If you look at pages pages 7 and 8, it lists the power requirements as 182W for the "typical configuration" and 279W for the "high performance system".
(Yes, going by the maximums for all devices is probably a bit over engineered. The AMD document does recommend 80% of all non-cpu devices. But playing it very safe for the moment...)
I don't know where they got their current draws for each device. But assuming they're realistic, I would think a many users would be in 200W+ category. A power user could really hit 300W with the numbers in that PDF. Dual Athlon systems would certainly break that, right?
Back to the subject of testing power supplies; it seems that it would be nice to have the equivalent to dynometer for cars. There would probably be more graphs... some of the spec sheets would provide some of that information, but they don't seem consistent..... But it would be nice to be able to verify their claims too.
The 80% guide brings those numbers down to 162.47 and 241.91 -- well under 300W.
If you think about it, even their 80% rule is something of an overkill: you really can't have 100% CPU usage AND access all the various drives & perihperals. It's just not possible -- at least not in any version of Windows I have used. Perhaps Linux is better. Multitasking simultaneously with 100% usage in any version of Windows just bogs the system down so much you hardly get any productivity, so in normal usage, I never try to do that. Like if I am doing a major transform function in Photoshop on a large image file, there's not much point trying to do a search on the HDD. It's just slows everything down to a crawl. I'm better off to wait the 30 secs.
With the test PCs, at 100% CPU load with various testing utilities, I opened both CD drawers & inserted a data disk in each & closed it at the same time. Not very productive, as it took forever to read the CDs, but even then the momentary peaks never hit more than ~135W AC power draw. That's like 100W power delivered on the DC lines. Now it is very possible the kill-a-watt meter can't repond fast enough to catch really fast peaks that ran higher, but almost any PSU can deliver much higher than rated power for a few milliseconds, I would think.
I guess what I am saying is that if you look at PSU requirements from the POV of getting the least amount heat so that it can be cooled with the minimum of fan noise, and consider typical usage on typical desktop systems, current recommendations for PSUs aren't that sensible. Manufacturers have to be a bit like lawyers -- they look at worst case scenarios, and that's what most PSUs are designed for (although I bet many don't cut it anyway). In real world desktop apps, it just seems almost impossible to reach such high power draws.
It would make more sense to me if PSU makers made PSUs specific to desktop use, with very clear definitions about what this means, and other PSUs specific for "industrial" use -- ie, as they are currently made. While we're at it, why not also do a complete remake of all the critical power connectors in a PC for minimum losses. There have got to be better connectors that the ones used today.
That's about all for now, as I balance precariously out on this long thin limb, scanning the skies anxiously for the attacking crows soon to come...