shown: whether the training output actually competes with datacenter compute quality-wise. still no independent verification
honestly if the AI part turns out to be real that's a genuinely interesting approach to the compute access problem. if it's not then it's just mining with extra steps. someone needs to actually benchmark the output against known baselines
#technology
| 2,000 mlx | |
| 2 | |
| 0 |