I'm looking to upgrade an Athlon XP+1700 (desktop PC) and Duron 800 (HTPC pc w/2+ tuners) with modest CPU power upgrades (both have enough power for what they are used for, but faster is always nice) and something that will bring idle power consumption down as low as possible.

Right now each of those systems uses about 90w at idle which increases to 110w under full CPU load (using mprime). I'm aiming to get idle power below 40w. Full CPU power isn't too crucial since the amount of time spent at full power is very low compared to the time spent idle.

So here's my considerations:

Athlon 64 3000+ Venice Socket 939 - Would consider socket AM2 but heard it uses more power and then I can't reuse my DDR.

Sempron 64 3000+ Palermo Socket 754 - Is idle power under CnQ going to be significantly lower than the A64? Would also consider the 2600+ S64, but from what I understand any Sempron 64 under 3000+ doesn't support CnQ. Does that include the Sempron 64 2800+ Manila AM2?

AMD Geode NX - Seems like a good low power option, but again, will idle power draw be significantly lower than a A64? Has limited MB options and MBs seem to come with onboard graphics meaning it would probably be better for low power servers. CPU power should be slightly better than the Duron 800 but probably worse than my XP1700+ and my Duron replacement should have at least 3 PCI slots (HTPC computer w/2+ tuners) which rules this out.

VIA C3/Epia - Too expensive compared to Geode and can only get in MiniITX w/1 PCI slot, though I have one w/a 800MHz for a server that only draws like 26-7w at idle if I remember correctly so I know that has low idle power. But CPU is slow.

Any other ideas? I'd consider an Intel solution but I'm a bit of an AMD fanboy so the Intel solution would have to be considerably better or cheaper for me to consider it.

I'm leaning towards the A64 simply because I don't think that idle power will be more than a couple watts different than the Sempron but this is mostly a guess, I would love to see some actual numbers!

-Dave

Best Performance/Power:

E6 S754 Sempron 2500+
-needs motherboard w/ undervolting in BIOS

My motherboard (TForce 6100) let me undervolt down to 0.88V and no lower, and the Sempron runs just fine at default speed (1.4GHz) at 0.88V.


Alterative for C'n'Q enable motherboards:

"D0" Sempron 3000+ or higher (StarMicro has 64-bit versions)
-use RMClock or CyrstalCPUID to roll your own C'n'Q settings

Possibly saves a few Watts at idle if you can get it down to 0.850V at 800MHz compared to the E6 Sempron 2500+ running all the time at 1.4GHz, but will use considerable more power at full load (1.8GHz @ ~1.275V or more).

Really no need for C'n'Q unless you need more than a 1.4Ghz K8. A 1.4GHz K8 is easily twice as fast as the fastest VIA C3 while using less power.

AM2 as a platform uses more power than S939 which in turn uses more power than S754, while giving very little additional performance (~5% better). Chipset choice is even more important, however.

PS: Of course the TForce 6100 can actually run that Sempron 2500+ pretty fast. The CPU HT link gets flaky somewhere north of 360MHz, but 2.5GHz is doable. Power use will suffer.

For some reason I read elsewhere that Nvidia chipsets seem to be power hungry. The motherboards I was looking at use the ULi M1689 chipset, not sure how this chipset does in terms of power consumption, anyone know?

While I certainly don't necessarily "need" more power than a 1.4GHz Sempron, having the option without having to do anything is always nice which is the main benefit of C'n'Q for me. I'm not worried about power consumption at full speed as time spent at full speed is normally so low compared to the amount of time spent at idle.

Any idea on how much different power consumption is between 754, 939 and AM2? 2w, 5w, 10w? Varies depending on chipset?

Thanks for the feedback!

Not a lot of data on that. I'd guess less than 10W.

All the K8 sockets (940/754/939/AM2/etc.) can use the exact same chipsets. The GeForce 6100 seems to be the coolest running nVidia chipset, and it is just as fast as the nForce4/5, and also the best overclocking chipset. VIA/ULI/ATI all seem to have cool runnning chipsets. Motherboard makers can choose what voltage to run the chipset at so power use isn't consistent anyway.

S754 is the most efficient since it has the least number of traces--for example you only use one big DIMM instead of two. Most people have figured out that SLI isn't worth the extra power usage, but yet still want dual-channel DDR/DDR2 even though the same power/performance applies there.

The only thing that needs tons of memory bandwidth is integrated 3D graphics. Otherwise single channel DDR400 is not a performance limitation.

As far as 1.4GHz K8. You need an additional 1GHz or so to even notice any performance difference. I wouldn't bother with the differences between 1.4GHz and 2.0GHz K8's. At most it is a 40% performance difference, and generally much less. For general use 1.4GHz K8 is very snappy.

The voltage required for stability increases pretty linearly with clockspeed, and power increases faster than V squared. So basically power increases in propotion with clockspeed squared at minimum stable voltages. That pretty much means you want the lowest usable clockspeed for power efficiency.

The current K8 C'n'Q chips lock out core voltages under 1.1V which keeps them from having the best performance/power. The AM2 Semprons might be fairly good. The AM2 Sempron 3000+ does NOT have C'n'Q.

The Tforce 6100 lets you store sets of BIOS options, so you can have a 1.4GHz @ 0.88V option saved, and say a 2.1GHz @ 1.1V for when you want a little more power. Just requries are reboot. Not as convinient as C'n'Q, but usable.

Can anyone clear this up once and for all? What is the status of AM2 Semprons vis-a-vis CnQ? Anyone have an official AMD line on this, or even better have one of the aforementioned chips?

Thanks again for the feedback. So it looks like if I don't want to mess with undervolting, I want to stick with socket 754 and something that supports CnQ, GeForce 6100 or other brand chipset.

But the Athlon 64 90nm chips aren't available in socket 754, so if I want the performance of Athlon 64 I have to go to socket 939.

So the last question is? What the difference in power requirements between Sempron 64 3000+ Palermo and Athlon 64 3000+ Venice?

The Sempron 64 3000+ Palermo is:
1.8GHz, 1.4v, 64KB+64KB L1, 128KB L2, 1600MHz HT

The Athlon 64 3000+ Venice is:
1.8GHz, 1.35/1.4V, 64KB+64KB, 512KB L2, 2000MHz HT

Close to the same specs, but the L2 cache is much bigger an the HyperTransport runs faster on the Athlon. How can I tell what voltage CPU I'm getting when getting a boxed CPU? It looks like the Sempron is stepping E3 and the Athlon is stepping E6 looking at AMDCompare.com

AMDCompare.com also lists some interesting 35w Sempron and Athlon 64 processors as well which I can't find...

On the K8 architecture, all cache really does it make the processors hotter and more expensive. Considering the performance you gain from a larger cache (none for most apps, a few percent for gaming and a few others), it really isn't worth it IMO. Sempron64 sounds like it would be more than enough for what you plan to do with the system.

But if you really want Athlon64, 90nm Venice chips are available on S754. You can find Athlon64 3400+ Venice (2.4GHz) for $99 shipped at StarMicro. 3000+ (2.0GHz) and 3200+ (2.2GHz) flavors are also available.

If it was my money, I'd get a Sempron64 3000+ and 6100-based motherboard. Only real downside I think to S754 is there's no upgrade path for dual-core, but I have a feeling this isn't an issue for you.

Nope, dual-core isn't an issue for these machines. Thanks for the advice, Sempron 64 3000+ will be ordered shortly.

-Dave

For the lowest possible idle power consumption you’d want a Sempron as they can run as low as 800 MHz, whereas Athlon 64’s are typically limited to 1 GHz. If you get a Sempron that allows you to undervolt below 1.1V using CrystalCPUID, then even better. I’m not sure if the E3 revision allows this, the Dx (not sure what x is, possibly 0) does though.

I’d keep away from an NF3 chipset motherboard as the one I tested had a high power consumption for such a modest board.

Yeah that's why I grabbed a D0 Sempron, my board doesn't have BIOS Vcore/multi control, so I needed full adjustability via software. But Ex seem to undervolt better, and have SSE3 and 64-bit. So as long as your board has good BIOS options, I'd go with a Sempron64.

Well, I ordered this Sempron 64 (Rev E6): http://www.amdcompare.com/us-en/desktop ... 3000AIO2BX

Along with the Asus K8U-X motherboard (ULI M1689 chipset): http://usa.asus.com/products4.aspx?mode ... =14&l3=200

Both from NewEgg. Should have it Tuesday, we'll see how it compares to my XP 1700+ and if it does well, I'll get another to replace my Duron 800.

I've just picked up a DFI RS482 and a second-hand Venice 3000 (E3). They should make a nice upgrade to my fileserver, and the board is compatible with Mac OSX too. Previous server was a Tualatin Celeron 1.2, but since it has taken up TV encoding duties it was just too slow!

I'd have gone with Conroe if not for the expense, and with AM2 if not for the requirement to buy new RAM.

The actual voltage and HT speed are both functions of the chipset NOT the processor. The newer chipsets/chips (like the TForce 6100 uses) support the 5x multiplier (5x200MHz = 1000MHz) HT speeds on S754. Also some motherboard will undervolt by default. For example my TForce 6100 sometimes undervolts as much as 0.1V--which is good for keeping things cool--not great for overclocking.

All this makes you think about how much the Athlon X2 3600+ with Sempron style chache will undervolt...

And as it will be an Athlon it will support CnQ

This link shows the X2 3500+ as being a 35W EE part and costing $147. It may be a typo and it's really a 65W part, otherwise it's gonna be a big favourite around these here parts.
Although there's speculation that it will be an OEM part only, which would be a shame.

I got 2 Sempron 3000+s have them installed and am satisfied with the power savings and additional performance, but of course, I'd like more, but I don't think I'll be able to get much more without another significant cash outlay.

My Desktop had these components:
Athlon XP1700+, ATI Radeon 9600, Seagate 7200.7 80gb drive, MS-6764 motherboard, Windows XP.

Idle: 67w
Full CPU: 84w (using mprime)

Swapped in Sempron 3000+ mentioned above with Asus K8U-X MB.

Idle no CnQ: 60w
Idle w/CnQ: 47w
Full CPU power: Didn't write it down, but it was about 30w higher.

My MythTV system:
Duron 800, GeForce FX 5200, 2 Hauppauge PVR-250s, Seagate 7200.9 250gb drive, MS-6738 motherboard, Fedora Core 5

Idle: ~100w
Full CPU: 111w

Swapped in Sempron 3000+, MSI K8MM-V motherboard

Idle w/CnQ: 74w
Full CPU: 105w

So I saved 20w with my desktop and 25w with my MythTV system while significantly upgrading CPU speed.

I was hoping for more, but I think I need to use onboard video to save more power with my desktop. If I could undervolt the CPUs more than the default 1.1v @ 1GHz in CnQ idle mode, that would help, too.

Another area I saved with my desktop is in sleep/standby power util goes to just a watt or 2, where the old XP1700+ setup still used 37w when sleeping.

I also took the time to see how much power each component draws using the MythTV system starting from just the MB using the onboard video.

FX5200: +14w
PVR250: +12-13w
HD: +7w

If I could get a VGA->RCA or S-video converter I might be able to ditch the video card. I also wonder if the PVR500 (2-tuners in one) uses less power than 2-PVR250s. If I can get the HD to spin down when idle, that will help, too.

You didn’t mention whether you were under-volting! That’s a cheap way to drop the power consumption
If you aren’t, then CrystalCPUID or RMClock will help to reduce consumption at load quite a bit.

Yes, definitely undervolt those bad boys. On the Pundit in my sig, undervolting reduced idle from the wall by ~4w and load by ~11w.

Well, one machine is a Linux (Fedora Core 5) machine, MSI K8MM-V motherboard, I didn't see any undervolting options in bios, and not sure how to undervolt from Linux.

The other is Windows XP with Asus K8U-X but RMClock doesn't let me run it any lower than 1.1v at 800MHz. I'll try CrystalCPUID when I get home later. Is there a way to run at those undervolted levels when the software isn't running? RMClock only seems to work when it's running.

What tools you use to ensure that it's stable when undervolted, is mprime good enough? It seems like you'd want a tool that would let you excersise CnQ by varying CPU load periodically as well.

The software is actually doing the under-volting itself, it needs to be running permanently.
1.1V is the lowest voltage that you can reach on most if not all recent AMD chips. I don’t know about AM2 CPUs and the EE chips are a special case and some are lower than 1.1V at stock even at full load.

I use Prime95, one instance per core, for 12 hours or more. I also use Memtest if I have any doubts about my RAM.
I don’t know of any utilities that specifically test CnQ. If you choose your custom CnQ settings wisely, I can’t see there’s anything to worry about.

Unfortunately, as mentioned earlier, the revision E Athlon64 and Sempron64 do not allow Vcore <1.1V via software. This means you can't do much to reduce power consumption at idle, but undervolting the CPU at CnQ load can reduce power consumption a good deal. For example, my Sempron is able to undervolt to 1.225V at 1.8GHz. It's an older D0 revision -- your newer CPU should be able to undervolt even further.

Here's a thread with some information on CnQ undervolting in Linux.

Link

Thanks for all the info, guys. I'll play with undervolting full-load speeds when I get home.

Hi Frosted.

I noticed in your sig you are using a 40G Fujitsu 2.5" HD.
I have an 80G Fujitsu 2.5" that I'm planning to use in my next system.

Can you please tell me the speed of yours? (4200, 5400 rpm)
I got a 4200 for noise considerations.

Thanks!

[quote="drees"]I got 2 Sempron 3000+s have them installed and am satisfied with the power savings and additional performance, but of course, I'd like more, but I don't think I'll be able to get much more without another significant cash outlay. [/quote="drees"]

Thanks for the report drees.
I'm considering a Sempron 3000+ system too, as was recommended.

You mentioned power/performance but haven't said anything about noise yet.

To me this is the most important item. I know that power/perf. are directly related.

Can you tell us what your old systems were like noise-wise and what your new systems are like? Anything subjective or objective measurements would be appreciated.
Can you hear your new system?

I think Anandtech has power consumption figures for dual tuner TV cards, and IIRC double tuner cards draw ~ twice that of single cards with onboard encoder/ decoders. One option for the HTPC is, seeing's though all the TV stuff is unloaded off the CPU, to underclock the CPU and undervolt it as much as you can. Shame you can't undervolt more though. Or, replace the PVR250s with simpler cards that don't have onboard encoders - that should reduce the idle figure at the risk of increasing the load power draw, aside from any other consideration.

What PSUs are you running? Something like the PicoPSU should easily run either of your rigs, particularly if you drop the discrete graphics. A separate small 2.5in HDD for the HTPC operating system/ programs would be an idea, allowing you to spin down the larger HDD.

Starting to get into the realm of ever decreasing returns though - but hey, what else are we here for?

Noise is pretty low, but yes, I can hear it. Really the only noise maker difference is the CPU fan, the OEM fan provided by AMD is actually pretty good, especially with the BIOS lowering fan speed based on CPU temps. I think that if you had a cooler with a big 120mm fan you could keep it nearly inaudible. The fans seem quieter when spun down than my old fans. It takes some time of full CPU power to heat up the CPU enough to spin them faster, which doesn't normally happen for me.

The other big noisemakers in my system are the HDs (mounted in ZALMAN ZM-2HC2) and the PSU. The PSU is a Seasonic Super Tornado 300 which also keeps adjusts fan speed based on temps so it's pretty quiet as well.

That's a good idea, if I ran Bt848 or Bt878 capture card, that would likely only draw a couple watts at most of power, but I'll have to make sure it supports onboard recording as well. The Sempron 3000+ should be able to encode at least 2 mpeg2 streams at once. The alternative would be the PVR150, that one doesn't have a heatsink on the encoder, so it must use less power.
Good idea, I'll look into it. My desktop now sucks down 90w, my MythTv 110w. Assuming PSU efficiency of 77% (see http://www.silentpcreview.com/article107-page4.html) that means max power draw of 69w and 84w respectively, just under what I'd want to run a PicoPSU w/120w brick at to keep the fan off. Idle PSU efficiency would go up from ~75% to ~85%, saving about 6w-9w each. Not quite worth the $90 to buy one, but I'll definitely keep it in mind.

Do you have a requirement for this system to be on 24/7? If not are you using S3 suspend? I'm running a dual tuner MCE system with a 3000+ venice. I'm not actually using CnQ because it can cause tv to stutter a little in media center. Idle is around 64W measured at the wall, which isn't too bad considering the inefficient PSU. I was initially concerned that this was too high - until I got s3 suspend working (ie so it wakes up automatically and sleeps again for scheduled tasks like tv recording). 90% of the day the machine sits idle in its S3 state using just 6W.
The 64W to approx 100W full load that the machine uses is easily cooled by 2 x 500rpm nexus fans (one in psu one case fan). I wuoldn't really benefit much from a less powerful CPU as it really is inaudible in a quiet room anyway.

I think it is important to prioritise power savings because it is very easy to end up spending a lot of time/money trying to save just a fraction of a watt. I'd say S3 suspend will give you the biggest saving if it is a 24/7 machine. Next undervolting, and then a PSU that is efficient at low power draw like the pico psu is going to offer you the biggest saving. Beyond that you are looking at the law of diminishing returns. You can switch off unecessary ports etc in the motherboards bios, change tv card etc, but its only going to make a very small difference.

Incidentally - its worth being aware that there is no guarantee that a sempron will be lower power that an A64 venice despite having less cache. If you look at the power consumption threads in the CPU/motherboard forum it varies quite dramatically from one chip batch to the next. In theory, as the cheaper model the sempron could well be using cheaper silicon and therefore draw more power than one of the better venice chips (some are rated as low as 30W).

Of course it is pot luck anyway when you by a venice and the difference isn't too much to worry about. Some of the EESFF chips on AM2 look promising though as these are guaranteed to be in a smaller power envelope.

Motherboard/Chipset seems to make a small difference too. The 6100/6150 nvidia boards should use less power than a normal NF3/4 as they are 90nm instead of 110nm. I have heard that the Ati chipset is meant to draw even less than the 61xx series, but I have both a 6150 nvidia board and RS482 ati board and the power draw is very similar. The Ati board is a very cheap ECS model and the 6150 is an expensive MSI board so this could be down to the voltage regulators being cheaper and less efficient on the Ati board, but really i don't think there is enough difference between them to worry about. They both are happy passive in a very low airflow case despite reaching high northbridge temps and thats the most important thing.

Marc

AFAIK there is no difference in the quality of the silicon used for Semprons and Athlons. All Semprons still in retail channels should now be Venice (ie E3/E6 stepping) or Palermo or whatever they call the 90nm core. The Desktop CPU Power Survey showed that the idle and load power consumption of the two procs is always within a couple of watts of each other (in an apples to apples comparison).

No, definitely not. If I could figure out how to get it to S3 suspend when it's not recording or not being used to watch recordings, I would definitely do it. I need to read up on the MythTV docs to figure out the best way to do it, I know that some people are. For some reason though, I was under the impression you needed an external source to wake it up? Could you enlighten me on how S3 suspend works or point me to some docs?


That was the impression I was under as well, though I figured the smaller cache of the Sempron would have lowered power draw some, at least under load.