What happened to the Power Consumption Woes?
11.13.06 - 09:53pm
Moore’s law is constantly mentioned whenever progress comes up in regards to computing. This simple law roughly states that computing power doubles every 18 months. Power consumption has followed in the steps of computing power and I would like to say it peaked 4 years ago or so with the onset of the Netburst(Pentium 4) architecture. Now that doesn’t make much sense, does it? Power consumption has continued to rise, so how could it have peaked years ago? One simple performance per watt PER core.
Power consumption per processing core peaked with Netburst(Pentium 4) when you had a single processing core capable of consuming over 200 watts in overclocked states. The focus for the last 2 years in processor development have been performance per watt, do as much work as possible on as few watts as possible. This trend lead to the development of the Dothan(Pentium M), Yonah(Core Duo), Conroe(Core 2 Duo), and now Kentsfield(Core 2 Quad) processor lines. Unlike their predecessors from the Netburst era that leaked electrons like sieves, the new like of mobile derived processors are rather energy efficient while performing enormous amounts of work but overall power consumption has still continued to rise, just not particularly within the processors.
Take a good look at todays modern highend motherboards. While they don’t describe the majority of boards out their, their attributes can be seen through ought the industry through a trickle down effect. A year ago you would have a hard time looking for a motherboard with a heatpipe based cooling solution along with an enormous heatsink on the southbridge. Over time motherboard chipsets have begun to seriously eat into your powersupply with the Nforce4 and Nforce5 chipsets being notorious for being power hungry beasts. Intels 975x chipset isn’t much better off as multiple 975x motherboards feature extreme cooling solutions and still hit record temperatures.
Besides the motherboards themselves, the RAM you put into that board has really started to add to your power consumption, especially if you are dealing with modern server grade memory. FB-DIMM based motherboards have been proven to consume up to 5 watts extra per ram module compared to standard DDR2 based motherboards. If you look at high end memory modules you will see elaborate cooling solutions featuring fins, grids, and sleeves in an attempt to keep the modules cool.
Five watts here, twenty watts here, those will add up over time but the largest increase in power consumption recently has been high end video cards. The 7900GTX and X1900XTX series cards consume a large amount of power with the cards consuming ~85 and ~125 watts respectively. Considering how the average power consumption on a stock Conroe is only 65 watts, if you have an X1900XTX it could be consuming nearly double the power of your processor. The new line of DirectX10 cards consume even more power than the previous generation however performance comes with this consumption. I haven’t got any numbers in regards to exact power consumption on the next gen cards but I can safely say they will offer atleast a 50% increase in most games along with a hefty power increase.
I really didn’t have much direction where I was going except to document what I noticed as a slow trend towards more power hungry architectures, once again. Unlike the previous power consumption “bubble”, clockspeeds aren’t the focus but total computing cores are. The new quad core chips are going to be beasts and consume beastly amounts of power and I imagine people will have trouble keeping them cool. I’ve seen reports of a Kentsfield at 3.6 GHz consuming 300 watts for the processor, motherboard, and ram. Considering how the board and ram might consume 75-100 watts, you are looking at 200 watts going through just the processor. I worry about how the power management systems around the sockets are going to handle these sudden surges in power consumption, looks like everyone and their brother will need phase change cooling just to keep their computers chill. By the way, I sell phase change cooling system. Enjoy the pictures and no I’m not obssessed..
Pictures from Tech Report, they wrote up a great review.
Well, I guess you could say that performance is directly dependent on power. And along with this new advent in performance, PSU’s have doubled and tripled in wattage… Alienware’s now selling a 1000 watt PSU as default with their Quad SLi. I wonder who’s going to have a 1.5k watt PSU first.
Quad SLI, something I detest to the utmost extreme. Performance is linked to power but more power doesn’t necessarily mean more performance. Right now the biggest culprits are parts that have been flying under the radar or have been considered acceptable to draw lots of power. I guess I just wish it was possible to have a high performance video card that drew under 75 watts, it’s been a while since that has been possible.
Where did you see the report about the Kentsfield at 3.6GHz? Everything I’ve seen so far shows it at 2.6. Was it an overclocked chip? On Tom’s Hardware, they say “In summary, we determined that the Core 2 Quadro requires the same amount of power in no-load operation that the Core 2 Extreme needs at full load.” Yeah, power consumption is getting crazy again. :-(
The Kentsfield was overclocked to 3.6 GHz which seems to be about where most air-cooled kentsfields will reach. Here is the link. Here is a link to a 4GHz Kentsfield but not stable. The chip is going to eat a lot of power meaning it will naturally run very hot and I doubt Intel is going to update its stock heatsink to deal with this issue so look for lots of OEM systems quietly baking away within their low air-flow cases ;)
The Kentsfield I have running at 2.66ghz with Crossfired 1900’s and 4 sticks of ram, Stays at 25c idling (95% of the time) and peaked at 41c during 3DMark06 using an medium level air cooled Zalman cooler.
Case and Cpu temperatures are very close to my P4 2.8ghz HT chip with a single gpu.