Notes from AMARBARO's projects — measured results, the protocol they were measured under, and the predictions that turned out wrong.
GPU idle was 60 percent of decode wall time, and the largest share was host-to-device copies running at 3.1 to 3.4 GB/s against 28.5 GB/s on a healthy boot. The root port negotiated 2.5 GT/s. Several timed A/Bs from that boot are void, one fusion ratio survives, and the machine is still on the slow link.
Std.math exp disagreed between the CPU and the GPU on 19,437 of 40,017 inputs. A shared exp written with explicit fma matched on all 40,017, at 3.6 times the cost in an exp-bound kernel. One attention layer then matched bit for bit, and a float Freivalds certificate caught every planted fault.
ReadSpreading activation helped only inside a small, dense topical neighborhood. The vault scored 23 of 29 with the network on and 23 of 29 with it off, and a 5 percent slice of Wikipedia diluted the gain to 0.7 percent. Adding hub links to every note changed nothing.
ReadThree model passes over one-line index rows flagged four families. Each failed against the operators the engine actually runs. The screen yielded no engine feature, and the four vero claims and the Float64 log fix both predate it.
ReadFour days of a wrong verdict ("unusable with our Mojo kernels") ended when we unbound the integrated GPU. A plain kernel then read SQ_WAVES 32,768 exactly, and a second fault, counters reading zero, came out from behind the first.
ReadMojo has no ray-tracing API, yet a 20-line probe emits image_bvh_intersect_ray for gfx1100. The probe is compile only, with an all-zero descriptor and one component set, so it proves reachability and nothing more.
ReadEight raw hidden vectors did not improve accuracy in a same-model 120-task rerun.
ReadFive paired 50-condition runs matched recorded metrics and gain vectors, with a 24.44x median runner speedup.
ReadA day with 135 GPU jobs and 34 failures, none of which needed the GPU to find. The fix is not a faster GPU, it's treating the card as a shared, budgeted resource with a preflight step before anything touches it.
ReadAn OS design in which the message between agents is a latent tensor with an identity, and text is one adapter among several. The transport and tiering numbers are real, the fork gate is resolved, and the thesis experiment has not passed. That experiment asks whether an untrained model reads its own hidden states as reasoning.
Read
