Confused about lowest-power AMD CPUs

[matt_garman]

I've got the ol' "building jones" and have started thinking about replacing the nforce2 ultra+athlon xp 1800 on my home fileserver. Of course, I don't actually need to replace anything; it's plenty powerful as-is. But if I can improve the processing capability to power draw ratio, then it's worth it (I also know people who could use the xp 1800; otherwise, see "building jones" ).

Note: this machine runs SETI@home 24x7, so, at least for me, loaded CPU power draw is most important.

Of course the Core2/Conroes look great, but I want to stay in the "value" category.

Many people have essentially the same post: "what's the current best low-cost-low-power-but-decent-capability CPU?" And most of the responses say socket 754 Semprons. Maybe I'm missing something, but it looks like most 754 Semprons have a TDP of about 65 W. But the latest (low MHz) 939 Athons have a TDP of about 35 W. So wouldn't, for example, an Venice core A64 3000 be an ideal processor for my type of application?

...Okay, I just read this thread, which says, in summary, that, at least with AMD processors, TPD is a theoretical max, typically for a whole CPU line. Still, my question remains, is there a nice ranking of typical power draw (preferably at load) for the current and recent AMD lineup? (i.e., S754, A939, SAM2, AAM2, etc)?

Finally, for what it's worth, the Sempron AM2 2800 plus the Gigabyte Tforce 6100 looks like a good starting point for my needs. Thoughts?

Thanks,
Matt

[floffe]

Check out the A64 TDP polls in this forum, and remember that those have plenty more cache than the Semprons, so the Sempron average should be a bit lower.

And also, the 35W AM2 ones aren't really low-cost either.

[jaganath]

Maybe I'm missing something, but it looks like most 754 Semprons have a TDP of about 65 W. But the latest (low MHz) 939 Athons have a TDP of about 35 W. So wouldn't, for example, an Venice core A64 3000 be an ideal processor for my type of application?

62W is the TDP for the fastest-clocked Semprons, the budget end of the range puts out maybe half that.

is there a nice ranking of typical power draw (preferably at load) for the current and recent AMD lineup? (i.e., S754, A939, SAM2, AAM2, etc)?

See SPCR's "Desktop CPU Power Survey" although IIRC no AM2 procs were tested, but power consumption has not dropped dramatically between S754/S939 and AM2.

the Sempron AM2 2800 plus the Gigabyte Tforce 6100 looks like a good starting point for my needs

You mean the Biostar Tforce 6100? Yeah, it's a good board, normally I would recommend a Sempron with CnQ but since it will be running at full load 24/7 it's not worthwhile.

[QuietOC]

Many people have essentially the same post: "what's the current best low-cost-low-power-but-decent-capability CPU?" And most of the responses say socket 754 Semprons. Maybe I'm missing something, but it looks like most 754 Semprons have a TDP of about 65 W. But the latest (low MHz) 939 Athons have a TDP of about 35 W. So wouldn't, for example, an Venice core A64 3000 be an ideal processor for my type of application?

Who knows where the 62W TDP number for the S754 Semprons came from.

I have the 67W 2.4GHz @ 1.4V "E3" revision S754 Athlon 64 3400+. The 2.2GHz @ 1.4V "E6" Athlon 64 3200+ is specified at 59W and the 2.0G @ 1.4V "E6" Athlon 64 3000+ at 51W.

The highest clocked S754 Sempron (the 2.0GHz 3400+) will use less power than the Athlon 64 3000+ at its stock 1.4V voltage.

The big thing is the slower default clocks of the S754 Semprons allow a lot of undervolting.

My 1.4GHz "E6" revision Sempron 2500+ ran great off 0.88V. Power usage/heat generation reduces more than the square of the voltage difference. So, basically the 2500+ runs at about 5W at full load.

None of the "E" revision C'n'Q enabled AMD chips allow undervolting under 1.1V. There is a large power difference between 1.1V and 0.88V and a huge difference between 0.88V and 1.40V. The non C'n'Q "E6" Semprons (2500+/2600+/2800+) can undervolt below 1.1V, but only if the motherboard BIOS has undervolting options (like the Biostar Tforce 6100).

The ability to undervolt below 1.1V is enough for many SPCR folks to get an older "D0" C'n'Q Sempron 3000+ or higher. Those can be undervolted/underclocked on any C'n'Q motherboard down as low as 800MHz @ 0.800~0.900V, but they generally need at least 1.250V to run at their default clockspeed of 1.8GHz (or faster).

The S754 platform has a whole lot less circuit traces and passive loads than S939 or AM2, and very little performance is lost with single channel DDR. Thus S754 + "E6" non-C'n'Q Sempron is the power performance leader.

You may be able to get close (power wise) with a AM2 Sempron, but it will be more difficult (higher default CPU clockspeed, two DDR2 channels to power). AMD does seem to have slightly improved the performance/watt in the new AM2 "F2" revision.

From what I have seen total power (motherboard + CPU) has slightly risen from S939 to AM2 (using the same chipset). People were misled by reviews comparing nForce4 S939 boards to nForce5 AM2 boards--while either chipset will work with any socket. If you value low power you will get neither of those chipsets, however. The 90nm 6100/410 or one of the ATI chipsets are much cooler. Also no motherboard extra feature is free power-wise, which is why a relatively simple board like Tforce 6100 is to be desired. Even further power reduction is possible with a completely simplified barebone like the ASUS Pundit P2-AE2.

[mb2]

TDP may aswell be witchcraft, they just don't seem to consistantly make sense. especially when u factor in the supposed rules of underclocking and undervolting (why do u say more than the square of the difference QuietOC?).
Eg, i have just got a via C3, its TDP is 12W. it stays in the 50s when priming for long periods of time, with no airflow around it what-so-ever (out of the box), and on a tiny s370 (i assume stock) heatsink, ambient in the low-mid 20s.
So going by that, your sempron should be easily cooled out of the box, with no airflow around it atall, with the much larger stock sempron HS with more surface area, when stressed for as long as u like. and with temps less than mid 50s? is this the case quietOC; if so, i think i'll go buy a sempron without CnQ!

as for the OP;
there are few factors affecting k8 power; the process (thickness?), cache, memory controller (DC presumably uses more power), whether it has a Vcore Lock, clock speed, Vcore and revision (newer tending to be more power efficient). (and quality/section of the wafer, which u can't control- unles u want to pay more for a 35wer, or deal with an A64-m or turion).

So getting the lowest power of all those your looking at a e6 s754 sempron 2600+, with a board that can undervolt below 1v.
i'm not sure about the AM2 semprons, whether they have a Vcore lock or not, or if they use more power or not; but using DDR2 would reduce overall system power draw.

[QuietOC]

why do u say more than the square of the difference QuietOC?

Because that is what the data shows.

Physics: Semi-conductors are not linear devices. Parisitic losses (gate leakage current, etc.) increase with temperature. Die temperature increases linearly with power output. If you could keep the die a constant temperature power would increase proportional to V squared, but in normal use (using constant temperature room air as a heat sink) it is actually closer to V cubed.

Get those CPU voltages (currents, really) down!

Eg, i have just got a via C3, its TDP is 12W. it stays in the 50s when priming for long periods of time, with no airflow around it what-so-ever (out of the box), and on a tiny s370 (i assume stock) heatsink, ambient in the low-mid 20s.
So going by that, your sempron should be easily cooled out of the box, with no airflow around it atall, with the much larger stock sempron HS with more surface area, when stressed for as long as u like. and with temps less than mid 50s?

Yes, The undervolted 1.4GHz Sempron 2500+ was much cooler than my 800MHz VIA CPU (and of course performs much, much better). I should have tried running it without a heatsink. It pretty much reported ambient temperature during Prime95 running passively. The GeForce 6100 Northbridge chip was much, much hotter than the CPU.

It is possible to buy C'n'Q enabled AMD processors without the 1.1V lock. The 35W X2 3800+ can undervolt well below that--also the mobile CPUs. I am not sure if the 35W AM2 Semprons (specified at 1.2V) can.

[matt_garman]

Edit: did some more research, answered some questions.

My 1.4GHz "E6" revision Sempron 2500+ ran great off 0.88V. Power usage/heat generation reduces more than the square of the voltage difference. So, basically the 2500+ runs at about 5W at full load.

That was for a S754 Sempron, right? Yes

Was it a 64-bit enabled Sempron? Not sure, but assuming yes.

Was that extreme 0.88 undervolt at the stock frequency (i.e. 1.4GHz)? Yes

What motherboard? Biostar Tforce6100 (I think).

How would the performance of a S754 Sempron 2500 compare to an XP 1800? (Better, assuming AMD's ratings are accurate across chip lines, I suppose.)

On another note: why doesn't anyone make a motherboard with integrated graphics on a full-sized ATX board? All the nforce6100 boards are uATX. I do have a Matrox Millenium 1 PCI graphics card (this computer is headless except in times of maintenance)... but now I'm back to hotter-running chipsets, like nforce4, etc.

[hammer]

There is an atx motherboard with integrated graphics:

Gigabyte GA-M55plus-S3G

The bios supports cpu voltage adjust.

(First post, figured I'd used the forum for reference a lot lately...)

[jackylman]

There is an atx motherboard with integrated graphics:
Gigabyte GA-M55plus-S3G
The bios supports cpu voltage adjust.

Does it adjust up AND down?

[jaganath]

Yes, Vcore adjustable from 0.8-1.55V.

GA-M55plus-S3G Manual

[matt_garman]

Hmmm... that S754 Biostar Tforce 6100 has tons of praise.

They also have a socket AM2 version.

As far as I can tell, the S754 and AM2 versions of that board are pretty much the same.

What I haven't been able to discern is undervoltability of the AM2 board.

Has anyone had a chance to test this yet? I'm this -> <- close to building a Tforce 6100 system. Obviously the S754 gets me the lowest power draw. But how low can I go with AM2?

Thanks!
Matt

Edit: changed "underclockability" to "undervoltability".

[QuietOC]

What I haven't been able to discern is underclockability of the AM2 board.

Has anyone had a chance to test this yet? I'm this -> <- close to building a Tforce 6100 system. Obviously the S754 gets me the lowest power draw. But how low can I go with AM2?

FWIW: The S754 Tforce 6100 is not underclockable. The lowest CPU HT clock is the default 200MHz. You can underclock C'n'Q chips with it by selecting lower multipliers, but that isn't very useful for undervolting because the CPU still has to boot at the default multiplier.

[matt_garman]

What I haven't been able to discern is underclockability of the AM2 board.

Has anyone had a chance to test this yet? I'm this -> <- close to building a Tforce 6100 system. Obviously the S754 gets me the lowest power draw. But how low can I go with AM2?

FWIW: The S754 Tforce 6100 is not underclockable. The lowest CPU HT clock is the default 200MHz. You can underclock C'n'Q chips with it by selecting lower multipliers, but that isn't very useful for undervolting because the CPU still has to boot at the default multiplier.

D'oh! I meant to use the term undervoltable. Lemme go edit that post...

MG

[matt_garman]

So... bump.

Any word on the power draw and under-clockability of a S754 versus AM2 Semprons?

Right now I'm leaning towards AM2, if only for future compatibility. Surely AMD will continue the Sempron line when they have a 65nm process. I'm (wishfully?) thinking that in the future, I can get a low-end 65nm Sempron for AM2 and quickly make up any extra power draw that current AM2s have over S754s.

Does anyone have experience with the Biostar Tforce 6100 for AM2? Any reason to get that over the Gigabyte GA-M55plus-S3G? The Gigabyte has more features, and someone already pointed out that it should have the same undervolting capabilities as the Tforce. Any thoughts on Biostar versus Gigabyte quality?

Thanks!
Matt

[jaganath]

Any word on the power draw and under-clockability of a S754 versus AM2 Semprons?

IMHO they're probably near identical in this respect, with the AM2 chips maybe being a little cooler. One thing that's not yet clear is whether AM2 Semprons have the 1.1V Vcore lock which was applied to E-stepping S754 Semprons.

Surely AMD will continue the Sempron line when they have a 65nm process.

Looks like it:

http://www.xbitlabs.com/news/cpu/displa ... 32709.html

According to a report from an Asian web-site, AMD Sempron processors will acquire second core as early as in the Q3 2007, a year from now.

“AMD’s entry-level processors will all feature dual-core architecture by the third quarter of 2007, when its last single-core CPU – the Athlon 64 – will start to be phased out in the market,â€

[QuietOC]

Does anyone have experience with the Biostar Tforce 6100 for AM2? Any reason to get that over the Gigabyte GA-M55plus-S3G? The Gigabyte has more features, and someone already pointed out that it should have the same undervolting capabilities as the Tforce. Any thoughts on Biostar versus Gigabyte quality?

You might want to look at what components each board uses. Gigabyte tends to use junky components. Their past boards are known for having poor quality--lots of RMAs. I am not saying that board is that way, but it is likely to be that way.

Biostars in general have been well made for their price--the Tforce line in particular. My experience is that they are better quality than boards from ASUS.

Most people here would rather let their system run a little warm rather than increase noise. Higher temperatures will kill cheap components faster. And also remember more features = more power used.

[matt_garman]

You might want to look at what components each board uses. Gigabyte tends to use junky components.

Yeah, that's kind of what I was getting at. What components are good, and which ones are junky? How do I tell which kinds of components a certain mobo uses (short of buying one and personally inspecting it)? Do you generally rely on reviews to help you determine if a board uses quality components, or some other, more reliable method?

(I vaguely remember, on another thread, you mentioning that the Tforce boards have good capacitors on them, but I could be wrong.)

Their past boards are known for having poor quality--lots of RMAs. I am not saying that board is that way, but it is likely to be that way.

Now that you mention it, I recall having a Gigabyte board with some weird grounding problem. Couldn't figure it out for the life of me!

Biostars in general have been well made for their price. The Tforce line in particular. My experience is that they are better than boards from ASUS.

At least judging by price, the Tforce series seems to be their higher-end line.

I read too many random posts all over the net, and have all this "knowledge"... it all jumbles together, and I can't remember verifiable fact from conjecture and speculation. Like with Asus, I could have swore I read a post (not on SPCR) from someone claiming to have worked for them. He said that the "Premium" line of boards are the ones they build with nothing but the best components. I think next in line was the "Deluxe" and at the bottom was the "X" line, which was effectively the el-cheapo stuff.

I probably shouldn't even post that because it's based on foggy memory of a non-verified post! Now someone else will be just as confused as me!

And remember more features = more power used.

Yup... since thinking about this, I think I'm actually going to try and leave the server as-is, and actually downgrade my power rig (way overkill for my needs). So the Biostar it is! I've got a Kill-A-Watt on the way, and am on a new mission to reduce my overall power usage in my house.

Thanks for all the help and feedback guys!

Matt

[matt_garman]

Any word on the power draw and under-clockability of a S754 versus AM2 Semprons?

IMHO they're probably near identical in this respect, with the AM2 chips maybe being a little cooler. One thing that's not yet clear is whether AM2 Semprons have the 1.1V Vcore lock which was applied to E-stepping S754 Semprons.

Hmmm, interesting. I'm suprised noone around here has tried that yet. Well, I may be the first. I'm going to review my financial situation, and hopefully build a low power system.

Thanks again for all your help!

Matt

[QuietOC]

Yeah, that's kind of what I was getting at. What components are good, and which ones are junky? How do I tell which kinds of components a certain mobo uses (short of buying one and personally inspecting it)? Do you generally rely on reviews to help you determine if a board uses quality components, or some other, more reliable method?

Website reviewers generally can't do long-term reliablity testing. They do some limited short-term reliability test (which many boards used to fail!)

A few sites are now noting things like capacitor quality since bad capacitors were a big problem during the K7/Netburst era. Biostar is not the only company using and promoting better quality components.

Like a lot of computer components, overclocking results are generally good indicators of quality (overbuilt) parts. For example my Tforce 6100 seems to be capable of running its CPU HT speed faster than any other K8 board period--though I was a little disappointed I couldn't get the Sempron 2500+ to boot at 2.8GHz (7 x 400 MHz).

[jaganath]

(I vaguely remember, on another thread, you mentioning that the Tforce boards have good capacitors on them, but I could be wrong.)

This is true, the Tforce 6100 uses solid-state caps in the power regulation circuit, whereas the Gigabyte board appears to use wet electrolytics (cheaper, but more likely to fail prematurely if not cooled sufficiently).

[vg30et]

I recently made the decision to purchase the Gigabyte GA-M55plus-S3G over the Biostar Tforce 6100 for my AM2 A64 3200+. In the end, the additional features of the Gigabyte were more important to me than the supposed additional reliability of the Biostar.

Even with all the features and some posts indicating Gigabyte's power hungry tendencies, I am quite happy with my power consumption numbers. Using kill-a-watt and rmclock to undervolt in windows, I'm getting 52W idle @ .85v & 1ghz, 77W cpuburn @ 1.2v & 2ghz.

[matt_garman]

Even with all the features and some posts indicating Gigabyte's power hungry tendencies, I am quite happy with my power consumption numbers. Using kill-a-watt and rmclock to undervolt in windows, I'm getting 52W idle @ .85v & 1ghz, 77W cpuburn @ 1.2v & 2ghz.

What are the specs on the rest of your rig?

[QuietOC]

I recently made the decision to purchase the Gigabyte GA-M55plus-S3G over the Biostar Tforce 6100 for my AM2 A64 3200+. In the end, the additional features of the Gigabyte were more important to me than the supposed additional reliability of the Biostar.

FWIW: One of the Newegg user reviews of the AM2 TForce 6100 did note that the board runs the DDR2 at 2.2V. The S754 Tforce 6100 only allows a few DDR voltages with 2.6V being the lowest. 2.6V vs. the normal 2.5V DDR voltage is not a big difference, but 2.2V is a considerable overvoltage from the 1.8V DDR2 specification, and a large (>V^2) increase in power/heat.

[vg30et]

What are the specs on the rest of your rig?

P180
Seasonic S12-380w
3 nexus case fans undervolted to ~5v using zalman fan controller
Gigabyte GA-M55plus-S3G
AM2 A64 3200+ (Orleans core - ADA3200CNBOX)
Scythe infinity w/ undervolted nexus 120mm
2x 512mb DDR2
2x Seagata 7200.10 250gb SATA2 in RAID 1 config

System is a 24x7 web/mail/file/rdp server with just a power and lan cable hooked up to it. Onboard audio and all other non essentials disabled from bios.

TCaseMax v1.19 does not recognize the cpu core correctly but reports a 49C tcasemax, no TDP reading. Not sure if this is accurate but it if is, I believe it corresponds to a ~35w TDP.

For comparison sakes, out of the box numbers using AMD Power Monitor:
Windows Idle (1.1v @ 1ghz): 62w
Windows w/ CPUburn (1.4v @ 2ghz): 96w

[QuietOC]

Using kill-a-watt and rmclock to undervolt in windows, I'm getting 52W idle @ .85v & 1ghz, 77W cpuburn @ 1.2v & 2ghz.

Did you verify those CPU voltages (specifically the 0.85V setting)?

Please report what the hardware monitoring programs report for those settings. It should be a range of voltages, and probably not exactly 0.85V.

Do the normal "F2" stepping AM2 C'n'Q processors not have the 1.1V lock then?

If so then that's a good reason to go AM2.

Thanks!

[vg30et]

Did you verify those CPU voltages (specifically the 0.85V setting)?

RMClock and AMD's power monitor software confirm CPU voltage dropping from 1.1v to .85v. At the same time, CPU core temps dropped a couple degrees and kill-a-watt displayed a reduction in power consumption.

RMClock actually allowed voltage adjustment lower than .85V but anything below .85V was unstable (Prime95 errors) and anything below 1.0ghz/.80v led to an immediate lockup.

[QuietOC]

Did you verify those CPU voltages (specifically the 0.85V setting)?

RMClock and AMD's power monitor software confirm CPU voltage dropping from 1.1v to .85v. At the same time, CPU core temps dropped a couple degrees and kill-a-watt displayed a reduction in power consumption.

I know RMClock does not report actual voltages (it merely shows the setting.) I am pretty sure the AMD utility is similar. Use Speedfan/MBM/Sandra--anything that will read the actual information from the hardware monitoring chip. (It is questionable how accurate this is, but few people are willing to stick a voltmeter on their CPU socket.)

The drop from 1.1V to 0.85V should be good for much more than a couple of degrees. It should reduce CPU power usage about in half.

[vg30et]

QuietOC: You're right, Speedfan and CPU-Z report the voltage as .864V, a little higher than rmclock's .85 setting.

CPUZ:

Speedfan:

TCaseMax:


I retested the temps and noticed little to no difference between 1.1v and .85v at idle despite the 3-4W reported savings by killawatt. I believe this is because CPU temps are already quite low - ~8C above ambient.

[corevalue]

Hi Matt, Im a Newbie here, but I got an amost similar dilemna to you. I need to build a PC for use as home entertainment centre, and to do "real stuff", like CAD occasionally. Snag is, it's going on a boat where power is SEVERELY limited. So it's got to be capable, but must be (a) quiet and (b) frugal.

My take on TDP (as an ex-electronics engineer) is that the figure is a guide to the guy designing the mobo - this is the worst case dissipation @ flank speed at the specified vcore/freq. In practice, no processor is expected to reach it. I used to have to proffer up similar figures to the mechanical design guys in my past - usually rendered with a grunt and oh, I dunno. Not more than XXX Watts, for sure. These days, the simulator used during the chip design will spit out a fairly accurate number, but that has to be weighted with how much work the processor will be expected to do, in it's worst environment, and the quality of the silicon. So AMD take the worst case for those figures, add 20% margin, and hey presto, a TDP, which is probably 100% higher than you would ever see a chip really dissipate whilst running the sieve of eratosthenes.

AMD now seem to be testing their part for efficiency, and binning them accordingly. I note, having tried to buy one, that the 35W TDP (part numbers start with ADD for the Athlon, or SDD for the Sempron) are as rare as rocking horse sh*t, and twice the price of the "normal" ADO ~70W TDP parts. Presumably, at the moment, the fab hasn't slickened the process well enough to get a good yield of low-leakage parts. Time will tell.

I originally thought to get a 35W 3200MHz Sempron - various benchmarks seem to indicate that it's only just behind the Athlon in performance now, but confusion about CnQ only applying to certain parts, and discovering I could get a similar 35W 3500MHz Athlon for less changed my mind. However, I can't get either, nobody seems to stock them. How the hell do AMD expect to sell them at 100% premium over the "normal" chips?

The board I intend to use, after much searching, is the Abit NF-M2 nView, with on board nVidia 6150 graphics, and according to the manual, everything is adjustable. In this, I'll stick a cheap CnQ capable Sempron, probably the SDA3200CNBOX. The main TV source is a Technotrend Skystar 1 DVB-S satellite TV card, this has it's own MPEG decoder, using it in my Athlon Thunderbird at 800MHz, the processor loafs at 10% usage!

Incidentally, to all you green computing lovers. Ever looked at the standby power of your computer? You'll probably be shocked.

[jaganath]

I originally thought to get a 35W 3200MHz Sempron ...I could get a similar 35W 3500MHz Athlon

The "Performance Rating" (3200+ and 3500+) doesn't actually refer to the clock speed; it's a relative benchmark based on some reference proc, can't remember which. So the Semp 3200+ runs @ 1.8Ghz and the A64 3500+ runs at 2.2Ghz.

How the hell do AMD expect to sell them at 100% premium over the "normal" chips?

As usual, green consumers are forced to pay a premium for environmentally friendly products; also, presumably they reckon the higher purchase price will be offset by the savings in electricity costs. Of course the reality is that the normal parts can be got down to EE levels pretty easily by undervolting.

The board I intend to use, after much searching, is the Abit NF-M2 nView

There is a thread devoted to this board in this forum; apparently it doesn't allow you to undervolt lower than 1.35V in BIOS, which is bad for power consumption; but CnQ + RMClock/CrystalCPUID should enable you to undervolt by software.