Gigabyte GTX 1080 G1 Gaming 8 GB

[HumanSmoke]

1200$ for the card in Japan... People have no common sense anymore, nor respect for the buyers...

Amazon has always preyed on consumers with poor impulse control. Hardly newsworthy at this point - it's like pointing out that eBay is infested with opportunists.

 

[N3M3515]

This 1080 and 1070 seem to be poor overclockers.......
Gigabyte G1 and MSI Gaming versions of the 980Ti where much faster than the stock and overclocked way better.

 

[Assimilator]

Could be a revisiting of Maxwell. A rather tight spread of OC range. Jacob Freeman at EVGA said that 2200MHz was the target of their Classified card, with hopefully options for a more relaxed sub-zero BIOS option. The latter I presume will require some extensive testing - but at least it is underway

Pascal is very much feeling like Maxwell 3 with higher % base clocks and resultant lower % OC headroom.

 

[Dethroy]

Review results from all over are pointing to the same thing: Pascal is limited to 2.1GHz on the core, regardless of how many extra power connectors and other bling you strap on. I wonder if this is a limitation of the new node, or Boost 3.0, or something else entirely?

From what I've seen the problem seems to be the voltage limitations of the nvidia design as its locked at 1.25v i believe. Should be fixed up with some bios's/Galax HOF style cards i hope, otherwise its going to need a hardmod to get past that cards are binned well, but voltage holds it back, as GPU boost 3.0 is limited to voltage points, with all chips seemingly capable of around 2100mhz thus far.

So for now all we can do is leave the extreme guys to do their thing and figure out how to break the volt cap, should get things rolling fast i hope.

Probably restricted on purpose to have more headroom for the next round of cards.

 

[trog100]

Pascal is very much feeling like Maxwell 3 with higher % base clocks and resultant lower % OC headroom.

yes the reference clocks are higher than in the past which simply means less overclocking headroom..

some headroom has to be there simply for reliability reasons.. 10% would be a reasonably figure.. assuming the TPU review card isnt a poor example 5% is a little too close to the fall over point for comfort..

pascal basically is a die shrunk maxwell that can clock much higher.. how much user overclocking there is comes down to how close to the fall over point the partners dare go.. too close and the failure rate will be too high.. i recon within 5% is too close..

trog

 

[newconroer]

Oh look, an aftermarket cooler arrangement that out performs Nvidia's reference model.
Imagine that...

 

[TheinsanegamerN]

From what I've seen the problem seems to be the voltage limitations of the nvidia design as its locked at 1.25v i believe. Should be fixed up with some bios's/Galax HOF style cards i hope, otherwise its going to need a hardmod to get past that cards are binned well, but voltage holds it back, as GPU boost 3.0 is limited to voltage points, with all chips seemingly capable of around 2100mhz thus far.

So for now all we can do is leave the extreme guys to do their thing and figure out how to break the volt cap, should get things rolling fast i hope.

Even if we get past the 1.25 volt barrier, what of the voltage limits for 16nm? Can a process that small take 1.3v+ for any extended amounts fo time and remain stable, even under water? That seems to be pushing it.

It couls also be we are just seeing the limit of the arch itself. volta may be needed to hit above 2 Ghz reliably. Especially after seeing two third party cards hit the exact same OC, that seems like a arch limitation.

Custom BIOSes may prove this wrong, but I feel if nvidia was limiting clock rate in the bios, then somebody would have found that by now.

 

[Vayra86]

Pascal is very much feeling like Maxwell 3 with higher % base clocks and resultant lower % OC headroom.

This is exactly what I've been saying even long before the release of Pascal. Nvidia ate up 'our' OC headroom, and architecturally, the majority of perf gain comes from FinFet, not from 'Pascal' being so great. This started already with Kepler v2, and every iteration after that has been streamlining the GPU for gaming purposes only, allowing higher clocks / more TDP budget for shader power and connected resources.

It's a smart move to 'hide' the lack of a more efficient SMX and the AIB cards bring that lacking improvement to light. Basically we are paying the premium for Nvidia overclocking our cards out of the box, or in other words, we are paying for their implementation of GPU Boost 3.0, while we get less of everything in terms of 'metal' - a tight bus, relatively low amount of shaders, etc. The real gain architecturally has already been implemented in Maxwell with the delta compression that they now just added another layer to.