Pascal Rumours Who's Excited

About six months between 780 and 780ti.


I find it hard to believe that small Pascal will be 30% faster than a 980ti. Although I felt the 980 was a little underpowered, the difference between the 780 and the 780ti was not as big as the difference between the 980 and the 980ti. With that in mind, if small Pascal is 30% faster, conservatively, that would be a huge increase. Depending on the benchmark, 680 SLI was more powerful than a 980, at least at launch. If small Pascal is 30% more powerful than big Maxwell, that would make the '1080' a lot more powerful than 780 SLI. That does not seem feasible in my opinion, especially if big Pascal is going to be 70% faster than a 980ti. A 980ti is not 70% faster than a 780ti, right? I know it's a die shrink with HBM2, and it would be awesome to see, but is it really going to happen? Would nVidia blow their load so quickly when they could easily tease that kind of performance out of another generation?

Forget the fact that Pascal will use HBM as I think that will contribute very little (look at the Fury X). It will be the die shrink that makes the difference with NVidia able to pack twice as many transistors in.

With Pascal NVidia will give it all the performance they can, if they don't they will leave the door open for AMD with new designs on the die shrink.

NVidia may be ahead on 28nm but that will end with 28nm, the upcoming die shrink will put them back on a level playing field with AMD.
 
I do have the feeling that HBM2 will make a difference as compared to previous technologies, however I cannot believe that the card will be as powerful as stated a few posts back. It would put the bar too high for Volta, be too much of a leap rather than the incremental progress hardware always gets. So a moderate leap, yes, but not too large, IMHO.

Six months between vanilla and Ti is still too much, probably ha ha, maybe I'll go vanilla - > Ti. But man, they must make them of similar size as Maxwell cards, a small card would look awful IMO. Big cards look powerful, pure aesthetics.
 
I do have the feeling that HBM2 will make a difference as compared to previous technologies, however I cannot believe that the card will be as powerful as stated a few posts back. It would put the bar too high for Volta, be too much of a leap rather than the incremental progress hardware always gets. So a moderate leap, yes, but not too large, IMHO.

Six months between vanilla and Ti is still too much, probably ha ha, maybe I'll go vanilla - > Ti. But man, they must make them of similar size as Maxwell cards, a small card would look awful IMO. Big cards look powerful, pure aesthetics.

It is not going to be an incremental step though, it is going to be on a die shrink which will give it the extra performance. Big Pascal will be like running GTX 980 Ti's in SLI as it will have nearly the same number of transistors as a couple of GM200 chips.
 
Forget the fact that Pascal will use HBM as I think that will contribute very little (look at the Fury X). It will be the die shrink that makes the difference with NVidia able to pack twice as many transistors in.

With Pascal NVidia will give it all the performance they can, if they don't they will leave the door open for AMD with new designs on the die shrink.

NVidia may be ahead on 28nm but that will end with 28nm, the upcoming die shrink will put them back on a level playing field with AMD.
You might be might. Seeing as it's all stipulation at the end of the day, you could well be right.

Also, I agree that HBM hasn't helped AMD that much. Either they focused so much of their time on developing the technology—so that they had the theoretical advantage later down the line—that they didn't spend enough time doing what nVidia did—which was to squeeze every ounce of power out the 28nm process—or the technology is still too infantile to actually impact the performance in any tangible way beyond high resolutions and VR, neither of which are very common right now. HBM2 might change that dynamic though.

But the die shrink, based on when nVidia and AMD moved to 28nm from 40nm, the performance gap wasn't huge. It was the same as the transition from the 680/7970 to the 780ti/290X, which were both using the 28nm process. In fact, SLI 580 3GB GPU's was a lot more powerful option than a 680, while a 980ti is not far from being twice as powerful as SLI 780ti—although that's partly due to the VRAM and the relatively poor driver support for Keplar.
 
I would like to add that it seems to be the general consensus that nvlink will be on the consumer cards as well, whereas that feature will actually remain exclusive to the professional compute cards. I've read that somewhere from a reliable source, but don't recall where, sorry.

Also, we might want to distinguish between the usage of HBM1 versus HBM2, the latter of which could make aaa difference - but that's just a feeling and I'm sure many of you loads more about this than I do.

Why did they swap Pascal with Volta by the way?

As for Pascal = ~980Ti SLI, I presume you're right. Hence I need to correct my earlier saying incremental step forward.
 
I would like to add that it seems to be the general consensus that nvlink will be on the consumer cards as well, whereas that feature will actually remain exclusive to the professional compute cards. I've read that somewhere from a reliable source, but don't recall where, sorry.

Also, we might want to distinguish between the usage of HBM1 versus HBM2, the latter of which could make aaa difference - but that's just a feeling and I'm sure many of you loads more about this than I do.

Why did they swap Pascal with Volta by the way?

As for Pascal = ~980Ti SLI, I presume you're right. Hence I need to correct my earlier saying incremental step forward.

AFAIK I think they just changed the codename for it. In the roadmaps they have released both were supposed to come with stacked DRAM. One could have been less ready so they switched it, no one will ever know why though unless they say why.

As far as HBM1 vs 2 goes, HBM2 will have a bigger impact. First off because no game atm is GPU Vram bandwidth limited(afaik and understand atm), they are limited by the CPU. DX12 should help exploit the advantages over GDDR5 with HBM1 but too soon to tell. Secondly, HBM2 is twice as fast, so that clock speed increase alone would give it a huge boost. Also along with other improvements(Latency being one if they decrease it further) would boost it more than say allowing for more data through at once compared to V1 and in addition slightly improving the maxmium bandwidth it can push at once. So it adds more lanes for data to go through while doubling the lanes speed. While adding more lanes won't really help currently, doubling the speed will for whatever actually does utilize the lanes boosting overall throughput.
 
Thanks for that explanation NBD, you just got me more excited (I truly am a GPU wh*re).

I think I am dead set on the ASUS RoG Swift PG278Q... I know the PG27AQ is around the corner but it's also 300 euro more expensive, and I can live without IPS and with 144Hz instead of 165Hz. More on that in the thread I'll make come November (purchase will be December).
 
Thanks for that explanation NBD, you just got me more excited (I truly am a GPU wh*re).

I think I am dead set on the ASUS RoG Swift PG278Q... I know the PG27AQ is around the corner but it's also 300 euro more expensive, and I can live without IPS and with 144Hz instead of 165Hz. More on that in the thread I'll make come November (purchase will be December).

Don't get to excited, still won't see that much of an leap in performance. It'll be more than HBM1 but don't expect a magical 10FPS increase:p
 
Don't get to excited, still won't see that much of an leap in performance. It'll be more than HBM1 but don't expect a magical 10FPS increase:p
I would'nt be too sure about that you know how much Nvidia love to smack at AMD by saying something like "See this is how it's supposed to be done" and release a beast card.
I just wish though that one day the tables get turned thoroughly around and who knows with Microsoft getting involved they may have the R&D funds to do so.
Still it's an interesting time ahead that's for sure
 
I would'nt be too sure about that you know how much Nvidia love to smack at AMD by saying something like "See this is how it's supposed to be done" and release a beast card.
I just wish though that one day the tables get turned thoroughly around and who knows with Microsoft getting involved they may have the R&D funds to do so.
Still it's an interesting time ahead that's for sure
I would be both apprehensive and excited if Microsoft took over AMD. It would shift the console market massively, as well as the PC market. But at least if Microsoft did purchase AMD, I would hope they'd simply offer funds and marketing rather than try to change the core dynamics of what they do and stand for. Because as it is, AMD are making bold moves. Ultimately they're not very bankable, at least not immediately, but they have an identity. I would hope that Microsoft would allow the AMD division to remain unique and innovative.
 
16GB is the max I expect for next gen cards and only for the top tier models. However expect supply to be super low again unless they get massively better yields or more buyers for increased mass production.

Maybe that's:

1060 8GB
1070 16GB
1080 16GB
Titan 32GB

Or maybe 32GB is reserved for quadro cards.

I know that won't be the names :)

What do you consider top tier models?
 
Maybe that's:

1060 8GB
1070 16GB
1080 16GB
Titan 32GB

Or maybe 32GB is reserved for quadro cards.

I know that won't be the names :)

What do you consider top tier models?

As in the TX and the Fury X. Nothing over 16GB. 8GB is more than enough now, 16 for epeen. I suspect it'll be
1060 4GB
1070 8GB or 7.5 lol:p
1080 8GB
Titan 16GB
Of course this is more of hopeful wishing then anything. I think 16GB will be more common for professional GPUs. Imagine 32GB HBM2 for the Firepros? Damn... It's possible to because they already made a 32GB GDDR5 model. They would sell like hot cakes for super computers or something along those lines:p
 
Last edited:
As in the TX and the Fury X. Nothing over 16GB. 8GB is more than enough now, 16 for epeen. I suspect it'll be
1060 4GB
1070 8GB or 7.5 lol:p
1080 8GB
Titan 16GB
Of course this is more of hopeful wishing then anything. I think 16GB will be more common for professional GPUs. Imagine 32GB HBM2 for the Firepros? Damn... It's possible to because they already made a 32GB GDDR5 model. They would sell like hot cakes for super computers or something along those lines:p

Yeah yours sound more reasonable than mine. I must have been drunk when I concocted mine.

12GB for a Ti model...?

Also, your ideas keep the cards affordable.

Yeah that'd make for great compute cards!!
 
I wonder if I can still use my trusty 2600K with Pascal.

PCPER did a good segment on how Sandybridge is good for the moment but it is starting to show signs of age with the latest cards even in single GPU configurations.

Upgrading to something newer did show significant signs of improvement mainly in the frame times department and Crossfire-X/SLI card setups saw a good performance gain of up to 40% and frame times improved too.

I imagine this will be a bigger improvement once AMD's and Nvidia's next line comes out -

 
Last edited:
As in the TX and the Fury X. Nothing over 16GB. 8GB is more than enough now, 16 for epeen. I suspect it'll be
1060 4GB
1070 8GB or 7.5 lol:p
1080 8GB
Titan 16GB
Of course this is more of hopeful wishing then anything. I think 16GB will be more common for professional GPUs. Imagine 32GB HBM2 for the Firepros? Damn... It's possible to because they already made a 32GB GDDR5 model. They would sell like hot cakes for super computers or something along those lines:p
This sounds the most likely to me as well.

I wanna know what AMD have cooking personally, I'm looking forward to seeing both camps though :)
Yeah, now that I'm tied to FreeSync, I'm more interested in seeing whether AMD can truly regain control of their fair market. I think they'll always have very good budget options, but they really need to work on their top-tier stuff. Personally, I think they should do away with the niche products unless you have the same power available from more widely appreciable products. The Fury X, for instance, should have had a $50 cheaper air cooled version, with the Nano being $50 cheaper and the Fury being $50 cheaper. This may not have been possible with how expensive HBM was to develop and produce, but it may have given AMD the push it needs and deserves.

PCPER did a good segment on how Sandybridge is good for the moment but it is starting to show signs of age with the latest cards even in single GPU configurations.

Upgrading to something newer did show significant signs of improvement mainly in the frame times department and Crossfire-X/SLI card setups saw a good performance gain of up to 40% and frame times improved too.

I imagine this will be a bigger improvement once AMD's and Nvidia's next line comes out -

I'll probably be upgrading my chipset from a Z97 4670K to a Broadwell-E or something like that next year some time.
 
Back
Top