Originally posted by: Fox5
1. Your Athlon XP cpu, at a low clock speed and low voltage can be very power efficient. You should be able to get it to the point where a small passive cooler will be enough to keep it running (though not necessarily cool). However, the rest of the system is not power efficient. As far as I know, the old VIA motherboards implemented very little of the power efficiency stuff for shutting down unused components or going into suspend. Additionally, a low speed athlon xp has very little performance, you'd probably have to run a low end Linux system to get any kind of usable desktop out of that. (or an older version of windows, which are insecure, buggy, or lack power management features or some combination thereof)
2. In-order cpus benefit way more from hyperthreading than out of order. Out of order cpus support instruction reordering and are already able to extract some parallelism out of code. Considering the x86 assembly doesn't provide much to help extract parallelism from code explicitly, threading is one way of doing so. All in-order architectures benefit more so from SMT, but the x86 architecture is extra difficult to extract parallelism from the code so it benefits more so. BTW, the new i7's get way more of a boost from hyperthreading than the pentium 4 ever did, but they also have way more execution units available than the p4 did along with a very large L3 cache to keep them fed.
Multi-threading gain of various architectures on SpecInt2KRate:
Power 5: 21%
Pentium 4: less than 5%
Pentium 4 single core to Pentium D dual core: 76%
Pentium 4: 15%
Pentium Extreme Edition(Smithfield): -7.6%
I think the implementations of multi-threading on Atom and Nehalem will change people's perspectives on Hyperthreading. The gains are simply amazing here. In-order does take advantage of HT better, but Atom architecture is further optimized to maximize the benefits.
The xbox 360 cpu cores are also in order cores with SMT. From developer comments, it appears that in general, a single core performs about as well as a very low end p4, but with SMT performance increases to the level of a mid to high end Pentium 4 and can even approach clock per clock performance parity with it. (though for floating point SIMD, xbox 360's cores blow away a Pentium 4, I believe they're rated at ~128GFlops combined, whereas i7 is at ~55 GFlops for all 4 cores, though I bet i7 comes much closer to its theoretical rating than the 360)
The Niagra server processors are also in-order and can handle up to 8 threads per core. For certain work loads, for far less power, a low clock speed, and a tiny size per core, they can outperform even high end opterons.
Atom sees some decent games from hyperthreading, but is still outperformed by both semprons and celerons. With hyperthreading supported apps, the atom can often catch up to the 1.2Ghz Celeron though. Of course, there are plenty of apps where hyperthreading doesn't help, and you've got something half as slow as the 1.2Ghz celeron. In many apps, atom looks to provide poor performance/watt.
Also, the Sempron with cool n' quiet (or a celeron clocked really low) use about as much power in idle as the atom. They use much more under load, but if you're concerned about power efficiency and not a tiny form factor, you'll spend most of your time idling anyway.
Edit: Note that's Atom with an inefficient desktop chipset. Power consumption should drop with a more power efficient chipset, though I'd imagine the same can be said for the Intel and AMD platforms. Not to mention they have a real performance advantage, and could be downclocked to their minimum and still perform as well as or better than atom.