Page 1 of 1

Athlon 64 power reduction: C'n'Q, Vcore, mobile, tools

Posted: Thu Mar 31, 2005 10:00 am
by jojo4u
Athlon 64 power reduction: C'n'Q, Vcore, mobile, tools

1. Cool'n'Quiet
2. Beyond Cool'n'Quiet
3. Mobile Cores
4. BIOS undervolting
5. Software for changing multiplier/voltage/FSB
6. Interesting threads


I believe this information could be very useful for Athlon users. I will update this list when I encounter new information.
This thread is replacing the old one in the "Silent Front" forum.

1. Cool'n'Quiet
What, Which, How?

Cool'n'Quiet varies the frequency and the voltage of the CPU depending on the actual load.

It reduces the power in 2 ways:
1. It reduces the leakage power at idle. The Athlon 64 puts itself to sleep when the HLT/STPGNT instruction is sent. At this stage, only the leakage power is drawn. By reducing the voltage and frequency, this power is cut down significantly.
2. It reduces the power needed during light and medium load. When the CPU is able to cope the work at a low frequency/voltage level it stays at this level.

The perfomance penalty if this solution is negligible.

All Athlon 64 and all A64-based Semprons >=1,8Ghz support this feature. The registered memory used by the socket 940 CPUs Athlon 64 FX (old) and Opteron does prevent this feature. The new upcoming E3 revision Opterons have Optimized Power Management (OPM) which is the same technique. Apparently, the mainboard has to be compatible.

According to AMDZone, you may experience performance loss using dual-core AMD processors with Cool'n'Quiet enabled on Windows XP SP2. Microsoft is reporting the problem in a knowledge base article, but doesn't let you download a hotfix, you need to contact them directly to get it...
However, a patch is available from AMDZone, which improves CPU performances and addresses some Cool'n'Quiet issues.
Please read the instructions carefully if you decide to apply this patch, since you need to change registry settings also. That patch seems to be a beta one, and this is probably why microsoft is asking you to call. Use it only if you experience instability or performance loss, or if Cool'n'Quiet is not working correctly. A restore point should be made before applying the patch, just in case... (text by Slaugh)

There are known problems with socket-754 boards and 2+ DIMMS. You can find compatibility notes on some boards in the recommended lists. You can also ask in the forum if someone has a stable operation with 2 DIMMs. According to Aopen, 3 DIMM socket-754 configurations are incompatible with Cool'n'Quiet.

When you are choosing an Athlon 64 now, go for the 90nm cores:
Socket-754: "Palermo" (Rev D0) Semprons (which are made out of a cache-halved "Oakville" core) with a TDP of 62W available at the moment. The OPN ends with -BA.
"Palermo" (Rev E3) Semprons are available too. TDP: unknown. The OPN ends with -BO.
Socket-939: There is the "Winchester" (Rev D0) core available. The TDP is 67W and the OPN ends with -BI.
"Venice" cores (Rev E3) are available too. TDP: 67-89W. The OPN ends with -BP. The Venice core is said to consume a tad less power then the Winchester. Sources 1 2
Cache-doubled "San Diego" (Rev E4) are available too. TDP: 89W. The OPN ends with -BN.
If you can't get a 90nm core, get a least a core with the CG-stepping.

Examples of the benefits:
Computerbase got 7W AC drop with a 2800+ (socket-754), 17W AC drop with a 3400+ (socket-754) and 22W AC with a FX-53 (socket-939).
hardwareluxx.de examined socket 939 CPUs and got 23W AC drop for a 3500+ and 35W for a 3500+.
hardtecs4u.de shows 8W DC drop for a 3500+ 90nm Winchester.
The faster your CPU, the more important Cool'n'Quiet gets. All Athlon 64 of the same stepping seem to have the same practical Cool'n'Quiet idle power regardless of their frequency.

Installation instructions here. You can download a dashboard from the AMD site in order to check the correct function.

Athlon 64 drivers here.
Athlon 64 X2 drivers here.
Opteron drivers here.


Troubleshooting

Athlon.de has a hudge german thread about Cool'n'Quiet problems. Common symptoms are: instant crash after activation, crash after some hours, problems with Windows starting.
The following fixes solved some problems:
- do an image of your system partition. Sometimes a malfunctioning Cool'n'Quiet demolished Windows installations.
- get the newest production BIOS. If it doesn't help, get a Beta-Bios.
- install the newest driver from AMD.
- the RAM is quite cruical. If you have a socket-754 board try only one stick. Lower the timings and/or the frequency.
- shift around the RAM sticks.
- Win XP Service Pack 2 seems to make problems.
- try RMClock since it apparently alters the frequency more softly.

Mainboards supporting Cool'n'Quiet

AMD has a list of recommended mainboards. Click on mainboard to see if it AMD validated Cool'n'Quiet on this board.

Some manufacturers (at the moment: Asus, Epox, MSI, Abit-usa) list Cool'n'Quiet compatibility on their specification sites of their products.

2. Beyond Cool'n'Quiet

Tweaking of the P-States

Cool'n'Quiet sets the speed and the voltage to defined levels (P-States). It does not work together with any custom BIOS-voltage or multipler/voltage changing tool. The power user may want to tweak this. For example:
- You want to save even more power by setting the Vcore to the lowest possible level for each state.
- You don't need the maximum power of your CPU (e.g. small server, HTPC) and lower the maximum state.

If you want to play around with the multiplier, you probably have to enable Cool'n'Quiet in the BIOS nevertheless.

Two tools are available for this task:
RMClock or CrystalCPUID. Information about the differences of these two utilities can be found in the SPCR-Forum.
Clockgen is able to adjust your reference clock (aka FSB) via keyboard shortcuts.

Mainboard does not support Cool'n'Quiet

You are not lost. Your energy hunger is still comparable to a Intel Prescott with C1E or EIST since the Athlon 64 still has disconnect on HLT/STPGNT enabled. Here, the CPU basically goes down to a very low frequency (4-250Mhz) while it recieves the HLT or STPGNT instruction. The voltage stays the same, though.
Note: When Cool'n'Quiet is enabled, the disconnect feature is still active.

There is the possibility to underclock your Athlon 64 by adjusting the reference clock (aka FSB). This may be handy in order to limit the heat output when the cpu is busy. There shouldn't be any effect on the idle temperature.
Dynamic underclocking via the reference clock (aka FSB) is not supported by any software afaik.
Clockgen is able to adjust your reference clock (aka FSB) via keyboard shortcuts.

3. Mobile Cores

There is a comprehensive post in the anandtech.com forum.
Mobile cores don't support Cool'n'Quiet out of the box. I suspect that RMClock and CrystalCPUID are still able to set the P-States dynamically. Even Windows XP might be able to access the power profiles - please report in the forum about this.
The current situation in terms of low-power is as follows:
04/2005: There is a 90nm socket-754 mobile version called "Oakville" existing. It has 1,35V, which is not a big step compared to the 1,35/1,4V of the "Venice". Since it is extremely difficult to find a motherboard supporting mobile and 90nm, just get a desktop 90nm.


4. BIOS undervolting

You may have noticed some discrepancies in the recommended Cool'n'Quiet list and the recommended undervoltable list. In theory, every mainboard supporting Cool'n'Quiet should be able to undervolt. But manufacturers often don't include the voltage adjustments in the BIOS.
You may want to have BIOS voltage adjustments when you want to use Linux or when you fear that the booting of you PC may cause you CPU to overheat with the current cooling.
Once you are in Windows, you should be able to adjust the Vcore with the usual utilities including the ones supplied by the manufacturers. CrystalCPUID and Clockgen are providing a command-line interface. If there are any exceptions from this rule, please report.


5. Software for changing multiplier/voltage/FSB

RmClock
http://cpu.rightmark.org/products/rmclock.shtml
This tool allows user adjustable Cool'n'Quiet voltage/frequency settings.

CrystalCPUID
http://crystalmark.info/?lang=en
This tool allows user adjustable Cool'n'Quiet voltage/frequency settings. You need a Athlon64 or mobile/modded XP.

CpuMsr
http://www.cpuheat.wz.cz/html/Tweaks.htm
His software CpuMsr is able to change many settings of you CPU. For example, you can adjust the divisor of the S2K disconnect feature as well as the clock change speed.

ClockGen
http://www.cpuid.com/clockgen.php
Multiplier/voltage/FSB adjustments by keyboard shortcut.

SpeedSwitchXp
http://www.diefer.de/speedswitchxp/indexd.html
Powerful replacement for the Windows XP built-in energy control.

http://www.cpufsb.de
Supporting many mainboards and CPUs.

Speedfan
http://www.almico.com/speedfan.php
IMHO the best temperature logging utility.
Btw, interested in better fan speed algorithm?

SetFSB
Supports changing the FSB on the fly. It is no longer in development.

6. Interesting threads

DFI nForce 4 boards allow overvolting while sticking to AMD Cool'n'Quiet driver. Link


----
8th apr - added OPM
22th oct - added X2 compatibility

----
Suggestions are very welcome.

Posted: Thu Mar 31, 2005 11:26 am
by Coolin
Wow very nice post. :)

I was wondering, is it possible to BIOS undervolt AND enable Cool n Quiet at the same time? I have an Athlon 64 3000+ and I have it running at the stock speed of 1800MHz at 1.168volts. But when I use CPU-Z to check on Cool n Quiet, I see that it lowers my clock speed to 1GHz, but it no longer lowers my voltage... :( I believe my voltage while Cool n Quiet was enabled (before my underclocking) was at ~1.07.

Posted: Thu Mar 31, 2005 1:34 pm
by jojo4u
Coolin wrote:Wow very nice post. :)
I was wondering, is it possible to BIOS undervolt AND enable Cool n Quiet at the same time?
Thanks for your appreciation.
Regarding your problem: "2. Beyond Cool'n'Quiet - Tweaking of the P-States" in the 1st post shows you a solution.

Posted: Thu Mar 31, 2005 1:43 pm
by Rusty075
Coolin wrote:Wow very nice post.
Indeed.

Congrats Jojo, you have the very first sticky for the CPU forum. :lol:

Posted: Fri Apr 01, 2005 12:13 am
by Coolin
jojo4u wrote:
Coolin wrote:Wow very nice post. :)
I was wondering, is it possible to BIOS undervolt AND enable Cool n Quiet at the same time?
Thanks for your appreciation.
Regarding your problem: "2. Beyond Cool'n'Quiet - Tweaking of the P-States" in the 1st post shows you a solution.
Woops. :oops:

Anyway, I got my voltage down to the exact same as when Cool n Quiet is enabled. So right now it's at 1800MHz at 1.072 volts (according to CPU-Z). It's Prime95 stable for 5 hours so far, but I'll keep it running through the night. Perhaps I'll push it further and get cooler than Cool n Quiet temps at higher speeds. :lol:

I'll have to learn how to use those programs and find out if I can get my processor running at 1GHz and 0.8 volts or something. But both of these programs look so confusing to configure. :shock:

Posted: Thu Apr 07, 2005 6:57 pm
by Mats
Great post! :D
The registered memory used by the socket 940 CPUs Athlon 64 FX (old) and Opteron does prevent this feature.
AFAIK, Opteron E got CnQ. (As a side note, AMD removed Opteron D from their lists a while ago.)
Besides, I've seen Opteron mobos with regular RAM support (Abit), dunno how common they are though.
hardwareluxx.de examined socket 939 CPUs and got 23W AC drop for a 3500+ and 35W for a 3500+.
hmm... :P

Edit: I guess Opteron E got PowerNow, not CnQ.

Posted: Sat Apr 09, 2005 4:59 pm
by csimpson
Hey, great post!

Just a question, is it possible to alter the voltage of the P-states?

For example, some method to automatically tell my cool&quiet to volt to 0.9v instead of the 1.1v it volts to when C&Q kicks in?



Thanks for the help, and sorry if I'm making no sense :D

C

Re: Athlon 64 power reduction: C'n'Q, Vcore, mobile, tools

Posted: Sat Apr 09, 2005 5:17 pm
by StarfishChris
Read the whole post.
jojo4u wrote:Tweaking of the P-States

Cool'n'Quiet sets the speed and the voltage to defined levels (P-States). It does not work together with any custom BIOS-voltage or multipler/voltage changing tool. The power user may want to tweak this. For example:
- You want to save even more power by setting the Vcore to the lowest possible level for each state.
- You don't need the maximum power of your CPU (e.g. small server, HTPC) and lower the maximum state.

If you want to play around with the multiplier, you probably have to enable Cool'n'Quiet in the BIOS nevertheless.

Two tools are available for this task:
RMClock or CrystalCPUID. Information about the differences of these two utilities can be found in the SPCR-Forum.
Clockgen is able to adjust your reference clock (aka FSB) via keyboard shortcuts.
n.b. you have to enable Cool 'n' Quiet to change multiplier (and voltage?) on my motherboard at least.

Posted: Sat Aug 06, 2005 12:00 am
by DyJohnnY
ok, seems i discovered crystalcpuid but not this thread.
1. crystalcpuid - excellent tool.
here's what i did.
first of all my processor may be the biggest reason behind all my results, it is one of the best overclocking stepping , LLBE week 17, and s it seems, it likes undervolting also.

so getting back to the story.
I have an epox 9nda3j [nforce3 ultra, w agp]
as people have told me, and as i have noticed myself this board undervolts the cpu. for example, when running at stock volts. i was getting 1.344V under load and 1.378 in idle. so you could say from .25 to .50V undevolt capability.

a few days ago i discovered crystalcpuID.

about my bios settings - i don't know wether usage of crystalcpuID is truly related to Cool'n'quiet being enabled on the board, because for my overclocking i had to disable it, otherwise, as i've read, OC'ing would be complicated.

CPu specs: venice 3000+ at 2520MHz [ram@280Mhz]

I decided to use crystal and bring down the vcore for my venice, and see where it would be stable.
my venice runs at 2520 with voltages varying from 1.26 and 1.286
- this was tested by running large fft's+round off checking in Prime 95.

As my board undervolts, the speeds in crystal cpu-id are set at 1.325V, and the epox voltage reporting utility indicated the v.core voltages .5V lower under load.

now for the temp impact:
-i'm using an Xp-90+a tek-chain fan of 120mm set at it's lowest speed, guessing it's somewhere at 5V-6V, fan is oriented to suck air through the Xp-90.
-the pc is in open air, my case is not complete yet. ambient temperature is somewhere around 30 degrees. the bios of the board reported 36degrees, but i guess the sensor is placed in a warmer spot of the board, and is probably a better judge of temperature of the air beeing sucked through the HS.
-at 1.325 in crystalcpuid. bios reported alternating between 1.260 and 1.286 - i got 52 degrees while running prime 95 with the above mentioned settings.
at 1.4 in crystalcpuid, bios reported between 1.325 and 1.344 under load, with 1.376 in idle - temps were of 54 degrees while running prime 95

i'm pretty happy with this, a computer that's running at 2500mhz on just 1.27Vcore i managed to get it 2 degrees cooler. also idle temps are 3 degrees lower, at 40degrees.

Now comes the part where i have no idea what to do, and i'd like some advice:
1. if my bios settings disabled CNQ how come i can still change voltages?
2. will i be able to change multi's with CNQ disabled from bios?
3. if i cannot change multis, and i enable CNQ, is there any program that uses it by default in windows = sets my cpu to the settings it has by default [i'm worried about crashes due to raised mhz's and no volts to run them]
4. while running the tests, since i never used crystal before, and was worried about stability issues, I set all of the 3 states, in multiplier management to the same voltage, and ofc left the multi's alone.
now the question:
IS there a program that load the PSU with a requested %? i know folding at home has the setting "requested CPU usage" but i was wondering if there is another alternative
as i said i'm a noob when using crystalcpuID, and i'd like to test the voltages i set for him in the 3 states, with a utility that loads the CPU by an exact amount, the percentages u can set in multiplier management.

5. just so i understand this: lowering just the multiplier for the CPU from 9[default] will not cause stability issues if the cpu was stable with voltage X at multi 9, and u don't modify voltage ofc?
to me it feels like common sense, but i'd like some confirmation also :)

Posted: Mon Nov 21, 2005 9:31 pm
by Mats
This is no breaking news but I think some of you haven't seen it. There's a way to determine TDP for new K8 CPU's that doesn't have it printed on the lid. Here's an example of what you'll see:

Code: Select all

AMD64 TCaseMax - v1.13
----------------------

CPU Information (CPU #1):
Standard CPUID: Family: F, Model: 7, Stepping: 1
Extended CPUID: Family: F, Model: 27, Stepping: 1
CPUID String: 20F71
Processor APIC: 0
Processor: AMD Athlon 64 3700+ (San Diego)
CPU Speed: 2880.13 (320.01 x 9.0)
Revision: SH8-E4
Platform: Socket 939
Startup VCore: 1.400v
Maximum Case Temperature = 55C
TDP: 38.2 Watts
TDP varies from CPU to CPU, and once again we see how hard it is to figure out the correct TDP. You actually have to use the CPU to find it out! The listed TDP for a 90 nm 3700+ is 89 W, the same as for the 90 nm 4000+, but also for the 130 nm 4000+. In other words, completely outdated. (Mobo makers can get future TDP guidelines from AMD, the actual CPU doesn't need to have that info, only the current TDP.) This is the reason why we see new threads from newbies every week asking about how to cool down a A64 and worries about the TDP, it is not easy to find it out.
A 38.2 W TDP (= 30 W, Intel TDP style?) is not really hard to cool down, even without undervolting.
Very interesting to see some dual core CPU's with a TDP = 35 W while this slower one uses 84.4 W! :D :?

Posted: Mon Nov 21, 2005 10:04 pm
by Mats
Found some interesting reading in this Opteron document, Thermal Profile looks different to me. Ambient temp = 42° C, I guess that's a typical server environment... :wink: Another reason for taking those TDP's with a grain of salt. What temp do they use for A64?

Posted: Fri May 26, 2006 5:44 pm
by wwenze
I'm having trouble selecting below 1.100V in CrystalCPUID (anything lower it sets to 1.100V), anyone has the same problem?

Never mind, found the reason, I'm using an E6 CPU.

Posted: Thu Aug 10, 2006 1:28 pm
by QuietOC
wwenze wrote:I'm having trouble selecting below 1.100V in CrystalCPUID (anything lower it sets to 1.100V), anyone has the same problem?

Never mind, found the reason, I'm using an E6 CPU.
E3's also have this problem. Get a "D0".

Posted: Sat Sep 02, 2006 9:18 am
by frostedflakes
Was messing around with RMClock on an X2 3800+ AM2 system, and looks like I can go <1.1V. In fact, RMClock gives me options down to 0.55V (must be a glitch, because I didn't think K8 motherboards have hardware for <0.8V). I tried a few lower options, though, such as 1.075V and 1.05V, and according to monitoring software Vcore actually was <1.1V, they weren't just phantom settings that did nothing. The only problem is after 10s or so at any setting below 1.1V the computer crashes.

So I dunno, just thought it was kind of interesting. I always figured the <1.1V limit was something hardware based, but it looks like it might be possible to get around it. Maybe a talented programmer / K8 P-state guru could figure out what the trick is and write a program to take advantage of it??? :)

Changing undervolt

Posted: Tue Sep 05, 2006 8:08 am
by NeilBlanchard
Greetings,

I just had to reduce the undervolting setting on my X2 4200+ -- it was at 1.275v (stock is 1.35v), and recently I have been getting freezes; and it has failed to "wake up" on occasion. I have just bumped it up the 1.3v (the smallest increase possible on my Gigabyte motherboard), and so far so good.

I'm thinking that this is most likely due to changes in the PS (SeaSonic S12 430) during it's first summer, but it might be changes in the CPU, I suppose?

Posted: Wed Oct 18, 2006 3:44 am
by burebista
@frostedflakes Mine AM2 X2 3800+ is perfect stable from FID 4x VID 0.900V to FID 10x VID 1.125V on ABIT KN9. Playing with RMClock, of course.
At VID 0.800V I'm getting errors in Orthos, but in desktop is fine.

@NeilBlanchard An interesting thread.
It seems that there is a protection diode between the memory controller and the CPU that can forward bias if the memory voltage gets too high. This leads to a quick death of the CPU. AMD won't provide numbers -- they just recommend increasing Vcore.
Consensus from the thread is that Vdimm/2 - Vcore < 0.325V (that's 1/2 the memory voltage minus the CPU voltage should be less than 0.325V)...
But I guess on DDRII Vdimm 1.80V should be fine with Vcore 0.900V. :)

Cool n Quiet and Linux

Posted: Fri Oct 27, 2006 9:11 pm
by damien
I saw that the giant list of programs for tweaking were all Windows.

Given that there's more than one mention of MythTV in the forums, what is the current status of CnQ for Linux.

Last time I really played with it was back with (a now ancient) 2.6.15 kernel, where my Pentium M laptop refused to boot with ACPI enabled - and other power saving features weren't well supported.

Does anyone know much better these days? I'm spec-ing out a new system and I have been burned before on hardware that was "supported"... well, sort of.....

The Linux Question

Posted: Wed Nov 22, 2006 12:52 am
by codysnider
Ok, to answer the MythTV question regarding Linux and C'n'Q:

I use an AMD Turion MT-40 and an Ubuntu Edgy Eft dictribution (stay away from Dapper Drake for the latest in MythTV, problems communicating with a MySQL 5 database...which MythTV relies on heavily.)

Checking from when I started this post without a VMware Windows XP and Mac OSX setup running, I was sitting at 1.01 on the vCore and after starting both those systems it jumped me to 1.31. Also, the far throttles automatically from the 900s to about 1300 when I approach 50 degrees. All these can be monitored through the console command: sensors (if you have lm-sensors installed.)

I believe the Edgy Eft distribution is running a 2.4.17 kernel under the X.org window manager. I'd highly recommend using this distro if you plan to use MythTV on either a server OR client box.

Any more questions: Visit my Linux HOWTO and hardware blog CodySnider.com

Hope that helps. Cheers!

AMD power improvements

Posted: Mon Feb 12, 2007 2:09 pm
by NeilBlanchard
Hello,

An interesting glimpse of some (possible) improvements in future AMD CPU's:

http://theinquirer.net/default.aspx?article=37574

Posted: Mon Feb 12, 2007 3:01 pm
by jaganath
With Barcelona/K10, each core can change speed as needed potentially saving a lot of power. The only problem is that they can not vary voltage independently on each core so more power is consumed than necessary.
That's a shame, it's the reduction in voltage that brings the really big savings.

AMD keeps pushing power savings but Intel is right there with them in that area, and in the performance arena there's just no contest. It's as if AMD is abandoning the retail battleground and focusing on servers etc.

SOLTEK SL-K8AV2-R1L Sempron Socket 754 motherboard

Posted: Thu Feb 21, 2008 2:51 pm
by sdgengineer
Good article. I have the subject motherboard, and experienced no problems with it, until I attempted to add an additional 512 MB DIMM on the motherboard, It has two DIMM sockets (DDR 400). After I installed the DIMM the computer would not boot.

Based on your excellent article, it appears that I will have to live with the existing memory. Do you know of any solution? Would a single 1 GB DIMM work, if I wanted to get one.

Again thanks for your excellent article. I realize I could buy a new motherboard, but I was just trying to boost performance of an older computer

Scott

Posted: Thu Feb 21, 2008 3:00 pm
by jojo4u
This article is mainly obsolete for today's Athlon CPUs but your combination still covers it. Hmm did you try it with the new stick only? Perhaps it's defective? If it boots with only the new one, get into the BIOS and set every RAM setting to lowest speed. Lower lower MHz and and put command-rate to 2T. Then try to raise the timings. Now you can try with two sticks again.

Sempron Socket 754 motherboard

Posted: Thu Feb 21, 2008 3:09 pm
by sdgengineer
Thanks

I swapped it out, and don't believe the memory is bad. I realize its somewhat obsolete (its a Sempron 1.6 GHz processor) I will try your suggestions.

Regards

Scott