Undervolting an AMD 939 via software

Steerpike · Post by **Steerpike** » Fri Mar 24, 2006 9:35 am

Every time I decide on a configuration, some key component is not available from the vendor I want to use, and I end up re-thinking. My latest hope (I'll call it that rather than plan!) is to get a pretty standard AMD 939 processor / board, rather than trying to build a Turion based desktop (last straw was that the DFI Lanparty UT nF3 250Gb board was out of stock at NewEgg!).

I've seen several posts that suggest that an undervolted 939 will offer similar heat benefits to a Turion and I'm willing to go that route.

Since I'm building an HTPC, and want a 'narrow' case, the line of boards that include the nVidia 6150 is appealing, especially since I'm planning on using the Silverstone LC11 case which requires micro-ATX.

But finding a micro-ATX that has both the embedded 6150 AND is undervoltable seems to be a challenge in itself ... so ... I've also seen some reference to the fact that one can control the V-Core in software. Is this true, and, would such software run on a mobo such as the Asus A8N-VM CSM? This would seem to be an appropriate solution.

Tibors · Post by **Tibors** » Fri Mar 24, 2006 10:15 am

Read this SPCR article: CrystalCPUID: User Configurable Cool 'n' Quiet

Basically any mobo that has a working C'n'Q implementation should work with that program.

widman · Post by **widman** » Fri Mar 24, 2006 7:32 pm

Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.

defaultluser · Post by **defaultluser** » Sat Mar 25, 2006 2:01 pm

widman wrote:Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.

That's a VERY SERIOUS accusation with huge repercussions for all of us here at SPCR. The least you could do after making such a general sweeping statement is provide us a link.

You sure it's not your motherboard's lack of undervolting support?

dragmor · Post by **dragmor** » Sat Mar 25, 2006 2:39 pm

defaultluser wrote:
widman wrote:Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.
That's a VERY SERIOUS accusation with huge repercussions for all of us here at SPCR. The least you could do after making such a general sweeping statement is provide us a link.

You sure it's not your motherboard's lack of undervolting support?

He is right. My 3500+ refuses to be set below 1.1v via software (Crystal or RMClock), but is quite happy to be set below that in the bios. I'm not sure if the software is reading a value on the cpu and refusing to set it below or not. But the software reports a minimum voltage of 1.1v.

Things I learnt when undervolting my CPU
1) Software will no set the voltage below 1.1v
2) My CPU will not hardboot below 1.2v
3) If I go into the bios and change the CPU voltage or multiplier and then save the changes I can boot to whatever I want. And my CPU is 48hrs P95 stable with 0.8ghz @ 0.8v.
4) CnQ on my system just drops the multipler to 5 and the voltage by .3v. So I'm quite happy with 2.2ghz @ 1.2v (to hardboot) and 1.0ghz @ 0.9v.

widman · Post by **widman** » Sat Mar 25, 2006 8:06 pm

defaultluser wrote:
widman wrote:Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.
That's a VERY SERIOUS accusation with huge repercussions for all of us here at SPCR. The least you could do after making such a general sweeping statement is provide us a link.

You sure it's not your motherboard's lack of undervolting support?

I have been using Clawhammer, Manchester and Venice in the same motherboard, MSI K8N Neo2. Clawhammer can be set to 0.850V @ 4x multiplier. While Venice and Manchester refuse to set below 1.10V by software CPUCrystalID or by BIOS.
here user experience
http://forums.silentpcreview.com/viewto ... highlight=
here implicitly say so, no voltage below 1.1V
http://forums.silentpcreview.com/viewtopic.php?t=28920

But winchester core possible to volt until 0.8V. And Venice core onward, it seems AMD set 1.1V as lowest voltage.

You can find a lot of threads here regarding 1.1V

defaultluser · Post by **defaultluser** » Sat Mar 25, 2006 10:00 pm

widman wrote:
defaultluser wrote:
widman wrote:Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.
That's a VERY SERIOUS accusation with huge repercussions for all of us here at SPCR. The least you could do after making such a general sweeping statement is provide us a link.

You sure it's not your motherboard's lack of undervolting support?
I have been using Clawhammer, Manchester and Venice in the same motherboard, MSI K8N Neo2. Clawhammer can be set to 0.850V @ 4x multiplier. While Venice and Manchester refuse to set below 1.10V by software CPUCrystalID or by BIOS.
here user experience
http://forums.silentpcreview.com/viewto ... highlight=
here implicitly say so, no voltage below 1.1V
http://forums.silentpcreview.com/viewtopic.php?t=28920

But winchester core possible to volt until 0.8V. And Venice core onward, it seems AMD set 1.1V as lowest voltage.

You can find a lot of threads here regarding 1.1V

Thank you. I just wasn't happy with a passing reference, since this is the first I've heard of it.

I have a feeling AMD did that just to prevent OEMs from making "Turion equivilant" notebooks using desktop processors.and software like CrystalCPUID. Thus, you can undervolt all you want in the BIOS, but the software hooks are disabled.

widman · Post by **widman** » Sun Mar 26, 2006 12:31 am

defaultluser wrote:Thank you. I just wasn't happy with a passing reference, since this is the first I've heard of it.

I have a feeling AMD did that just to prevent OEMs from making "Turion equivilant" notebooks using desktop processors.and software like CrystalCPUID. Thus, you can undervolt all you want in the BIOS, but the software hooks are disabled.

I can set voltage lower than 1.1V in bios. But when do monitoring again it say 1.1V. Temperature reading is indicating 1.1V, not lower than that. Somehow, the cpu set the voltage or I will say cpu refused voltage less than 1.1V by software or by bios.

I think the reason is not "Turion Equivalent", but more voltage reason. Due to small process, a 90nm process is not that robust for voltage different. There are many reports say that CPU killed by high vdimm voltage. So maybe AMD set a lower limit 1.1v due to vdimm 2.6V

http://www.xtremesystems.org/forums/sho ... hp?t=79509

dragmor · Post by **dragmor** » Sun Mar 26, 2006 2:04 am

widman wrote:
defaultluser wrote:Thank you. I just wasn't happy with a passing reference, since this is the first I've heard of it.

I have a feeling AMD did that just to prevent OEMs from making "Turion equivilant" notebooks using desktop processors.and software like CrystalCPUID. Thus, you can undervolt all you want in the BIOS, but the software hooks are disabled.
I can set voltage lower than 1.1V in bios. But when do monitoring again it say 1.1V. Temperature reading is indicating 1.1V, not lower than that. Somehow, the cpu set the voltage or I will say cpu refused voltage less than 1.1V by software or by bios.

I think the reason is not "Turion Equivalent", but more voltage reason. Due to small process, a 90nm process is not that robust for voltage different. There are many reports say that CPU killed by high vdimm voltage. So maybe AMD set a lower limit 1.1v due to vdimm 2.6V

http://www.xtremesystems.org/forums/sho ... hp?t=79509

I've got no problems setting it lower in the bios and/or via CnQ and I'm running my ram at 2.8v (default for the sticks) with my CPU mostly idle at 0.9v.

Although my CPU is one of the earlier E3 Venices.

buzzlightyear · Post by **buzzlightyear** » Sun Mar 26, 2006 6:18 am

Strange.

I ran an Opteron 144 (Venus E4 core) at 1.075v @ 1.8ghz using CrystalCPUID. In fact, I am using that computer now.

I also had an Opteron 165 (Teledo E4?) and I could run @ 1.025v @ 1.8ghz using CrystalCPUID. In fact, I Primed it for 24 hours with that setting.

In both cases, the ram was set to 2.6v by default. All the monitor software (SmartGuardian, CPU-Z) reported the correct voltage.

If I use C'nQ, then it would set it to 5x multi and 1.1v for both CPUs.

The motherboards are DFI RS482 and Biostar Tforce6100-939

widman wrote:Just warning, new amd core - venice, san diego, manchester, toledo refuse to undervolt less than 1.1V.

buzzlightyear · Post by **buzzlightyear** » Sun Mar 26, 2006 6:21 am

Deleted

jaganath · Post by **jaganath** » Sun Mar 26, 2006 6:31 am

I ran an Opteron 144 (Venus E4 core) at 1.075v @ 1.8ghz using CrystalCPUID. In fact, I am using that computer now.

I also had an Opteron 165 (Teledo E4?) and I could run @ 1.025v @ 1.8ghz. In both cases, the ram was set to 2.6v by default.

Can I ask, what kind of cooling did the Opterons require at that Vcore? Did you have to run the heatsink fan at 12V, 7V, 5V? Any chance of cooling them semi-passively that way?

buzzlightyear · Post by **buzzlightyear** » Sun Mar 26, 2006 6:41 am

I used the AMD stock 4-heatpipe HS for both PCs. On the DFI RS482, I use the Smartfan (in the Bios) and the fan is rev'ing at 1400 rpm when idle. With the Biostar, I use the Biostar verion of Speedfan and keep the fan rev'ing at around 1600 when idle.

Temperature at idle is around 36-37c, about 10c higher than my Mobile Oakville rig.

I don't think I can cool the Opteron semi-passively.

May be I will try a Ninja.

jaganath wrote:
I ran an Opteron 144 (Venus E4 core) at 1.075v @ 1.8ghz using CrystalCPUID. In fact, I am using that computer now.

I also had an Opteron 165 (Teledo E4?) and I could run @ 1.025v @ 1.8ghz. In both cases, the ram was set to 2.6v by default.
Can I ask, what kind of cooling did the Opterons require at that Vcore? Did you have to run the heatsink fan at 12V, 7V, 5V? Any chance of cooling them semi-passively that way?

buzzlightyear · Post by **buzzlightyear** » Sun Mar 26, 2006 7:42 am

Actually, I just made a good point without realizing it.

With my 35w Mobile Oakville, even at full speed 1.475v (262 x 9=2.34ghz) and the same AMD stock heatpipe heatsink at low fan speed 1300rpm, it idles at 26-27c. At 1.1v and 1.5ghz, it idles at 2-4c above room temp (like 24c).

With my Opteron 144 @ 1.075v at just 1.8ghz and the same heatsink at low speed, it idles at 36c (repeated the same setup on two computers).

Based on my limited experience with 3 computers, I don't see how a s939 Venus/Telero core CPUs can match the Mobile Turion (or Oakville in my case) in term of running at low temp.

jaganath · Post by **jaganath** » Sun Mar 26, 2006 9:06 am

That's a very interesting finding; even deeply undervolted, a standard desktop chip is beaten on a performance per watts basis by a mobile chip at stock volts. This supports the notion that mobile chips are "cherrypicked" silicon, which presumably offer excellent overclocking and thermal characteristics. The only fly in the ointment is the spotty support for Turions on desktop boards, which seems to be a problem for Intel mobile chips as well, incidentally.

buzzlightyear · Post by **buzzlightyear** » Sun Mar 26, 2006 12:36 pm

Slight correction.

Actually, even deeply undervolted, a standard desktop chip is beaten on a performance per watts basis by a mobile chip OVERvolt'ed. My Oakville is overvolt'ed from stock 1.35v to 1.475v. And it still run cooler by almost 10c than my other two Opterons undervolted to barely over 1.0v and at stock speed.

mb2 · Post by **mb2** » Sun Mar 26, 2006 6:38 pm

it may be the case with normal a64s too.. but aren't optys hotter than the standard a64s?

Mats · Post by **Mats** » Sun Mar 26, 2006 7:48 pm

I think Opterons have no 1.1 V limit, but on the other hand they tend to run hotter (at least single core) than an A64 counterpart. My old 146 used to run 2 GHz stable at 1.0 V.

Tzeb · Post by **Tzeb** » Sun Mar 26, 2006 10:18 pm

My opteron can go as low as 0.8V. I keep it at 270x4 with 0.8v at idle and 270x10 with 1.4v at full load, all curtesy of RMclock. My old Venice E6 could not go lower than 1.1v, even if i set a lower value in bios, it would still work at 1.1

Steerpike · Post by **Steerpike** » Sun Mar 26, 2006 11:19 pm

Well, I went ahead and ordered (after a month of vacillation) the s939 mobo, CPU, and case yesterday, only to read the updates to this thread today!

I guess I will live with 1.1V ... I guess I'll HAVE to live with it!

I also discovered, after ordering, that the motherboard I picked is incompatible with the risers that are needed in the LC11 case (my oversight) ... so I either have to pick a different case, or a different mobo - or live without any cards (which I may be able to do). Quite amazing!

morkys · Post by **morkys** » Tue Mar 28, 2006 8:57 am

I have another post asking about the most efficient CPU for my PC:

http://forums.silentpcreview.com/viewto ... 134#255134

Drop in and help me out if you like.

...now back to your regularly scheduled thread

smilingcrow · Post by **smilingcrow** » Tue Mar 28, 2006 9:15 am

jaganath wrote:That's a very interesting finding; even deeply undervolted, a standard desktop chip is beaten on a performance per watts basis by a mobile chip at stock volts. This supports the notion that mobile chips are "cherrypicked" silicon, which presumably offer excellent overclocking and thermal characteristics.

Interesting, I interpret the same data as supporting what I thought was a given, which is that Turions use a different ‘design’ to the desktop chips, as opposed to being ‘cherry picked’. By different ‘design’ I mean that in the design process they chose circuits that are optimised for lower power rather than speed.

jaganath · Post by **jaganath** » Tue Mar 28, 2006 9:46 am

I interpret the same data as supporting what I thought was a given, which is that Turions use a different ‘design’ to the desktop chips, as opposed to being ‘cherry picked’. By different ‘design’ I mean that in the design process they chose circuits that are optimised for lower power rather than speed.

There was some debate as to whether Turions were simply undervolted A64 cores. The above info shows this is not the case.

Your interpretation is odd in suggesting that the circuits are optimised for lower power rather than speed, as buzzlightyears experience suggests that no clock speed has been sacrificed to get the lower power; rather the opposite in fact, the overvolted Turion puts out less heat than the undervolted Opteron while operating at 540Mhz faster clock speed.

nutball · Post by **nutball** » Tue Mar 28, 2006 9:48 am

Steerpike wrote:I also discovered, after ordering, that the motherboard I picked is incompatible with the risers that are needed in the LC11 case (my oversight) ... so I either have to pick a different case, or a different mobo - or live without any cards (which I may be able to do). Quite amazing!

I made the same mistake with my mobo and LC04, but I just bought a flexible PCI riser -- it's not ideal because it doesn't give the PCI card much physical support internally, but it gets the job done and it's a lot less hassle than changing case or mobo.

smilingcrow · Post by **smilingcrow** » Tue Mar 28, 2006 1:06 pm

jaganath wrote:There was some debate as to whether Turions were simply undervolted A64 cores. The above info shows this is not the case.

Now I’m completely confused, not that this is unusual.

When you said, ‘This supports the notion that mobile chips are "cherrypicked" silicon’, I thought you meant that Turions were simply desktop chips that were chosen for their ability to run at lower voltages and power consumption. I now assume that you meant the opposite! Your choice of language was semantically confusing to me!

jaganath wrote:Your interpretation is odd in suggesting that the circuits are optimised for lower power rather than speed, as buzzlightyears experience suggests that no clock speed has been sacrificed to get the lower power; rather the opposite in fact, the overvolted Turion puts out less heat than the undervolted Opteron while operating at 540Mhz faster clock speed.

Oops, I didn’t make myself very clear at all. My understating is that the Turion was designed so that the trade off between ‘circuit switching speed’ and power consumption is more balanced towards lower power consumption than in the desktop chips. I wasn’t referring to clock speed, but the circuit switching speed, which I assume means that Turions have a lower IPC than desktop chips!
How this choice relates to the maximum possible clock speed of the Turion I don’t know. This could be more limited by AMD’s chosen power envelope for the Turion rather than a limitation of the actual design process.

In AMD’s own words:

‘More specifically, the process used to manufacture AMD Turion 64 mobile technology has been optimized for thermally efficient processor operation, enabling reduced power consumption during various system performance states and sleep states.’ See this URL:

http://www.amd.com/us-en/Processors/Pro ... html#95417