Diff of Autogollark at d7869b0

@@ -40,3 +40,3 @@ Autogollark currently comprises the dataset, the search API server and the [[htt
 * Maybe compute grants are available for training.
-* Substantial bandwidth bottleneck on CPU. Specdec/MTP would be useful.
+* Substantial bandwidth bottleneck on CPU (230GB/s nominal; 200GB/s benchmarked; 100GB/s per NUMA node, which llama.cpp handles awfully). Specdec/MTP would be useful.
 }