TR says that a single core can pump 1W according
to their JS number , wich is a single threaded task ,
hence 4 cores can use 4W on peaks , far from
Anand s number wich should be of equal amplitude
given that CB is FP heavy and FP ops are power hungry
tasks.
Quite true. But it's also the case that we don't have benchmark results for Baytrail-T running a multi-core workload while drawing 4+ watts. We only have benchmark results of multi-core workloads while consuming roughly 2.6 watts. Heh, and this is one of the reasons why TDP is a difficult figure to specify - sure it might be possible to hit a certain value, eg 4 watts, but if the power control logic doesn't let the SoC run in any state that goes beyond 3 watts then what's the correct TDP number to report?
Which pretty much mirrors what we see in the benchmark results - the multi-threaded cinebench results for Baytrail appear to be about 90% of what we'd expect to see if frequency remained constant. Of course that's assuming that the single-threaded result was running at a frequency that's near the maximum and requires an extra 0.1-0.15 volts (if the design had more frequency headroom then we'd likely see some of the many higher power parts take advantage of it.) Since yeah, running at 90% frequency and dropping from 1.1V to 1V would already bring 4W down to 3W.