More Recent Learnings from SPCR
The Landscape of CPU energy utilization has been fundamentally altered with the new generation of Intel Sandy Bridge Chips.
My rule of thumb is
1. Cache is good for gaming
2. Number of cores is good for encoding
3. GHz is good for responsiveness.
4. memory frequency/timings are important in all three
5. Memory amount only matters when its not enough.
Its a quick and dirty summary, can't be right in all cases.
Btw Sandy Bridge processors break these rules in many aspects.
First, its cache system and memory controller are very efficient. So size of L3 cache (LLC) does not play a major role anymore.
And effect of memory speed and timings is diminished.
Then Hyperthreading kicks in and blurs number of cores effect. i7-2600 and i5-2500 have 4 cores and similar speed, in encoding - very different.
And then SB are 20% faster per MHz ( I saw up to 80% faster compared to phenoms), can no longer be compared directly to other processors.
It used to be if you wanted a quiet computer, you would select not the fastest chip, but one that would be just fast enough, and then you would under clock it, then starve it of fan driven air... intentionally letting it run a bit on the hot side. If you failed to take this route, you would get high levels of thermal load inevitably accompanied with unavoidable noise.Some New Data
The AMD Fusion on an Asus E35M1-M Pro Motherboard uses
a total of 18 watts at idle, and
a total of 28 watts under an average load (H.264 playback) http://www.silentpcreview.com/article1167-page5.html
(measured at AC which includes 40-60% inefficiency at the PSU level)
On the other hand
The Intel Core i5-2500K running on an Intel DP67BG motherboard uses
a total of 16 watts at idle, and
a total of 23 watts under an average load (H.264 playback) http://www.silentpcreview.com/article1148-page5.html
(measured at DC which doesn't include PSU AC to DC conversion inefficiencies)
Adjusted for PSU AC to DC conversion inefficiency
estimated 23 watts at idle, and
estimated 32 watts under an average load (H.264 playback) Here is the Kille
r, most of the Core i5-2500K wattage is not in the Intel Sandy Bridge CPU which idles at around 4 - 5 watts. http://www.xbitlabs.com/articles/cpu/di ... html#sect0
What if you go to something like the 2400S or 2300. It makes not Difference... idle is still about 4 watts. http://www.xbitlabs.com/articles/cpu/di ... html#sect0NEW RULE NUMBER 1 - It's not how big it is, it's how you use it.
No longer does it make sense to get the smallest weakest CPU you can get by with. Most of the time your computer is operating at idle. That is the usage level that determines your electricity bill at the end of the month. If you had a 0.00 watt chip, all you would save over the Core i5-2500K (which I think more powerful as a 6 core AMD and which is more powerful than all but a handful of Intel super CPUs) is about 5 watts. If you compare it with a real live AMD Fusion or ATOM, the savings is less.NEW RULE NUMBER 2 - It's the Motherboard.
The motherboard is now the greatest source of heat (and therefor noise). The difference in performance between an Intel Core i5-2500K running on an Intel DP67BG and an Asus E35M1-M Pro (basically an Atom with IGA on steroids) is outlandish... Tom Thumb standing next to Andre the Giant.
YET......... the difference in wattage between the two CPUs is less than 4 watts. The difference between wattage between the two systems is much larger. This is because of the motherboards (probably in large part the accompanying Intel Chip set).
There is one indirect heat source driven by the 2500K that is not matched by the AMD Fusion. The AMD Fusion can use a small PSU and use it at levels where it can run at high efficiency. The 2500K, even though it is always running at just a few watts has a higher top end. It is capable of burning up 95 watts if you let it rip. So it needs to be provisioned with a larger PSU even though it doesn't need it in real life. That larger PSU will generally be less efficient at 15 to 30 watts of output than the smaller PSU needed by the AMD Fusion system.
MikeC on low power / high efficiency PSUs (I think maybe contradicting the above... and if so disregard what what I say on the subject and listen to what he says):
NEW RULE NUMBER 3 - We need to start paying even more attention to the efficiency of Motherboards and PSUs
One of the most efficient ATX PSUs is the Seasonic X400. Compare that to the SS-400ET:
SS-400ET: 32W AC = 21.2W DC ; 58W = 41.6W
X400: 31W AC = 21.6W DC ; 53W = 42.6W
At 21-22W output, the difference in AC in is almost too small to be significant. At 41~43W, yeah, there is a couple watts difference (about 4W), but that isn't the power level you think is significant (and neither do I).
So how could we get say under 30W AC power draw at 20W? Anything more than ~28W would probably not be worth the pain of the change, as a lot of the low power boards would have to be retested for future comparisons to have much relevance. And that ain't no easy task, as lots of them have been moved on, no longer in our hands.
The only PSU that has ever drawn lower than 32W AC input for 21-22W DC out is the picoPSU (with a high enough efficiency 80W brick): http://www.silentpcreview.com/article601-page3.html
Which is "an oddball PSU solution". We could use it... but I would need to identify/shortlist/test some high quality AC/DC bricks with enough power capacity and good voltage regulation so it would not taint the max power readings on higher power boards/CPUs/systems.
If you want a quiet computer the action is now shifting from the CPU to the Motherboards and PSUs. SPCR has always paid attention to these kinds of things. I guess they have never been unimportant. But they are now becoming ever increasingly important to anyone who is building a quiet and efficient computer. NEW RULE NUMBER 4 - Economic Tradeoffs are Changing
For fewer and fewer people is a small anemic, but good enough, single purpose computer going to make economic sense.
The difference in cost between the AMD Fusion / Asus E35M1-M combo and the Intel 2400 / Intel DP67BG combo is about $100. That is less than the cost of the operating system. The two core Sandy Bridge CPUs will likely bring that cost differential down to about $30. Will this vary from retailer to retailer and country to country... sure, but not that much.
Funny thing is that the two core Sandy Bridge CPUs are not likely to be much more energy efficient than the four core ones. How low can you go when you are already that close to zero.NEW RULE NUMBER 5 - Supplemental - You can no longer talk about performance without talking about SSDs
Disk I/O is as important to much if not most usage as CPU cycles. Having the fastest SSD isn't so important as having one as opposed to relying on a traditional hard drive. Within the next 12 months, relying on a trusty old hard drive will be similar to what people did when they chose to rely on their trusty old floppies instead of those new fangled hard drives. It is the same story... repeated over thousands of years since the construction of the pyramids.... only the names have been changed to protect the innocent. ERRATUM - Corrections Welcome
Did I overlook some details or subtleties? No doubt. I just through this together like an email.... with all its typos.
Nit picking is welcome, supplementation even more so. But please attack what is being said, not the person saying it. I know this message threatens some personal belief systems out there. But hey this is a hobby not a religion. Don't take it too seriously or personally.... and count to 10 before you get out the flame thrower.
So please let me hear what you think about the above.