AMD 90nm Evaluated

muyuubyou writes "The Tech Report has measured the new 90nm A64 3500+ against its 130nm counterpart and a Pentium 4 3.6Ghz 90nm. AMD looks way ahead in the 90nm process especially when it comes to power consumption. Note these are consumptions for the entire system including GeForce 6800 GTs and hefty PSUs. RTFineShortArticle for more detail on the configuration. Leaving the PC on overnight is probably not a good idea with these new Pentium 4s."

Comments

wow

[hruntrung]

That's impressive. Of course, since it's total system wattage, it'd be nice to have some information about disk usage over the period of time, etc.

I like, though, that the 130nm Athlon 64 is still better than the 90nm P4. It might just be time to buy another desktop.

Re:wow

[Tumbleweed]

I'd wait until the Athlon 64 PCIe boards come out before buying a new system, so as to prolong useful system life. Nvidia's nForce4 chipset should be out in 4th quarter or thereabouts. _Then_ you can jump safely into a new system.

Power consumption

[TrollBridge]

"Leaving the PC on overnight is probably not a good idea with these new Pentium 4s."

It' snot going to make THAT much of a difference on your electric bill.

Now what I want to see is an analysis of the possible benefits to notebooks, specifically in extending battery life. Intel's Centrino seems to be doing fairly well in that department, but where is AMD's response?

Re:Power consumption

[aldoman]

Yes, actually it is.

Let's say your average 'gamer' system uses 500W of power, including monitor.

At 10c per KWh, that is going to be 5c/hour, or $37/month.

Re:Power consumption

[wherley]

say the P4 uses pretty grossly 100 watts more than the AMD - and you leave that on for 10 hours overnight each day for a year. say you pay about 7 cents a kilowatt-hour. then you end up paying .10 kilowatts * 10 hrs/day * 365 days/yr * 7 cents/kilowatt = ~ 25 bucks a year extra to run the P4.
say there are 100,000 P4 users doing this - there goes 2.5 million USD worth of electricity up in heat! :)

Re:Power consumption

[ivan256]

In fact the spike in power use from turning a system on in the morning is often higher then the amount of power it will draw all night while doing nothing.

That is complete bullshit.

Let's say your PC spikes to 500 watts for an entire minute before settling down to 50 watts. This would represent a worst case senario.

In this case, your couputer would use as much power in 10 minutes as it did powering up. Show me a modern PC, laptops included, that idle at less than 50 watts. Low end centrinos idle in the low 60s.

Re:Power consumption

[Minwee]

That's being pretty generous -- The power supply of that 'average gamer system' would have to be running at peak capacity 100% of the time to use that kind of power.

A large screen CRT monitor uses somewhere around 50-70W when active, and 1-2W in sleep mode. LCD displays use less power, but they're not what the average gamer uses.

Steady state usage for the computer itself is more like 200W than 500 -- The 500W capacity on your average gamer's power supply is equal parts peak capacity for boot-up and lies told by marketing, and you would have to be playing Doom 3 all day long, every day to keep that up for the entire month. Even if you disabled power management and just let it idle all night long it would still use less than 100W.

Using these numbers, and assuming that your average gamer is playing twelve hours a day and in class or sleeping the other twelve, we're looking at an average power consumption of 175W for a total of fourty-two cents per day or $13 a month at your rates.

The back of the envelope rests, your honour.

Re:Power consumption

[dougmc]

200 mWatt or so

Yes, it does use less power when running the HLT instruction, but not *that* much less. Half the power wouldn't surprise me. Modern cpus use over 50 watts of power -- I remember when the Pentium first came out, and people were amazed at it's 13 watts of power use. If only they knew!

Now, going into suspend or sleep mode, that saves a lot more power, but it can't do that thousands of times a second.

And in case people don't realize it, running things like Seti @Home or RC5 *do* cost them money. Their computer will probably use somewhat less energy if idle than it does when busy.

Also, it gets worse. Not only do you have to pay for the extra power used by your computer, but if you live somewhere hot, you'll have to pay for the extra air conditioning needed (after all, 200 watts of power used by your computer = 200 watts of heat generated.) Somebody told me that as a rule of thumb that 5x the amount of power used to generate the heat is needed to remove it via air conditioning -- so 200 watts of computer = 1000 watts of A/C needed to keep it cool. Can anybody confirm or deny this rule of thumb? -- it sounds like too much to me.)

Re:Power consumption

[Remlik]

Ok, I have a P4 3Ghz, ATI Radeon 9800, 3 HDD's (which do not sleep) and a 21" CRT monitor. The only power saving feature I use is putting the CRT into sleep mode after 15 minutes. Otherwise the computer and drives run full time.

My electricy bill is at times lower that $30 a month. No, I do not use the spread your payments out option.

Re:Power consumption

[Entropius]

"Low end centrinos idle in the low 60's".

That's incorrect, from my experience.

My mother's Athlon XP-M laptop idles around 25 watts. This is the complete power draw, counting the disks, LCD (on dim), and everything ... as reported by /proc/acpi. (acpi doesn't totally work in my Athlon 64 laptop, but from relative battery life I imagine it pulls around 30-32.)

Also, let's do some math. Its battery has a capacity of 4.4 amp-hours, and can get about 3 hours on a charge while idling.

The battery puts out 14.1 volts, so its total capacity should be 62.04 watt-hours. Thus, to idle for three hours, the computer has to use about 21 watts.

Now, this is a monster laptop: large disk, 15 inch widescreen, and a non-Centrino processor. The battery, natch, is pretty big as well. I think the whole thing clocks in at a little under eight pounds.

My friends' small iBooks get ~5 hours to a charge, and I bet they have lower-capacity batteries than the big M5312. They have, of course, smaller LCD's and efficient G4 processors. I wouldn't be surprised if they draw 10-15 watts.

Re:Power consumption

[default luser]

Actually, for CRTs, active power usage depends on the rate of the dot clock.

The power usage of a monitor will increase linearly with dot clock (with some minimum accounting for the brightness of the display).

Most high-end 19" monitors (with high-speed dot-clocks) have a maximum power usage of around 140w. Those numbers you have quoted are for THE HIGHEST supported resolution and refresh rate, with the maximum brightness...they vary because the maximum brightness and maximum dot clock speed vary among them.

On the other hand, most people use the recommended resolution and brightness set by the manufacturer. That is usually 1280x1024@85Hz on a 19" monitor, for a dot clock of around 111MHz.

For comparison, if you run your 19" monitor at 1600x1200@85Hz, you'll see a clock of 163MHz, and a proportionate increase in power usage.

For example, my monitor (Vision Master Pro 454) has a maximum rated output of 135w. If we ignore the brightness issue, then we assume that at maximum frequency (1920x1440@85), or 235MHz, the power usage is 135w.

So, scale down to a more reasonable resolution like 1600x1200, and we're only using ~ 93w. Or use the recommended resolution at 1280x1024, and we're sipping a cool ~ 63w.

Of course, these numbers are probably a bit higher due to components I have not taken into account. I do recall that the instruction booklet for my 454 lists power usage at multiple resolutions, and they did display this linear relationship, but I don't have access to it now.

Re:Power consumption

[iwadasn]

The biggest thing you can do for your electric bill is get rid of your incandescent bulbs, compared to that, nothing else comes close. Replacing a 60 watter with a 20 watt CF will net you 40 watts each, and you get about 1.5x the brightness.

your average house has something like 10 replaceable bulbs, so that's something like 400 watts, more than even a couple large computers.

How do we power these systems?

[aldoman]

With every computing consuming more and more power, its looking like we will need a wind turbine or solar PV array for anyone to run a decent sized network of computers at home.

Anyone currently doing this? I'm thinking of installing a turbine, but unsure of where to start out.

150 watts just to do nothing?

[JUSTONEMORELATTE]

Can someone elighten me on this? Is there a reason why the SpeedStep and other power-saving methods that are used in most laptops can't be adapted to desktop systems?
The old joke is that all CPUs sleep at the same speed, but after seeing the power consumption graph on this site, it's obvious that "power-hungry CPU" doesn't just mean high heat during gaming. This suckers are hungry even while doing nothing at all


--
Free gmail invites

Re:150 watts just to do nothing?

[News for nerds]

Cool'n'Quiet, baby.

Hopefully this will kill "AMD is hot" 'jokes'

[hattig]

Buy an Intel Prescott based system if you live in the Artic Circle ...

Looking at the data in the article, would I be mad in assuming that a 90nm 3500+ uses around 23W in idle mode?

Assuming power supply is 75% efficient:

112W * 0.75 = 84W getting to system
179W * 0.75 = 134W (130nm under load, near TDP of 89W, let's assume 84W)

134W - 84W = 58W Mobo, Gfx, IDE, etc power consumption

84W - 58W = 26W
26W * 0.9 (motherboard VRM efficiency) = 23W

I suppose that system power usage also drops in idle mode though as well.

Yes, these figures are extremely dodgy and vague and aren't worth much more than the speculation they are. It looks like the 3.4GHz P4 uses over 100W under load though - that is shockingly high.

Wow...

[T3kno]

The ambient temperature in his office was 85 degrees F? I'm breaking a sweat at 72F. When the A/C turns off in our office over the weekend the ambient climbs to about 85 and all of my servers fans are on overdrive. I wonder if that had anything to do with the power consumption in this test, I'm curious to see what the diference is at a more normal operating temperature, say 69 degrees F.

if intel were another company...

[etaluclac]

they'd bill this as a "feature." Buy the processor and we'll bundle the radiator for free. Remember, supplies are limited, so hurry before winter approaches.

Re:How about performance?

[eddy]

>How fast can the 90 nm Athlon 64 core go before it dies?

At least 3.6GHz. That's a 130nm(? I assume) Athlon FX-53, so the 90nms will do better.

Overclocking

[Sivar]

A friend of mine, an overclocking expert (inventor of the "Goldfinger devices" if anyone remembers those) said that the new shrunk cores overclock to around 3GHz if you can get your FSB high enough (though this won't be an issue with the FX chips, which aren't clock-locked).
To those paying attention, 2.2GHz in an Athlon64 can generally outperform a 3.4GHz PentiumIV, so this is a big deal.

Re:Electricity cost may be more/less than you thin

[NerveGas]

The one place where your figures aren't quite right is in the air conditioning department. An air conditioner, being a heat pump, just needs to move the heat from one spot to another, and the "typical" phase-change A/C unit is fairly efficient at it.

To put some figures on it, an air conditioner with an EER of 12 means that it can move 12,000 BTUs with 1000 watt-hours of electricity.

Now, 12,000 BTUs is equivalent to 3516 watt-hours of heat. So for every 3,516 watts of heat generation, you'll be expending 1,000 watts to move that heat to the outside of your building. And that's with an EER of 12, some units exist with EERs as high as 17.

So, for every 150 watts of power your computer is using, figure 40 to 60 watts for your A/C.

On the other hand, were you using a peltier device for cooling, you'd be in bad shape. If the EER figure were applied to them, it would be less than 1. For example, to move 30 watts of heat across a peltier, you'd need to apply approximately 45 watts of heat to it - meaning you'd be removing 30 watts from the cold side, but you'd need to remove 75(!) watts from the hot side.

steve