Block a user
research: Goldilocks multiplies on the INT8 tensor cores
Go/no-go micro-benchmark: NO-GO
Ran step 1 on beast (one 5090, miner paused, 348,160 threads x 2000 iterations x 12 constant multiplies each, both paths ending in the same fused Goldilocks…
research: Goldilocks multiplies on the INT8 tensor cores
perf: fused-PTX field arithmetic, +54..66% on every card
perf: fused-PTX field arithmetic, +54..66% on every card
perf: parameter-bank uniforms and per-arch carry path (+20.5% sm_120)
perf: parameter-bank uniforms and per-arch carry path (+20.5% sm_120)
Node metadata is only reachable through the mining inference, so most nodes are invisible
Node metadata is only reachable through the mining inference, so most nodes are invisible
Cache the node metadata the telemetry feed already sends, and show it on the miner route