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

[Legend]

I expect the x80 Pascal card to be shown at GTC, they will tell us that they will be available within 8 weeks which will bring us up to just before computex. Then over the next few months there will be a few of the smaller Pascal based cards, along with mobile parts. Come October-December time the big Pascal part will make an appearance in the Titan form, it will be mind bendingly fast and cost two arms and two legs, probably with 32GB of HBM2, then in March/April the consumer big Pascal will make an appearance with 16GB HBM2.

I have been thinking the very same myself, and from a business perspective it makes perfect sense for Nvidia to do just this. I really can't get my head around why anyone would think they'll come flying out the gate with an all singing and dancing 16GB HBM2 GPU for £500-600!! HA! Just hilarious. Will. Not. Happen.

 

[AllBodies]

I mean, if you want to be really comprehensive about the performance comparison, we should look at an average over lots of games, and with mature drivers (so it's mature vs mature, instead of mature Fermi vs launch Kepler) so:

From here

So we're talking a 20% performance increase over full-fat Fermi from cut-down 'medium' die Kepler.

Also there are two companies competing. So maybe if one can't manage 980 Ti performance for around £300, perhaps the other can

Also also, didn't GK104 have its compute cut down and was gaming focused? So there's no reason GP104 won't be the same, and so won't lose relative performance over the 980 Ti in that way.

 

[lumik]

So add this all up, Pascal isn't nearly as big a architecture change as Kepler was to Fermi, not by a mile. It will have to add things back in stripped out vs Maxwell cards, this will negatively impact overall gaming performance increase as it will take up die space. imho.

how do you know this?We know exactly nothing about pascal.
I expecting GP104 should be around 300-320mm2.16nm have 2x density of 28nm.
TITANX is now around 35% faster than GTX980.So they need increase performance vs GTX980 only by 60% to achieve performance of the new 1080 30% above TITANX.
Thats really not that hard if you have new Node with 2x density.

 

[Smrtka]

I agree if you increase >>>desnity<<< (as mentioned above) you get more performance per nm...math guys math

 

[bru]

http://www.anandtech.com/show/5818/nvidia-geforce-gtx-670-review-feat-evga/5

Aside from the fact that there are loads of cases where the 670 wasn't nearly as far ahead as you believe, you're all comparing very different situations.

The 580gtx had nearly no architectural improvement on the 480gtx. Nvidia spent a year trying to fix the 480gtx, it always was a mostly basic respin. The 670 brought a very large, and long coming architecture update from Fermi. This time around Maxwell is that architecture boost, it also had the hardware scheduler and compute stripped out... the new parts on 16nm will have both of those added back in and won't be nearly as big as an architectural shift. So the chances of gaining the same amount as effectively a 'full' 480gtx to a 670gtx, and a 980ti to a xx70 is nearly non existent.

That is before we get to the fact that people are basing their assumptions on the 970 price and going with £300.

680gtx, $500/£420 actual price(though on average at launch more like £450), 670gtx $400/£330. 980 $550/cba'd to check £, 970 $329/£275 ish for some apparently.

So lets ignore that a 670gtx wasn't actually £300, let alone £270, it was a good chunk more on a MUCH cheaper process. The 970 was cheaper because it was a much more heavily cut down part that performed way worse in comparison to the fullfat part than the 670 did.

670 was about 10% tmu/shader reduction, same rops, same memory bus as the 680. The 970 had an almost 20% drop in tmu/shaders wile rops were almost 15% lower and don't get me started on the memory bus.

The actual price for the full fat part went UP, the 970 was cheaper because it was much more significantly cut down.


So add this all up, Pascal isn't nearly as big a architecture change as Kepler was to Fermi, not by a mile. It will have to add things back in stripped out vs Maxwell cards, this will negatively impact overall gaming performance increase as it will take up die space. The 670 was £330, not £300 and the 16nm finfet process is significantly more expensive than 28nm planar.

Not 100% sure on your pricing there DM. but interesting all the same.

Here is the 670 just 10 days after launch.

Spoiler

My 970 just 3 days after launch

Spoiler

Expect $550-650 for the new xx80 part and $425-500 for the xx70 part, imho.

Of course that is entirely possible, personally I think your wrong, just as I expect personally you think I am wrong. The end result will probably be somewhere in the middle, with one of us thinking, 'well that could have been better' and the other thinking, 'well that is better than I expected'.

 

[Dicehunter]

I'm going to wait for the GP100 full fat Pascal part myself.

 

[Jono8]

Expect $550-650 for the new xx80 part and $425-500 for the xx70 part, imho.

and what performance do you expect from both?

 

[Mauller]

1080 rumoured to be a touch slower than 980ti. But Pascal should not be as overclockable as Maxwell due to readdition of DP. But depending on how they do DP there could be dark silicon which still sucks power or they use some kind of shader fusion method like AMD does with GCN.

 

[Perfect_Chaos]

1080 rumoured to be a touch slower than 980ti. But Pascal should not be as overclockable as Maxwell due to readdition of DP. But depending on how they do DP there could be dark silicon which still sucks power or they use some kind of shader fusion method like AMD does with GCN.

If that is even close to being true then the 1080Ti is going to have to be something special, otherwise people that already have a 980Ti won't see a point in upgrading. So maybe 1080Ti would be about 20-30% faster than the 980Ti? I think almost everyone was expecting more than that.

 

[Ayahuasca]

It'll be interesting seeing comparisons against 1500mhz clocked 980ti's.

Unless full fat Pascal has a feature set I need or a big leap over 980ti/TX I'll be waiting till the next gen or when nVidia gimp us maxwell users

 

[Ash0787]

Ok lets assume the new 1080 card is a little below a 980Ti in performance
and costs over 400, why would you not buy a 980Ti instead ( which should fall below £500 soon unless nvidia artificially inflates the price )
so unless people really care about power consumption these cards will need to be relatively cheap ( and if they did care about power draw they are still getting screwed because no HBM2 )

 

[FoxEye]

1080 rumoured to be a touch slower than 980ti. But Pascal should not be as overclockable as Maxwell due to readdition of DP. But depending on how they do DP there could be dark silicon which still sucks power or they use some kind of shader fusion method like AMD does with GCN.

Sounds more realistic, but what's the source for this rumour?

 

[Rroff]

Sounds more realistic, but what's the source for this rumour?

Even taking into account adding some stuff back in it doesn't make sense.

Cost could be another issue as unlike in the past there are a few factors like the size of the transistors/pitches that will tend to make the first generation(s) of products a lot less cost effective to produce than 28nm.

 

[FoxEye]

Even taking into account adding some stuff back in it doesn't make sense.

It only doesn't make sense if you have competition pushing you all the way.

Look at Intel CPUs and their 5% more perf each gen.

If the decision punters have is between a £400 1080 and a slightly better performing £500 980ti, nV are laughing all the way to the bank. Whichever one is better value, they get paid.

That only stops working if AMD decide to launch a 980ti perf card for £300. And how likely is that? With all the noises AMD have made about being a "premium brand" now.

e: Although personally I was expecting the 1080 non-ti to be a little better than a 980ti, not a little worse. I was expecting the 1070 to be somewhere between the 980 and the 980ti.

 

[Simon Belmont]

If the GTX1080 is slower the 980Ti there would be no point to even buying it? Just get a 980ti instead. I hope this isn't true?

 

[Rroff]

If the GTX1080 is slower the 980Ti there would be no point to even buying it? Just get a 980ti instead. I hope this isn't true?

Guess it depends if you are talking a scenario where they don't release a ti in the initial lineup and the "1080" is the fastest card in round 1, or a scenario where there is a ti and the "1080" is more in the 2nd place type position.

 

[Mauller]

If the GTX1080 is slower the 980Ti there would be no point to even buying it? Just get a 980ti instead. I hope this isn't true?

Maybe not for people with 980ti, but a far better deal for people with cards slower than a 980.

Depending on how they do it, the card will either be priced between the 980 - 980ti, or it will replace the 970. just my speculation when coupled with mid polaris performance and price rumours. If amd really do refresh their entire lineup and replace everything with mid polaris for around the $300 mark, with mid polaris having fury-980ti performance then Nvidia are going to have problems.

And the 1080 rumour is from Benchlife, a chinese source etc.

 

[Rroff]

And the 1080 rumour is from Benchlife, a chinese source etc.

I get a funny feeling that they've come up with some partial information about mobile Pascal and extrapolating the rest to desktop functionality.

 

[bru]

I really don't understand how people think that AMD can get to 980ti/Furyx performance with the midrange part but NVidia cannot. both sides will have not too dissimilar performance, as is the usual state of affairs.

 

[D.P.]

I really don't understand how people think that AMD can get to 980ti/Furyx performance with the midrange part but NVidia cannot. both sides will have not too dissimilar performance, as is the usual state of affairs.

Their love of AMD prevents any kind.of logic. According to the logic of certain posters the 1080 will be hardly any faster than a 980 and almost as expensive as a 980ti. Ignore them.