All is compiled kind of per the lowest common denominator, except for I think UGM which requires presence of SSE2 making it 10x faster. Don't know if the programmers test SSE2 and other features routinely for benefit, but since a small percent was lost already because of UGM, I would and count the pro/con... 10x more throughput by loosing 2% is not a hard choice. Other than that, the more a processor core can do in a cycle, the better, but how to determine cycle for cycle? Oh and of course watts per hour put in the equation.

BTW there was claim that HST1 was 20% faster [on Linux], suggesting there to be a higher integer component, a wild guess but I'm certainly going to try it out, if able to get work equal size on the different platforms.

I would consider 36 cores/72 threads @ 320 watts a huge advantage.

So you don't think a 72t machine that pulls the same amount of wattage as my i7-920 8t used to pull is not an advantage? Whatever. You obviously have no clue. O and O.

Newer architecture always increases performance by small margins. The OS has allot to do with it to. If your running windows maxed at all threads, and all the crapware in windows that wants some time too, you loose a little. Reason why linux runs better. No crap in it. I always leave 1 thread per processor open on my windows crunchers to let the OS process the crapware. Cost to much to keep up with modern processors. Find a good balance and crunch. My westmere procesors are still doing pretty good.

I believe that the performance/watt thing is a bit infected? It is simply not that simple.
Let's consider all the energy and resources (man hours, pollution of air, water, soil etc) needed to:
1. Manufacture that new server/PC (with high efficiency) with all the components that you want to buy (Motherboard, CPU, SSD, RAM, PSU etc).
2. Dig up all the precious metals from the ground (Gold, Silver, Palladium, Tantalum etc) from a mine in some country where Quicksilver also may be used and a lot of diesel from drills and CAT-machines.
3. Destroy the old hardware that you decide not to use (because you're buying new hardware). All the components will need to be recycled and taken care of.

I am convinced that it is the same as with new cars - they put out less emissions but usually you save the environment by using an already existing car until it is beyond repair.

(Also see for instance: http://www.lowtechmagazine.com/2009/06/embodi...f-digital-technology.html)

So, using old hardware (of course there is a limit as to /how old/) will probably be more net energy efficient than to buy new stuff and throw away the old.

Let's just take a simplified example:
If an old PC during an hour consumes 300Wh at WCG full CPU load and a new one consumes 200Wh to get the same amount of WCG crunching done you have a net win of 100Wh. If the PC would be crunching 24/7 it would mean 30 * 1000Wh = 30kWh/month or 360kWh in difference in a year. Let's say you use that old hardware for an additional three years, then it's 1080kWh in difference. That sounds like a drop in the ocean to me if considering the energy difference compared to manufacturing a new PC. Also, after those three years, even the new PC will not be /the most efficient one/ anyway...

Let's say you instead try to drive your car a bit less and when you do you apply eco-driving for instance, make sure you have optimum air pressure i tires etc. If you save 5 litres of oil per month by doing these things it would equate to 60 litres of oil in a year = 600kWh of energy in one year = 1800kWh during the three years of old PC usage as per the example above. In fact even much more since that oil needs to be refined etc.

What I am trying to say is that it is almost always better to cut energy consumption (for whatever purpose) by reducing consumption of already existing hardware that to buy new.