[eteknix] Gigabyte GTX 680 Superoverclock - Page 3

1) Because there are a lot of different factors determining overall power draw of the card. Inefficent design of the PCB might be one of them, the types of components used might be another. Maybe they found ways to reduce leaks (heat, ect..) in the non-reference card that were not available with the reference design.

You can also get into the ideas that they might also be lowering the volts used on the non-reference. You don't have to use the reference design's specs, that's the exact reason why they have non-reference. Maybe they've found ways of reducing the amount of volts used, a more stable delivery system, who knows (obviously somebody does, and I don't personally care to). Not to mention, what they report as TDP on a reference card isn't really the actual TDP. Just like CPUs, the manufacturer lists a max TDP that's probably well above what the actual chip TDP is. That's the highest TDP (probably an average) that they have gotten in testing, which means you could get a chip that has that TDP or you could (gasp) get a chip with a lower TDP. The same applies to graphics cards, or any other fabricated process involving micro processors (or similar technology).

let me put it this way, if you have a 5+1 phase, 8+2 phase, or 16+2 phase power supply on your MB for your nice i7, does that change the TDP of the chip? Does your i7 give off more or less heat with each power phase? No, it doesn't. It uses the same amount of power. Now, is there a possibility that one of those phases are going to use more power than the others? If the 16+2 is hotter than the 8+2, chances are your using more power. =O NO FREAKING WAY. Now that's not really why you use more phases, generally they end up being cooler as the load is split more evenly amongst the different phases. Either way, your chip/cpu/whatever will use the same amount of power regardless of how the power is being delivered. It requires that amount of power, it was designed to use that amount of power. Due to the fab process not being perfect, one chip/gpu/whatever might actually use less or more power than the other. Why do you think they cripple cards to make low end variants? It's probably because the card couldn't meet specs so in order to save it they cripple the worst performing part. Re-test, then label it as a lower spec design. Bam, now you meet specs.

2) That's a TERRIBLE way to measure power draw, you have no idea what the overall power draw of the card is. Your chip might be using a certain amount of power, but you have no idea if there are any areas in which power is being wasted. Your best bet to actually figure out how much power is being used is to get a volt meter and record what you are using.

Either way, your statement about how TDP reflects the condition of the VRMs being used is so terribly convoluted I can't even describe how bad it is. In my example with your i7, the better phases doesn't change the TDP of your processor. If your proc uses X amount of power during load, no matter WHAT type of system is being used to deliver that power your chip will ALWAYS use the same amount of power. ALWAYS, ALWAYS, ALWAYS. There is no exception, there will be no exception to that rule (outside of what I've already talked about).

I'll quote you whenever I freaking choose, cause obviously you need some assistance.

I dont care what the best way of measuring is, all i care is with no reference cards it stays at 80-90% with huge overclocks, 1300 etc, yet it goes to even 110% on non reference 680's like np on STOCK.

The most logical part is that the power delivery system - the vrm is far more efficent, which allow Asus lets say to make cards at 1200mhz/1300 boost without going over Nvidia strict rules about TDP of even non reference cards.

You can't deny that by saying they did something to pcb, the VRM is key and you know it

Didn't Nvidia said it will just void the warranty even if it is factory overclocked over 1150Mhz?

no

NO ITS NOT. There's oodles of ways to make efficiency better than just the VRM. You can shorten traces, find ways to use less traces. You could find ways to cut out caps, or use better caps (all just examples, there are TONS of things to do). There's so much crap you can re-design on a PCB to try and cut out parts of a design it's crazy. Not to mention, there is nowhere on the article that mentions TDP. Even so, the actual chips that are fabbed don't come from the PCB designers. Gigabyte nor Asus fab chips, which the fabrication process is what determines TDP and not the PCB or any component on the PCB. So I'm guessing none of the cards your thinking of even touch the offical TDP of the chip. (Because power draw =! TDP of a chip, that's JUST THE CHIPS HEAT, WHICH IS RELATED TO JUST THE CHIPS POWER DRAW NOTHING ELSE ON THE PCB).

The power draw is a direct relation to the speeds of the chip. There is no going around that, go ask ducky if you want. He's going to tell you that the power draw of the chip is directly related to the speed in which the chip is running. Now each chip may have a different operating range, that's because of the fabrication process which I said earlier. Once you get the standard range of the chip at "factory" settings, every speed increase above that is going to give a set thermal increase (TDP). As well as it's going to draw more power which will also be directly related to speed. Those are absolutes, there is no "DA VRM IS CAUSING DA PROBLEM". The power draw from the VRMs is going to be minimal compared to the power draw from the chip as you increase speed. Sorry mate, that's just how it goes.

[edit] Hell, overall the power draw from the PCB components total is going to pale in comparison to the power draw of the chip. When I mean pale, it's not even worth measuring or taking note of. You don't adjust power consumption numbers for the PCB, because it's basically nothing when you look at the power consumption of the chip/die (mem might be worth noting, still pales compared to the die).
Edited by mushroomboy - 8/6/12 at 4:12pm

We saw the preview of this card awhile back. The little mini fans are probably cheap. I'd throw it on water anyway, but if it on air, I'd be wary.

Uhh....

I just looked at the charts and such, being late I didn't feel like reading it. But um, why didn't they benchmark at 2560x1400 or similar?

You'd think that if they are going to take up three pci-e slots, then they could of at least provided more outputs for the user. Three dual-link dvi would have been good.