** The Official Nvidia GeForce 'Pascal' Thread - for general gossip and discussions **

[Rroff]

I am still wondering if Pascal is a coaxed die shrink of Maxwell and as such found the following snippet interesting.


Taken from Here’s How NVIDIA’s Simultaneous Multi-Projection Can Benefit 4K Gaming.

Its not a completely new architecture - but at the same time it would be wrong to write it off as just Maxwell with a die shrink.

 

[rene.agac]

I may be wrong but I really think we are looking at...
all-unlocked-Core2- series into locked 2500 and more expensive unlocked 2500k
... market segmentation again. This time with GPU. They arent going to bin their chips come on. They gonna lock it. And charge premium for unlocked ones. Worked for intel.

 

[Ayahuasca]

Nope

 

[pmc25]

I disagree with your opinion/pure speculation .
1070 and 1080 are still early cards. It is a different arch I believe. The competition will "probably" do no better unless they have low earning employees who are just much better than those of nVidia. Why do I say that?Because I doubt they have had the budget to spend.
The higher clocks probably means for nvida they can for the early cards make a bit more £ for a performance improvement, but that's me speculating (better performance from less cores via higher clock speed)

Pascal was hoped to fix many of the glaring holes of Maxwell and move to a more parallel design, per GCN. It categorically hasn't, and it seems that the bits & chips guy who was saying over a year ago that Pascal was FINFET Maxwell with re-added DP units for compute cards was completely right.

Pascal was a rush job, which everyone knows, but far more was expected of it. With those kind of clocks, there ought to be a far greater increase in performance than there is, with such a colossal improvement in process tech.

 

[pmc25]

What you just said is exactly the same for the Polaris 10 ( only fury x performance at best on release ) yet the 1080 has more than 980 Ti performance. So 1080 for me unless Amd have secrets up thier sleeves.

The Polaris 10 is a MUCH smaller chip, with a host of architectural improvements and new features.

 

[AlamoX]

Don't think they've said anything definite but seemed to be the theme of that bit of the presentation that those cards would be capable of hitting a certain level of performance and the non-FE cards would be more random.

they dont have to say much more than one 8pin for a 180TDP card, this feels like intel's K variant, you want 2x8pin to overclock pay extra £100, i think this is going the wrong way and might set a bad precedent, if we get actual hardware limit to GPU overclocking.

 

[chaparral]

It also makes the card more efficient at VR rendering (which is far less forgiving about low frame-rates).
Huang says the 1080 is twice as fast as the Titan X when using the multi-projection feature to render VR.

I know nothing about VR
So does this mean the GTX 1080 will give twice the amount of FPS compared to a Titan X when using a VR headset to play games


Edit:Sorry forgot to add the link http://www.engadget.com/2016/05/06/nvidias-new-gtx-1080-gpu-is-even-faster-than-the-titan-x/

 

[Ayahuasca]

Only in scenarios that use the multi-projection feature.

 

[Orangey]

We should call it GM304.

 

[queamin]

Someone posted on SA that Doom was running at 1080 on a projector because the projectors were only fullhd now that is strange for a card as powerful as a 1080 I wonder if at 4k it would tank because of bandwidth of the bus only being 256 and if so how will Polaris and all new card do the same at 4k?.

If the bus width is smaller than that of the 980ti and titan X and fury's might hold their own at 4k res.

 

[Rroff]

they dont have to say much more than one 8pin for a 180TDP card, this feels like intel's K variant, you want 2x8pin to overclock pay extra £100, i think this is going the wrong way and might set a bad precedent, if we get actual hardware limit to GPU overclocking.

1x 8-pin + PCI-e supply is comfortably over what the card requires for all but maybe the most extreme overclocking - TDP only loosely reflects the actual supply wattage requirement and is usually based on some overhead in an extreme situation as TDP is usually rated using blocks of around 20-30 watt or something - so the supply is probably capable of around +30-40% over what the card will nominally require.

 

[DarrenM343]

Pascal was a rush job, which everyone knows, but far more was expected of it. With those kind of clocks, there ought to be a far greater increase in performance than there is, with such a colossal improvement in process tech.

Pascal was over 2 years in the making apparently, 7000 people involved and billions of $. As they were first to market with the last gen I expect they probably started on the next gen before the competition did who were already struggling to get something to market.
The competition who have lost market share over the last few years are hardly going to claw that back if they are behind the curve of release and when they do bring something to the table it just able matches in some aspects (give or take a water cooler to help).I'm hoping AMD do spring a surprise this time around but i fear not.

I dunno, the 1080 is what 400 cores down on the TItan x yet faster, apparently by 20%+

 

[Rroff]

Pascal was over 2 years in the making apparently, 7000 people involved and billions of $. As they were first to market with the last gen I expect they probably started on the next gen before the competition did who were already struggling to get something to market

I think they are cheating a bit on the numbers there as they swapped out architectures at some point and probably counting all the work towards the other one as well.

 

[queamin]

I think they are cheating a bit on the numbers there as they swapped out architectures at some point and probably counting all the work towards the other one as well.

Yes more likely adding Maxwell and Pascal together.

 

[eazyrider]

I am still wondering if Pascal is a coaxed die shrink of Maxwell and as such found the following snippet interesting.


Taken from Here’s How NVIDIA’s Simultaneous Multi-Projection Can Benefit 4K Gaming.

I love the comments section on this article

 

[Minstadave]

Someone posted on SA that Doom was running at 1080 on a projector because the projectors were only fullhd now that is strange for a card as powerful as a 1080 I wonder if at 4k it would tank because of bandwidth of the bus only being 256 and if so how will Polaris and all new card do the same at 4k?.

If the bus width is smaller than that of the 980ti and titan X and fury's might hold their own at 4k res.

Bus width is irrevevant if your memory freq is high enough, GDDR5x isn't massively far behind the HBM in the Fury.

 

[queamin]

Hbm is very wide wait until hbm2 and compare when it as 8Gb not 4Gb, and a small bus does matter look a 960 with it cut down bus I am just wondering 256 hampers it at all at 4k, we will not know until benches really.

 

[queamin]

Because if it does matter at higher res then all new cards with 256 bus and lower might struggle at the higher res or might find frames drop of a lot more than a card with wider bus like 980ti.

 

[NirK]

Because if it does matter at higher res then all new cards with 256 bus and lower might struggle at the higher res or might find frames drop of a lot more than a card with wider bus like 980ti.

I think probably the 1070 will bottleneck a lot at high res but the 1080 should be reasonably OK, the 1080 is 320Gb bandwidth the 980ti is 336Gb.

 

[queamin]

Yes I thought it was 256 my wrong.