The Radeon 6850 looks like a great balanced card for silence + gaming.
I just picked up an Asus 6850 and I was curious what stable undervolts anyone is getting? I've gotten to 1.05v and it seems pretty stable, but I haven't pushed the boundaries yet since I haven't had the time to find a perfect stable mix. The goal in the end isto keep the core near stock and oc memory to around 1100 while undervolting the core as far as I can. Hopefully there will be bios editing support soon as well, but it's nice how many tools make it easy to undervolt.

I recently purchased a Sapphire 6850 and have been running through the standard tweaks and testing. After examining the undervolting/underclocking potential and benefits, I have concluded that for all practical purposes the benefits of undervolting/clocking via software are negated by the cards internal behavior.

While in 2D mode, the card downclocks the core and memory to 100/150 (vs. 775/1000 in 3d mode), and reduces the voltage from stock (1.149 on my card) to .945. In an ideal situation, there would still be potential gains to be had from lowering the voltage during 3d operation (& 2d mode but there is no software option that I'm aware of for that). However, the card does not reduce the voltage while in 2d mode once you change the 3d mode voltage to anything other than the default. Thus, unless your mostly in 3d mode, it is actually detrimental to manually undervolt the card b/c it will be operating more often than not at a higher voltage (a reduced 3d mode vs. the lower .945 2d mode).

This behavior could conceivably be overcome by making changes to the bios (e.g., if default 3d mode is reduced in the bios, the card may still reduce the voltage while in 2d mode). If you come across confirmation one way or another, please post.

As an aside, my card ran through 3dmark11 with no issues while undervolted to 1.03 (maintaining stock core/mem settings). Having discovered the nuance set forth herein, I did not try any lower voltage or any further stressing at 1.03.

My card overclocked to 940/1149 (core/mem) with the voltage at 1.174. This seems to be the standard OC without significantly increasing the voltage (according to Anandtech's findings). I may be able to achieve those clocks with a lower voltage but have not tried.

^^

I went further than that with my Sapphire HD 6850. It turned out that my card can work at 0.945 V even in 3D mode! Games, 3DMark 11, even FurMark(!) - everything works fine. So I left voltage at 0.945 (actually, at 0.943) in MSI Afterburner. I haven't tried going any lower, because, frankly, I don't think it's necessary. The improvement is already very significant: the card is reasonably warm and, thanks to a changed fan profile, whisper-quiet even in FurMark/MSI Kombustor at 1800 rpm and only 70C (24C ambient). There are only two issues:

1. Sapphire Trixx, the card manufacturer's new utility, doesn't support undervolting - only overvolting. I wonder what's the best way to contact Sapphire and let them know that undervolting may be necessary too.
2. The card resets voltage after waking up from sleep, and Afterburner doesn't correct it automatically, so you have to do it manually.

Interesting. I figured that such an undervolt in 3d was unlikely (and thus didn't mention it as a possible occassion that would nullify my conclusion). I'll have to try it out to see if your card is an anomoly.

Trixx' inability to undervolt caused me to switch to Afterburner. I rather liked the simplied interface of Trixx but I'll probably stick with Afterburner now.

The resetting of the voltage is another instance that could also be rectified through bios modification.

Although I set a very conservative fan speed map, I have not yet experimented with finding the optimal miminum settings to address load usage. Your setting certainly provides a good start (approx. 45% at 70) if you're maintaining those temps while running FurMark.

"I haven't tried going any lower, because, frankly, I don't think it's necessary."
Don't stop there. Afterall, undervolting isn't really "necessary."
What is the power draw under load before & after undervolting (e.g., measuring using a Kill A Watt)? Curious what the actual power savings are.

I was a little more optimistic from the start because I knew that the reference card's stock voltage is 1.094V, not 1.148V:

AMD Radeon HD 6850 Overclocking Roundup

It looks like all cards on the market are overvolted for AMD Overdrive upper limit (850 MHz), so there is a good possibility of undervolting.


My fan profile is based on two points: 20% at 40C and 50% at 80C. The latter is what I could maintain in FurMark prior to undervolting (barely tolerable temperature vs. barely tolerable fan speed ). But now I'm getting approximately 43% at 70C, and the difference between 50% and 43% is very noticeable.



Oh, I actually tried going lower immediately after posting my last post. It didn't work. It looks like 0.943 V is the lowest voltage that actually works. But even if my card is an exception, I think most cards can work at 0.96-0.97V. It is higher than the stock voltage at idle, but acceptably so.


I don't have the equipment to measure it, but, as far as I know, power consumption should be proportional to V^2.

(0.943/1.148)^2 = 67%

In other words, we should be saving up to one third of power. That's 40W out of 120W. Not bad!

Thanks for the fan profile recommendation. It should save me some time.

My card did not undervolt as well as yours. The card failed 3dmark11 at .945 and .975 (requiring a reboot). It was stable at 1v (I didn't not try b/w .975 and 1). Have you overclocked your card? I'm curious how well it overclocks given how well it operates while undervolted.

Measured energy savings from the wall measured by a Kill A Watt were as follows:
Test: Stock Voltage 3D ; Undervolted @ 1v
1: 195 ; 176-182 (177avg)
2: 183-188(185avg) ; 168-172 (169avg)
3: 184-188(186avg) ; 170-172 (171avg)
4: 187-188 ; 171-177(173avg)
Physics: 205-207(206avg) ; 208-213(211avg)
Combined: 244-260 (255avg) ; 237-250(244avg)

The system idles at 103w (when allowed to downvolt to .945 in desktop mode).

The Physics test suggests that the PC had warmed up a bit more for the undervolted test and was therefore a little less efficient. Adjusting the stock results by 5w, the savings attributable to the video card were consistent with the formula ~24% (1/1.148)^2). For, e.g., (200-103)-(177-103)/97.

Ultimately, if the card does not undervolt to the reduced 2d voltage, long term usage savings will be dependent upon the ratio of desktop:3d use. This can be addressed by saving a profile with reduced voltage in Afterburner that is triggered by 3d mode (found under the Settings -> Profile tab).

Of course, it will be up to each user to decide whether they want to save power at stock or set the 3d profile to be an overclocked profile.

No, I haven't. Well, I briefly ran it at 850 MHz and stock voltage to make sure it actually works, and I'm not going to push the card any further. What matters to me is cool and quiet operation, and this is what I'm getting: 775Mhz at 0.943V, and from 62C/1300 rpm to 65C/1600rpm in actual games. The card is whisper-quiet. The only thing that bothers me is RAM/VRM cooling. Is 1300-1600 rpm enough to keep them cool?


Personally, I'm not very interested in long-term power savings - I'm undervolting my card for a different reason. And it should be noted that, according to Afterburner help tips, it is not recommended to combine automatic profiles with voltage control.

With a similar fan profile, my results are consistent with yours. At stock voltage, Furmark maxes at 80C w/the fan at 48% (~2100rpm). At a reduced voltage (1v), it maxes at 74c w/the fan at 43%. The higher temp on my card reflects the higher undervolted voltage (1 vs. .945).

The card idles at 42c at both .945v and 1v.

I agree that the difference b/w 43% & 50% is significant.



To the extent the purpose of the undervolt is to control heat & reduce the optimal rpm of the fans (which was also one of my reasons for undervolting), what are the differences in temps/optimal rpms while in actual gameplay as a result of the undervolt? You indicate the temps/rpms after the undervolt but did not provide stock stats during gameplay.

After a few laps in Dirt2 I'm getting 65C at 2150rpm with stock settings (voltage and fan speed) and 64C at 1500 rpm with modified settings. Even stock settings are quite tolerable, but undervolting makes things even better. And there is quite a lot of room for adjustments - I could set higher temperatures (70 or even 75C) and lower fan speed, etc.

Those are solid gains. I tend to opt for higher temps in exchange for lower fan rpms/noise.

I was going to remove the fan assembly/heatsink cover to inspect the size of the heatsink (which requires removal of the entire heatsink) but found the following pic of the heatsink (source review). Considering how well the Accelero S2 performs on the 6850, I'm inclined to think that the Sapphire cooler with a larger fan (or 2) is likely to yield additional benefits over the single stock fan. It appears you might be able to fit 2 x 92mm fans (if not, certainly 2 x 80mm fans). It would also enhance cooling the VRM (see 2nd pic).


VRM cooling

I've tried overclocking/undervolting, and my card goes to 850MHz at 1.006V. Temperatures are also pretty nice: 67C at 1650rpm in Dirt 2 and 76C at 2000rpm in FurMark. Idle temperature is 45C at 1050 (only 1C higher than at 0.943 V).

The new version of the Radeon Bios Editor is out and it supports voltage modification on 6850's (Clock settings -> GPU registers).

Were anyone able to lower the idle fan speeds below 1900RPM? Asus SmartDoctor cannot go below 22% fan speed.

Also, at idle speeds (i.e. 2D mode), do you notice any whining noise from the GPU fan? Not sure if "whining" is a correct word, but there is some other sound in addition to the typical turbulence noise. My computer is just two feet away from me and its pretty annoying. Wondering if I got a bad sample. For comparison, the GPU fan "whine" is louder than the Scythe Mugen2B PWM fan @1300RPM, so its probably at or above 24dBA.

The reported 1900rpm at 22% suggests that there is either a configuration problem, a hardware fan controller defect or the software you are using is faulty b/c it extrapolates into a fan that spins at 8,636rpm at 100%. Needless to say, no gpu cooler uses a fan that natively spins that fast (the fan on my Sapphire is ~4,250 at 100%).

Do the following in order:
(1) Open Catalyst Control Center - > Performance tab -> AMD Overdrive. Make sure that "Enable manual fan control" is checked. If it isn't, check mark it, click apply and reboot your computer.
(2) If # 1 does not work, uninstall the Asus program and install either Sapphire Trixx or MSI Afterburner. Both allow you to adjust fan speeds to the lowest permitted by the bios (~1000rpm or ~20%). I would recommend MSI Afterburner since it also permits undervolting.

These steps should rule out a software issue.

If neither of these suggestions work, please identify the model of your card in your reply post.

Thanks for your suggestions! First, the video card is Asus EAH6850 DirectCU (the one that has a review here at SPCR). I've tried to lower the fan speed with AMD Overdrive down to lowest 20%, but it still stays at ~1900RPM. Whats interesting is that if I raise it to 100% (still in AMD Overdrive), the fan speed goes up to ~4200RPM. As you noted, that is a more realistic max fan speed, but 20% of it should be under 1000RPM, not 1900RPM.

Then I uninstalled Asus SmartDoctor, restarted the machine and tried lowering the fan speed using MSI Afterburner. The lowest fan speed it could do was 20%, but at that speed it sounded the same as before. Then I've enabled the "advanced fan controls" in MSI Afterburner and lowered the fan speed to be 10% and 0%. I couldn't hear a difference, but raising it to say 50% clearly spins up the fan. So it looks like, the card is ignoring fan speed settings below 20% from AMD Overdrive or MSI Afterburner.

Is this a faulty card or a software/drivers issue? I ever reinstalled Windows 7 to no avail.

Sorry to derail this thread a bit!

Given the max rpm of the Asus fan, it should be operating no differently than that of the Sapphire model I have and the lowest setting should result in the fan spinning at ~1000rpm.

You needn't adjust the fan with AMD Overdrive. Reset it so that it is at its default. You should just have the option to do so checked (although in most instances, fan control will still work without that box checked). When using Afterburner, what is the lowest rpm you can achieve according to the tachometer? If the tachometer is not showing, click on Settings -> Monitoring -> Active hardware monitoring graphs -> fan tachometer (make sure it's checked). Let's make sure the 1900rpm reading you received via Asus' prog isn't incorrect. If Afterburner shows 1000rpm, then it would appear the fan itself is defective b/c 1000rpm should be either close to inaudible or very quiet (depending on other features of the fan).

The card will not lower the fan below 1000rpm (or 25%). This is the hard limit set in the bios. Afterburner (or any software) settings below 25% is of no actual consequence. Your problem is that the card is not allowing you to decrease the fan speed below ~45% (1900rpm).

3rd Suggestion - uncheck "adjust fan" in AMD Overdrive and again adjust the fan speed using Afterburner.
4th Suggestion - I believe new drivers were released today. Uninstall the old (remove the settings if the option is provided; otherwise run Drive Sweeper) and install the new drivers. Perhaps there's an odd driver quirk affecting your setup (I highly doubt it given your description and the fact a clean install yielded no different result).

It sounds like the fan controller may be faulty and is not actually lowing the fan below 50%. In that case, return the card to the retailer (typically the quickest) or obtain an RMA.

Don't you think the Asus card could simply have a different hard limit?

I asked for the make of the card for this very reason. After learning the card, I discounted that possiblity b/c of the SPRC review:

"When idle, the stock heatsinks on both cards are very quiet, even by SPCR standards, with the reference model coming out slightly ahead; it is just a little bit quieter and smoother sounding. On load, the reference cooler is much louder, emitting 6~7 dB more than the amazingly quiet Asus DirectCU cooler. Our test system with the Asus card installed measured only 17~18 [email protected] and the heatsink kept the GPU substantially cooler. However you can expect higher system temperatures with the Asus card in a traditional tower case as it does not exhaust air out the back like the reference cooler."

A 1900rpm fan is unlikely to be described as "very quiet" by SPRC or measured as indicated. The article also states that the cards idled at the hard limit of 22% of max rpm. Since we know that the max rpm of the card is ~4200rpm, this would result in the expected idle of ~1000rpm. This is consistent with every other 6850 using a similar heatsink. Given the purported superiority of the DirectCU cooler I doubt Asus would deviate from the norm, let alone so substantially.

According to that article, the Asus card idled at 1560 RPM (and that's at 38C, so it may be the actual hard limit).

I think you will find that the 22% refers to PWM duty cycle, not to maximum rpm. What speed a PWM fan actually runs at at any point in the duty cycle depends on the fan profile, which typically is not entirely linear. While some duty cycles do go down to 0% this will normally be the minimum fan speed, it won't actually turn off. It is quite common for minimum or minimal fan speed to be held for the lowest part of the duty cycle, around 0 to 20% and for fan speed to then rise in line with increasing temperature to 100%. This helps slightly with fan noise at idle, but apart from being a major source of heat itself the graphics card chip is also influenced by the temperature of its intake air. So the balance can be between feeding cooler air to the graphics card e.g. by using a case with a side fan, which means that the gpu fan will run more quietly against the potential increase in noise than a side fan can generate.

Nice catch. But it doesn't explain why Givorg's card will not go below 1900rpm.




Correct. Your point would account for the higher idle of the Asus fan. I was working from assumptions which appear to have been incorrect in this instance.

As I posted on another site (of course: TRY AT YOUR OWN RISK):

I also have an Asus EAH6850 which stays sufficiently cool idle that I would like to change the idle fan speed (now PWM 22%, 2049 RPM minimum)

I tried changing the lower PMW fan setting from 22% to 15% (Duty cycle min %) using RBE version 1.27 (this was the only change)

Then flashed the card by booting to dos and using atiflash version 3.84:

- First:

atiflash -unlockrom 0

(0 = the number of my card, this is followed by the reply "ok")

- Second

atiflash -f -p 0 6850m15.bin

(followed by a successful flash)

Then reboot, according to RBE the value was changed, but the lower fan limit is still 22% (also using MSI afterburner and a custom fan profile)

Now back on the original BIOS, left rom unlocked.

Any idea about the reason? Override by the ATI driver? (Catalyst 11.1a, jan26 release)

(also my idle clocks are 300 300, should that not be 100 300? also according to the BIOS, a bit strange)

Any ideas?

Sounds like an Asus quality control issue or you 2 happen to share the same defect (b/c we know from the SPCR review that the min rpm is lower than what both of you are experiencing). If I understand correctly, flashing the bios did not enable a lower rpm. It doesn't seem like an ATI driver issue given that you're able to adjust the rpm's no matter the ATI Catalyst setting.

Do you have multi-display enabled? When using multiple monitors (or perhaps even having it enabled....I haven't tested either way), the card's downclock is less (300 300, iirc). I believe the standard single monitor downclock is 100/100 or 100/150 (I'm not in front of the PC with the 6850 at the moment).

Allwinner, thank you so much for your help. I've tried all four of your suggestions, but unfortunately they didn't work. Even without Asus SmartDoctor installed, the fan was ~1900 at idle. Confirmed by MSI Afterburner tachometer and GPU-Z sensors tab. Also, tried clean windows install again and used the current ASUS drivers instead of latest from AMD. Still the same.

So I've RMA'd it back to Newegg for a replacement, should be here sometime next week. Even if it would be ~1500RPM like SPCR review, I would be happy since dBA measurements at those speeds were nice. Will update once I receive it. Hopefully its just a bad sample or bad batch, and it won't be a gamble for others who are considering this card.

Thanks for the follow-up. I look forward to learning whether the replacement is any different.

Does it have to be a replacement? Can you order a different card instead? The Sapphire HD 6850 idles at 1100 RPM and 11dBa, so it's even better - and, as far as I know, there were no bad batches.

No, it can anything. How reliable is that 11dB measurement? Do you have a link to a review or something?

Here it is:

Das neue Element? NVIDIA GeForce GTX 560 TI

It's a different website, so the results aren't directly comparable, but the Sapphire HD 6850 is the quietest at idle among many cards.

Hey.
Having the same issues with the Asus DirectCu as you have discussed in the last few posts.
At 20% (and 25% actually) the fan wont go below 1800 RPM. Turning it further down in MSI Afterburner has no effect.

Doesn't sound like a rare occurrence at least.

Sounds like a card to avoid for those looking for near silent idle operation. Even if you one receives a unit that lowers the fan to ~1500rpm (as indicated in the SPCR review), that's still going to be significantly louder than the 1000rpm Sapphire (as indicated by my own testing as well as other posters in this thread).