Sign in with
Sign up | Sign in

MSI Caught Overvolting GTX 660 Ti, 670 Power Edition Cards

By , Benjamin Kraft - Source: Tom's Hardware DE | B 91 comments

If a company changes its product in the middle of a production run after an error becomes public, you know something’s fishy. If Nvidia steps in as well, that’s a good indication there’s something serious going on. Here’s what happened at MSI.

Our colleagues over at Tom’s Hardware Germany seem to have caught MSI red handed, overvolting the GPUs on the GTX 660 Ti and GTX 670 Ti Power Edition boards (Google Translate) to achieve a higher and longer lasting GPU boost state by basically circumventing the PWM controller. In other words, MSI was cheating. Perhaps no one would ever have known if it hadn’t been for one side effect. The increased voltage can cause the system to refuse to POST.

A quick search revealed that there are a number of threads on message boards where users report such problems, and our initial research led us to conclude that it was indeed these specific models causing trouble. What was obviously intended to provide a competitive benefit in benchmarks is apparently now backfiring.

To verify our findings and analyze their cause, we conducted some of our own research, enlisting the help of our own UK community member starcraftgod who was also affected by the issue. After comparing notes and measurements, we took a closer look at the circuit board itself. What we found there was unexpected to say the least, and it’s almost hard to believe an engineer would create something like this on purpose. In short, the PWM controller on said cards is overvolted by up to 88 percent using a small alteration to its power supply, which leads to the chip running way outside of its specifications and at significantly deviating characteristic lines. This means that too much voltage is being applied to the GPU in general and the voltage reduction under load (droop) is not as large as it should be -- and that's just one side of the story. The other is that MSI’s tinkering can cause serious side effects.


So what’s happening? A small component completely superfluous to the normal circuit in one of the ground connections causes major overvoltage in the PWM chip in question – instead of the 5 volts specified by Richtek, the chip is hit with up to 9.3 volts. That can’t be good for its life expectancy in the long run and also causes the issues we mentioned above. In combination with some PSUs, the system will be unable to start up if the 12 volt rail has to provide more that 12 volts to one of the PowerEdition models (even within the ATX spec). In other cases, MSI’s circuit trick can trigger a black screen or cause a driver crash when a change of load occurs. We find it hard to believe in a design accident here, since the circuit in question is a standard design – if implemented correctly.

MSI got back to us with a very enlightening statement. Here’s an excerpt:

We are currently not aware of any other vendor providing this same level of performance and we’ve worked with NVIDIA to ensure that new production models will limit this free overclock boost you currently get. Our new production models with normal GPU Boost function will be on sale next month.

Reading between the lines, we can gather that Nvidia is none too pleased with MSI going off on their own. The full article on Tom’s Hardware Germany (Google Translate) covers the topic in more depth.


Contact Us for News Tips, Corrections and Feedback

Discuss
Display all 91 comments.
This thread is closed for comments
Top Comments
  • 29 Hide
    zero messiah , October 1, 2012 4:16 PM
    *we’ve worked with NVIDIA to ensure that new production models will limit this FREE overclock boost you currently get*
    On behalf of consumers everwhere, i would like to thank you MSI for trying to make us feel guilty.
    And in the future, i will try not to feel entitled to having a working graphics card.

    `Sincerely,
    Joe Q Public.

    PS. Burn in hell @ 9.3 volts.
  • 28 Hide
    CaedenV , October 1, 2012 3:38 PM
    lol, 9.3V! That's an awful lot of power for small parts!
  • 28 Hide
    dextermat , October 1, 2012 4:17 PM
    xmas list:

    NOT buying msi video card : check
Other Comments
  • 28 Hide
    CaedenV , October 1, 2012 3:38 PM
    lol, 9.3V! That's an awful lot of power for small parts!
  • 13 Hide
    pjmelect , October 1, 2012 3:41 PM
    With the capacitor in the position shown in the circuit diagram the voltage regulator will not be doing anything and would not be needed. I think that this is a layout error and not deliberate and should have been picked up when the board was tested by the manufacture. It is amazing that the board worked at all with this error.
    It does however raise a more serious point that as the development cycle time reduces between boards being released the testing of the designs is being cut back on.
  • 22 Hide
    Darkerson , October 1, 2012 3:42 PM
    Having used MSI products in the past, I have to say Im very disappointed in them and will probably not be using them again.
  • 26 Hide
    csf60 , October 1, 2012 3:42 PM
    What the hell were they thinking??? Didn't they test these cards before release or what?
  • 24 Hide
    pacioli , October 1, 2012 3:51 PM
    That looks intentional...
  • 4 Hide
    alidan , October 1, 2012 4:03 PM
    so what was the end result of this in a gameing sense?

    not a system wont boot, or card dies sooner, lets assume everything went fine

    what did the card gain?
  • 29 Hide
    zero messiah , October 1, 2012 4:16 PM
    *we’ve worked with NVIDIA to ensure that new production models will limit this FREE overclock boost you currently get*
    On behalf of consumers everwhere, i would like to thank you MSI for trying to make us feel guilty.
    And in the future, i will try not to feel entitled to having a working graphics card.

    `Sincerely,
    Joe Q Public.

    PS. Burn in hell @ 9.3 volts.
  • 28 Hide
    dextermat , October 1, 2012 4:17 PM
    xmas list:

    NOT buying msi video card : check
  • 19 Hide
    proffet , October 1, 2012 4:18 PM
    BUSTED....
    And I have been a MSi TwinFrozr and PE (Power Edition) nVidia models fan in the past.
    I wonder if some legal action is soon to come like a class-action or something.?
    Consumers mad because they blew their hardware up..
  • 28 Hide
    officeguy , October 1, 2012 4:26 PM
    My momma told me if I don't have anything good to say, don't say it at all. So, I am not going to say anything.
  • 16 Hide
    dudewitbow , October 1, 2012 4:27 PM
    proffetBUSTED....And I have been a MSi TwinFrozr and PE (Power Edition) nVidia models in the past.I wonder if some legal action is soon to come.?

    at the least, they will probably have to do what evga did, and ask users with these models to return them and they have to ship them a new gpu(evga users returned an OC version for the FTW version for the 670)
  • 10 Hide
    Pennanen , October 1, 2012 4:31 PM
    Asrocks mobo capacitors, gigabytes pcie 3 mobos and now msi.

    I actually tought msi was one of the more trustworthy gpu companies out there.

    Oh how wrong i was.
  • 11 Hide
    Anonymous , October 1, 2012 4:35 PM
    Sure, MSI won't even include cpu vcore offset on their motherboards, but they'll cook the video cards. LOL
  • 21 Hide
    boiler1990 , October 1, 2012 4:38 PM
    While this wasn't a smart move by MSI, I think this points out a limiting feature of the Kepler GPUs - the voltage caps are very low. I think the 6xx GPUs would OC much better if there were more freedom in voltage tweaking - and I'm guessing that's why AMD's competing cards have better success in OCs
  • 0 Hide
    ricdiculus , October 1, 2012 4:40 PM
    Never cared for MSI since a tried to flash one of there motherboards and it bricked. They sent me a new bios chip and guess what, still bricked. These shananigins don't surprise me.
  • 20 Hide
    bystander , October 1, 2012 4:43 PM
    I don't understand how this was "cheating". They are selling an OC product. This isn't a reference design, so they don't have to adhere to specific clocks.

    However, they definitely screwed up by exceeding engineering specs.
  • 20 Hide
    jasonpwns , October 1, 2012 4:52 PM
    bystanderI don't understand how this was "cheating". They are selling an OC product. This isn't a reference design, so they don't have to adhere to specific clocks.However, they definitely screwed up by exceeding engineering specs.


    This gives nvidia a bad name too. Since if people get a dead card they usually don't care to investigate what's wrong with it and pass the blame straight to nvidia or amd. They usually never point the finger to the company they actually got the card from.
  • 12 Hide
    back_by_demand , October 1, 2012 5:02 PM
    csf60What the hell were they thinking??? Didn't they test these cards before release or what?

    Of course they did, but this reminds me of the scene from Fight Club
    Quote:
    A new car built by my company leaves somewhere traveling at 60 mph. The rear differential locks up. The car crashes and burns with everyone trapped inside. Now, should we initiate a recall? Take the number of vehicles in the field, A, multiply by the probable rate of failure, B, multiply by the average out-of-court settlement, C. A times B times C equals X. If X is less than the cost of a recall, we don't do one.

    Is the risk of getting caught worse than the financial benefits? It depends how many times they have done things like this before and not been caught
  • 4 Hide
    Anonymous , October 1, 2012 5:08 PM
    *sigh* just ordered the 670 P.E. last week... what to do now...
  • -8 Hide
    lamorpa , October 1, 2012 5:09 PM
    "That can’t be good for its life expectancy in the long run"?

    What could this statement mean? Life expectancy is the long run by definition.

    "That can’t be good for its life expectancy in life expectancy."?
Display more comments