[VRZ] Nvidia's Green Light Program Overclocking Limitations - Origins and Implications - Page 17

I'm just pissed they released their next-gen core as the 8800gt which was faster than the 8800gts/gtx (which broke the gs->gt->gts->gtx scheme), which caused them to release a second, g92-based 8800gts. It was confusing and deceptive. Then the 9 series was released with the same core, people bought 9800gt cards to replace their 8800gt's and discovered it was the same card. So we had different cards with the same name, and the same card with different names. Etc. All because ATi released the HD 3870. It was a knee-jerk reaction to put a next-gen core out before the marketing department figured out they should replace the 8 with a 9.

We, as computer enthusiasts, live and breathe this stuff, and even we were getting confused/arguing about which card was actually which. Imagine how lost (and open to being abused) the average consumer was.

I stand corrected My point remains, I feel that Nvidia tried to save money in the wrong place with 570 and 590 and it backfired. Now they took voltage adjustment out of Kepler cards in fear that their VRM design isn't robust enough for overvoltage. I don't believe for a second that 28nm process suddenly made voltage adjustment unsafe, there hasn't been reports of massive amount of AMD 7xxx cards failing.

Especially when they believe the BS reason nVidia gave despite it clearly being wrong.

And yours isn't? You have no proof beyond nVidia's PR whereas at least mine can be proven by looking at part numbers.
Now, Kepler uses 228w total power consumption assuming Techpowerups card only results are correct, as you can quite clearly see here, it has a 4 phase VRM (Count the R22 parts, those big grey things ), now tell me..how many good VRM systems for something that uses 200w+ of power only use 4 phases? I'm not a VRM circuitry professional but the only similar looking things are the eR33 parts near the top which are probably for the memory and auxiliary equipment..which means at best it's 4+2 phase. If I could read the actual part numbers, I'd definitely look them up and see exactly how much current/wattage each phase is capable of to see if its above adequate, you can also see that nVidia was definitely testing a 5th phase with the empty spot in the middle.

Erm, weren't Lightnings and the like with Voltage control hitting 1.3Ghz/1.4Ghz stable with air? That's significantly higher than the 1.2Ghz average for most 680s. Also, they're using the exact same process as the HD7k series which can happily have 1.25v 24/7, nVidia specifies 1.175v max..Why is it so low?

Yes, having a bad VRM circuitry on the 570/590 doesn't mean it's going to definitely happen on the 6x0 series...But considering that it was clearly designed to cheapen costs shows nVidia is (Obviously) concerned about board costs, as do many parts of the 6x0 series. (eg. 670 board size, using GK104 instead of GK100 as the flagship GPU, etc)

Exactly, nVidia cut part costs to a level which should be enough but couldn't handle certain problems such as bad batches of VRMs which is what actually bit the 570 and 590 in their rears, this is an exact

*applause*

If they had the solder stencil and everything set up for 5 phases, it's apparent that the decision to move to 4 phases was very late in the design cycle. Do we have any beta hardware photos? The card was obviously designed to run with 5 phases, but it was gimped likely as a cost saving measure.

wow.. you really honor your name

Do you honestly buy expensive motherboard or PSUs because of longevity? Decent VRM's on a motherboard will help with overclocking, because intel CPUs don't appear to be quite so susceptible to electromigration. Not to mention all the features of a decent mobo, like extra PCI lanes or SATA controllers.

Because electromigration occurs over time, and because you don't have access to NV's RMA statistics.

If there was not a high RMA rate (or a prediction of a high number of future RMAs), NV would not have taken action. It's a business, and taking action that will knowingly hurt sales for zero net gain is simply illogical.

Quite simply I believe them, because they have absolutely nothing to gain by lying.
Reference 680: 1.215v
Reference 7970: 1.175v

Might be the reason.

But if you can find evidence contradicting the NVidia spokesperson, please, be my guest. Show it. I will ask you though, what the hell do NVidia have to gain by lying? If it was simply cheap VRMs, they could sell a superclocked edition with decent VRMs. Or they could just say "yeah it's cheap VRMs", 99% of consumers wouldn't give a crap anyway because they don't know the difference between that and electromigation.
What do the GTX 570 or 590 have to do with the 680? All I said is that the reference 680 isn't a flawed or "gimped" design, it's a perfectly good design that's already been pushed to it's limits.

And don't get me wrong, I'm not defending the card, it's about on par with AMD's current generation. Maybe a little worse. I just disagree with the stereotypical angry internet swarm, who tend to be so quick to assume random corporate conspiracy and injustice is the cause of all issues.

I want to know this too.

Actually, yes, I'm going to spend extra on an 80 Plus Platinum PSU because I know that I'll still be able to use it in 8 years and it ends up way cheaper. Motherboards? Not so much, but it definitely factors into my motherboard choice. (Hence why when I reassemble my main rig, I'll be running a Gigabyte 990FXA-UD3 instead of the ASRock.)
Go on, buy a Shaw 860w PSU, it's cheaper and still has the same wattage, plus I know of someone who ran one quite well with a i7 920 and HD5870, quality be damned, right?

Electromigration has zero to do with it, life-span does, if you get two 850w PSUs, one with ultra high-quality components and one with merely adequate components the adequate one will fail much sooner, ever heard of redundancy? The GTX 680 uses ~230w of power, if nVidia has only designed a VRM rated up to 250w then in 4-5 years that will be really pushing the VRMs to power the GPU at stock speeds, I prefer to keep spare GPUs rather than buying cheap low-end ones all the time.

Yes, it does, and you get degradation at the same time, the chip doesn't just die one day, if you could OV a 680 and it was affected by electromigration, then the stable 1375Mhz OC might go to 1360Mhz, then down to 1325Mhz...Why aren't I seeing reports of that on 680s that are OVed?
Erm, yes, because if nVidia came out and said "Yeah, we're gimping board designs to save a few bucks" to their main market who also have a decent chance of understanding that this will negatively affect OCing is going to do nothing but benefit them...nVidia has also lied plenty of times in the past, they certainly do have stuff to gain from this.

Performance gains for the next generation, if Kepler could hit 1.3-1.4Ghz then it'd be no doubt very able to match or beat the newer products..A lot of people saw this with the GTX 460 vs the 560.
Also, nVidia's stance at first was that it's fine to make new designs that OV but not to allow software OVing of reference designs, hence why the Lightning and Classified supported it in the first place..Why did they change this randomly? There's definitely electromigration issues (albeit way overblown by a lot of people, much like how everyone thought that AMD 45nm CPUs were only safe up to 1.425v for the longest time until so many people ran 1.55v 24/7 that it became obvious they could) but it's also an artificial limitation to help nVidia make the GTX 780 look more compelling.

Nothing, but they are an example of what happens when you gimp on VRM designs too much.
I usually tend to disagree with them, however nVidia is nothing but the teenage girl of computer companies. (Remember Woodscrewgate? PhysX in 3DMark Vantage? GeForce FX failing so hard they started to do driver optimisations that reduced game quality so they could have higher performance? Not that AMD/ATI is innocent of that.)

Hehe, I guess some people think further ahead than me then I actually got a little annoyed my last rig lived as long as it did, but then I generally only have one or two builds running at the same time

Overvolted 680s have only been around, what, 8 months? We also haven't seen many reports of VRMs failing on them, my point is only that it looks like NV is expecting this to become an issue in the future.

It does not make sense for NVidia to claim there is an issue with their product, if there isn't an issue with their product. Whether they say it's electromigration, or blame it on manufacturers pushing crap VRMs too far, makes little difference. The backlash is always going to be the same.

(Also, on the overvolted boards, aren't the VRMs custom chips added by the manufacturers? IIRC the classified has a ridiculously high number of phases, yet NV still pulled RMA support for it.)

The only issue I have with this argument is that we see it every generation, every single generation people claim they're artificially holding back to prevent the cards outperforming the next gen. Again, from a sheer business perspective it doesn't make sense unless you have a massive lead on the competition. And NVidia doesn't.

That's kind of my point, why would they change it randomly? It's obviously going to damage their sales, and it was never going to go down well PR-wise, they have no motivation to disallow it unless they believe there is an issue. Again, artificially limiting the cards does not make sense unless they are miles ahead of AMD, and they aren't.


Hehe, the FX range. That takes me back, I seem to remember flashing a 5900 into a 5950u, and it still sucked in dx9 games

I reuse old parts all of the time, so longevity is important to me and the entire second-hand parts market.

Long enough to show electromigration, if the voltages eVGA says are safe (1.3v) are actually unsafe you'd see signs of it by now, they're trying to hide that there is one at all hence the electromigration, every enthusiasts that it's practically as set in stone as a law of physics..you can't beat it, whereas crappy VRMs can be improved.

They are on the OV capable boards, nVidia's original idea was that you had to have a custom PCB/parts and no software OVing..Then that changed randomly.

nVidia technically does, the GTX 680 would originally have been a GTX 660, I doubt GK110 will be in the 780 but they could have always bought out GK110 for that extra performance they'd need to beat AMDs cards in performance, and I have no idea, I'm just stating facts, nVidia's current position is right here in this thread and their original position is somewhere in Anandtechs review of the 680 Classy.

They're within pissing distance of each other, plus nVidia's main market is Tesla, Tegra and Quadro these days.