Power/voltage for video card

Quad core 45W cpu + 65W gpu on a 120W power brick

The gpu refuses to run at full clocks, but runs at low 3d mode. I look at the bios and it's 1V (full clocks) vs 0.85V (low 3d mode). If i lowered the 1V, would that make it run at full clocks? Or do i absolutely need a higher output power supply (i have a 180W coming in the mail)?

 

Let me get this straight...you want to lower the voltage to get higher clocks? Not going to happen most likely.

You're trying to use a 120W PSU for a system that has a 45W CPU, 65W GPU (110W total right now), probably 5-10W for hard drive and motherboard, and probably double digit watts for the LCD screen. It doesn't add up, the power isn't there. Try the 180W PSU.

 

I've seen people undervolt their video cards before, in fact, I got the idea form this thread: http://forum.notebookreview.com/showthread.php?t=401345

The guy lowers the full clock mode to 0.95V on a 260m, which is rated at 75W.

 

What do you mean by it refuses to run at full clocks? are you running GPU-Z while running a game to see if its clocking up correctly, also have you got your power profile set to max performance, as i would imagine regardless of what watts you are feeding it via power supply it would take the extra needed from the battery and just flatten it. Though i could be wrong.

 

True, but it is not guaranteed to work. Plus I'm not a huge fan of undervolting anything.

 

Explain please. I know that decreasing voltage can help reduce heat as well. I would like to know why you dislike it.

 

Undervolting takes a LOT of burn-in and testing and many times just won't work as soon as you start actually loading the chip that's undervolted, and if you do it wrong, it can make your machine unbootable. I personally undervolt my laptop CPU because I can squeeze an extra 20 or so minutes on battery out of it, but it's fairly dangerous waters.

 

Come on, cut the fear factor. You're all making it sound more complicated than it really is. Undervolting should be approached the same way as overclocking. We need to test and experiment one variable at a time in small steps. If we screw up, we can literally flash back to square one. As for Darth Bane's situation, I don't know what GPU he is trying to undervolt, but if it is a G92, then he can lower the clocks to say 500/1250/800 and get it to work with a voltage as low as 0.9v. All he would need to do is thoroughly stress test the current set of clocks at a given voltage at a borderline temperature to ensure that it is guaranteed stable. Then he can slowly crank up the clocks to discover the maximum frequency his GPU can operate stably at that given voltage.

 

The GTX 260m in our G51s are underclocked to 500/1250/799, whereas a standard 260m clock is 550/1375/950.

With the undervolt to .90 Extra, we are unable to get back to standard 550/1375/950 clocks.

 

what GPU is it darth bane?

 

IDK, mine doesn't add up either. 55w for CPU (OC'd) and 84w for GPU (OC'd), plus 5w for HDD, 25w for mobo and devices, and the 17-in screen to light up... My PSU is rated at 120w max, but the system still hauls when everything is being maxed.

 

I know it doesn't sound like much of a qualification, but since I've been working at Bestbuy I've almost NEVER seen a power supply on a desktop or laptop that has (as far as its rated) been sufficient to run a gaming machine. They all work fine, they just shouldn't based on the numbers.

Anyway. Yes, undervolting thing MAY allow it to work, and here is why:

Although you would think undervolting wouldn't allow you to go to higher clocks, it just might work:

Those higher 3d clocks will only run at a certain VID (voltage ID) that they are matched to in the vBIOS. If the card jumps up to that clock for gaming (and therefore the matching VID) and the power supply can't supply that VID, then you blackscreen. BUT, IF the card can run full power at a lower voltage (most can, mine runs fine at 10% undervolted) than you can change the VID for high power and it will work. Since it will be using a VID that it can supply.

The only thing you have to do then, is try to run the higher clocks using the VID for low power clocks. (good chance this will work.)

I can't understand why everbody isn't undervolting their GPU and processor (this will cut power and heat, increasing battery life). Undervolting has cut 10C off my processor and 14C off my GPU. (look at my sig)

And NO, you wont ever brick your card unless you're a total idiot, because you are usually able to modify only the VIDs that coorespond to 3d gaming, NOT BOOTING, which has its own unique VID. So just change the high power 3d VID and then the worst that can happen is you blackscreen in high power 3d, and you can always reboot and reflash...

Here is a picture of ATI bios editors GPU voltage editor for my 4650:



BTW the reason why the math so rarely makes sense is that manufacturers want lots of margin for error. Many periferals take lots of power (external HDs, DVDs, Webcams, Lights, Ipods (when your charging 'em) etc. These all have to be supplied with power too.

 

Regarding the GTX 260M undervolting, only some of us are able to get it running stable at 0.90v even with the reduced clocks compared to the clocks in NP8662s.

We believe that 0.90v is basically the very lowest that you can go with 500/1250/900 clocks and still be stable.

 

Give me a second to explain my position here...keep in mind that I am a trained electrical engineer and I do have some background in computer and silicon design.

These chips are designed to run at a specific speed, at a specific voltage, and for the most part these values are chosen based on performance desires, power requirements, other design decisions, and they factor in manufacturing variability. When they said that not all chips overclock or underclock the same, that is 100% true.

The voltages and speeds that the chips are chosen to run at (yes, chosen) are chosen because they have proved to be reliable on that silicon manufacturing technology and that particular chip design. Each transistor might be a slightly different size, use a little more or a little less power, or run faster or slower (especially slower) just because of these manufacturing variances.

There are literally thousands, if not millions, of signal paths and transistor connections in a single processor. An entire power distribution network in the chip that gives all these chips their power. And while it is an x86 processor alright, it is Intel's design solution for it or AMD's design solution or nVidia's design solution for it. They know information about the internal workings of the chip, they know how to best test the chips, and they spend a huge sum of money verifying the operation of these chips through hundreds or thousands of tests that are run on these chips during the manufacturing process. They also know now to weed out the bad chips. After all, in the consumer market 100% of the product needs to actually compute correctly 100% of the time without exceeding various specifications assuming that the chip is functioning.

We change the voltage, or speed, of the processor and you introduce all of those unknowns back into the equation. Only you do not have all the knowledge needed to verify the operation of the chip again, and you cannot necessarily trust someones results because each and every chip is definitely different. Sure, run Prime95...it works for 24 hours perfectly fine and you think the processor will run at 1.4V and 3.6GHz. Run 3dMark06 for 24 hours and you think the GPU will run with an extra 100Mhz. No, you know the math and processing units that Prime95 exercised work alright...what about the rest of the chip, the cache, the floating point units, the multimedia computational units, etc, etc? GPUs contain even more transistors than processors, and they can have many more units that were not tested in your benchmarking.

Bottom line, these chips need to compute correctly 100% of the time. Without all the knowledge about the chip that only Intel or AMD or nVidia has, without all the test routines, you cannot verify the chip works 100% of the time. Maybe you can verify 99.9999% of the chip, but when billions of things can be calculated in a single second...it isn't 100%.

I prefer reliability.

 

99.99% is fine with me I will just wait for my 180w and see what happens.

 

I understand what you're saying Greg, and whole-heartedly agree.

One thing you've left out though (or maybe you hadn't considered), is that these cards are mass produced. Upping the voltage passed the neccessary level is one way to decrease the amount of "dud" GPUs, since manufacturing defects can increase the required voltage for some cores. A voltage value is picked that is beyond what most cards need for partly this reason.

I guess I'm counting on the fact that manufacturers always leave a margin for error. I cannot resist the urge to exploit it to the fullest.

 

Update/Full Story:

The laptop in question is the MSI GX720, originally comes with 9600m gt. It uses the same motherboard found in the msi gt725, which uses the ati 4850. The 9800m gt I have has an extra tab (i think it's called "he") that the pci-express slot does not have room for. Anyways, even with a 180W power adapter, the card refuses to run at full clocks and any attempts to OC does not work (via bios flashing). What the hell is wrong.

 

I couldn't agree more; undervolting is essentially the opposite of overclocking, but without risking the hardware.

The challenge in undervolting the GPU specifically is that you must essentially edit, save, and flash a new vBIOS with your undervolting values; this is totally different from traditional undervolting because an unstable voltage results in a system crash which you can never recover from. What do you think happens if the vBIOS voltages are set too low and your GPU becomes unstable? Your system becomes worthless, that's what.

 

I've killed dozens of cards from undervolting in my line of work. It's probably takes no more than a minute to bring it back to life.

 

here's screenshots if anyone wants to help:
http://forum.notebookreview.com/showpost.php?p=5253594&postcount=1132