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

[andybird123]

It says a 2x density increase. You would expect more than that for a 2 node jump.

I think you're all setting yourselves up for a big disappointment.

I don't think ANY GPU released until say 2017 Q2 is going to be particularly impressive for high-end owners.

65 to 55 wasn't a big jump either, the big changes came at 40nm
In fact looking at it, nvidia did 65nm with the 8800's, went to 55 for the 9800 and then went back to 65 for the 280, before 40nm came in with fermi

The other thing is that Nvidia haven't got the power savings from HBM yet either, so as well as a say 45% improvement from the process, they also free up some power budget as well.

 

[bru]

Which ever card they launch has first is irrelevant to the news that big pascal is the first chip to be hitting production, according to the latest rumours.
Maybe they will have to go through a couple of respins or something, who knows, maybe it will work perfectly straight out of the gate. (I doubt it )
It will probably be first seen as a quadro card to begin with, but i can see it all being held up by slow hbm2 readiness anyway.

 

[queamin]

I agree hbm2 could hold pascal up a lot and not just pascal either.

 

[muziqaz]

Give in to the hype, embrace the (currently hypothetical) power!!!

No no, hype is when nvidia says pascal will be 10x more powerful than current cards Now that is a hype. What we are discussing here is much more realistic

 

[Orangey]

The other thing is that Nvidia haven't got the power savings from HBM yet either, so as well as a say 45% improvement from the process, they also free up some power budget as well.

People poo-poo'd this when it was said about AMD. Were you one of the poo-poo-ers?

 

[Stanners]

People poo-poo'd this when it was said about AMD. Were you one of the poo-poo-ers?

Can you keep this in topic, it is not a who said what about AMD thread.

 

[Gregster]

Can you keep this in topic, it is not a who said what about AMD thread.

Agreed.

I am hopeful that we will see Pascal running at least 60% faster than the current 980Ti/TX. Hopefully not that price though

 

[JediFragger]

Agreed.

I am hopeful that we will see Pascal running at least 60% faster than the current 980Ti/TX. Hopefully not that price though

I'm hoping my SLi setup will keep pace with one so I can justify keeping them a while

 

[andybird123]

People poo-poo'd this when it was said about AMD. Were you one of the poo-poo-ers?

Said what about AMD? AMD didn't have a process drop with FuryX so it doesn't apply. It's clear that AMD have managed to save power in FuryX by using HBM, it is just a shame they didn't manage a similar power/performance improvement from GCN itself.

The only thing I might have said prior to FuryX coming out was that I doubted that HBM alone was going to see the FuryX leapfrogging the 980ti, which it hasn't.

 

[shamus21]

What ever happens when Pascal is released next year your looking at the first generation of a shrink and new memory process. Their will be technical issues to start with, you will probably see only 8 and 16gb cards with say a Titan XXX for example with 32gb being a Pascal refresh mid 2017.

 

[Mr Oblong]

What ever happens when Pascal is released next year your looking at the first generation of a shrink and new memory process. Their will be technical issues to start with, you will probably see only 8 and 16gb cards with say a Titan XXX for example with 32gb being a Pascal refresh mid 2017.

I genuinely don't know what a 32GB card will be useful for, gaming wise.

I doubt we'll see games hitting 12GB by then (Maybe Star Citizen?), it feels like it'll take a new set of consoles to push things onwards.

 

[bru]

I genuinely don't know what a 32GB card will be useful for, gaming wise.

I doubt we'll see games hitting 12GB by then (Maybe Star Citizen?), it feels like it'll take a new set of consoles to push things onwards.

It is going to be interesting, as supposedly both AMD and NVidia will have completely new HBM2 line ups from top to bottom for next year. I believe (I may be wrong on this though) that HBM2 is the minimum of 4GB per stack, with the minimum of 2 stacks per chip configuration. So that means 8GB cards as a minimum, even if you can have single stack configurations that will mean 4Gb cards as a minimum.
Games being made from then on will have a lot of video memory to play with even on the lowest hardware. with it all coinciding with DirectX 12 as well, it is certainly going to make things interesting.

I do have to agree with the console part though.

 

[Mauller]

It is going to be interesting, as supposedly both AMD and NVidia will have completely new HBM2 line ups from top to bottom for next year. I believe (I may be wrong on this though) that HBM2 is the minimum of 4GB per stack, with the minimum of 2 stacks per chip configuration. So that means 8GB cards as a minimum, even if you can have single stack configurations that will mean 4Gb cards as a minimum.

I think it depends on how they both play with the HBM. If Hynix go for economics of scale and stick with 4 and 8 hi stacks. and exclude 2hi as an option. then AMD and Nvidia will more than likely create single and double stack low - mid end cards. But even then they will have far higher bandwidth than current low and mid end cards.

but if they went with 2hi then they could have cards with more than single and double stacks.

or they may just stick with HBM1 for low and mid tier parts. But then it goes back to the economic of scale issues. If HBM2 is just a straight progression in terms of production. then it will not be a cost issue to discontinue HBM1 production.

 

[Rroff]

It says a 2x density increase. You would expect more than that for a 2 node jump.

I think you're all setting yourselves up for a big disappointment.

I don't think ANY GPU released until say 2017 Q2 is going to be particularly impressive for high-end owners.

Hard to say for definite either way right now - nVidia have commercial obligations they can't meet with the compute cut Maxwell which should mean consumers see big GPUs sooner rather than later as their focus is going to be bigger cores though that could also mean less of the space used for game relevant performance.

The performance increase potential from 16nm FF+ should be fairly significant - 20nm allows for ~19% improvement on 28nm, 16 FF ~40% on that and 16 FF+ another ~15% - in testing with SRAM a direct design shrink from 28nm allowed for performance to be scaled up by 40-50% - but I've heard from a couple of people that the numbers banded about are currently only true for sub ~150mm2 and that above that it starts to drop off quite steeply due to leakage.

Plus as well as architectural changes to the main core itself HBM potentially allows a next generation GPU to utilise the advantages even further as mentioned above.

 

[muziqaz]

It makes sense to release budget minded cards with then cheaper higher yielding HBM. And leave more expensive HBM2 for high end.

 

[Boomstick777]

I think 32GB will just be for workstation cards.

My guess the Titan Pascal card will feature 16GB HBM 2.0 and an even higher price tag than previous Titan's. Maybe £950/£1050 in the UK at launch.

16GB HBM 2.0 VS 12GB GDDR5 Titan X

-----------------------------------------------------

The GTX Pascal X80 / X80 Ti could have 8GB HBM 2.0.

8GB HBM 2.0 VS 4GB/6GB GDDR5

-----------------------------------------------------

GTX X70 will have 4GB (with access to full 4GB) HBM 2.0.

4GB HBM 2.0 VS 3.5GB + 0.5GB GDDR5

-----------------------------------------------------

GTX X60 will have 2GB/3GB HBM 2.0...

 

[bru]

I think 32GB will just be for workstation cards.

My guess the Titan Pascal card will feature 16GB HBM 2.0 and an even higher price tag than previous Titan's. Maybe £950/£1050 in the UK at launch.

16GB HBM 2.0 VS 12GB GDDR5 Titan X

-----------------------------------------------------

The GTX Pascal X80 / X80 Ti could have 8GB HBM 2.0.

8GB HBM 2.0 VS 4GB/6GB GDDR5

-----------------------------------------------------

GTX X70 will have 4GB (with access to full 4GB) HBM 2.0.

4GB HBM 2.0 VS 3.5GB + 0.5GB GDDR5

-----------------------------------------------------

GTX X60 will have 2GB/3GB HBM 2.0...

I don't think you can do 2/3GB with HBM2 isn't it either 4 GB or 8GB per stack, with a minimum of 2 stacks. Grrr I'm confused now, not sure what the lower limit is. I cannot seem to find anything concrete on HBM2, even though I'm sure before I read that you can only have a minimum of two stacks, you couldn't have just a single stack. Cannot for the life of me find it now though.

 

[Simon Belmont]

I think 32GB will just be for workstation cards.

My guess the Titan Pascal card will feature 16GB HBM 2.0 and an even higher price tag than previous Titan's. Maybe £950/£1050 in the UK at launch.

16GB HBM 2.0 VS 12GB GDDR5 Titan X

-----------------------------------------------------

The GTX Pascal X80 / X80 Ti could have 8GB HBM 2.0.

8GB HBM 2.0 VS 4GB/6GB GDDR5

-----------------------------------------------------

GTX X70 will have 4GB (with access to full 4GB) HBM 2.0.

4GB HBM 2.0 VS 3.5GB + 0.5GB GDDR5

-----------------------------------------------------

GTX X60 will have 2GB/3GB HBM 2.0...

I don't think we'll need all this power for a few years yet.

 

[Kelt]

I don't think we'll need all this power for a few years yet.

Yeh, no.

 

[Boomstick777]

I don't think we'll need all this power for a few years yet.

Um wut?

4K needs all the power it can get