I didn't say power didn't matter. People generally don't use power consumption as a metric over performance as a bullet point unless their preferred card has already lost that performance race. Power is an important factor when reviewing a card as a whole but unless the TGP is totally f***** most people will buy for higher performance over lower power consumption.
Power draw by itself doesn't mean too much, but it does set the bar for heat and noise output. Noise can be dealt with at a cost if the available coolers aren't great, but generally that means that the heat output is felt even more.
This is also important for smaller systems. More efficient parts means less performance compromise for SFFs, whether that be due to hard cooling limits or just not wanting the system to sound like a jet engine. Also a pretty big deal for laptops though users usually don't get much choice there.
But if power draw is close, say within 10% under load? Point taken.
Good point.
I mostly see efficiency as it relates to being able to crank up the clocks more I guess, but your right, not everyone looks at it that way.
This, but note that being more efficient doesn't always translate into being able to crank up the clocks. It's pretty clear, for example, that AMD would probably give up no performance ground at all to Intel if they could get another 10% or so more clockspeed out of Ryzen regardless of power draw. As it stands AMD is likely to retain the efficiency advantage for a few more years while still seeing performance challenges from Intel despite Intel's fab troubles, simply because Intel can clock their parts higher.
With GPUs it's really all over the place though. Nvidia seemed to take the efficiency ground back with Kepler, but I think that the majority of that has been due to AMDs inability to keep up with Nvidia's pace of innovation since as they've only been able to present a challenge between Nvidia's release cycles and only by pushing out parts that stretched power envelopes and users' willingness to subject themselves to screaming blowers, like the R9 290 cards. And those were
only 250W stock!
For what it's worth, chiplets have a lot of promise, and AMD has shown with their CPUs that they can navigate around the many potential pitfalls of breaking up a processor without unduly affecting performance. The greatly increased memory latency on Zen 2 CPUs just doesn't seem to matter with all the cache they threw on the memory controller die, for example, and since GPUs are far less latency sensitive than CPUs, they very well could pull this off.
Still, I don't envy anyone who has to deal with the first round of drivers
