Nvidia Potential Roadmap Update for 2017: Volta Architecture Could Be Landing As Early As 2H 2017

Greebo · 26 Jul 2016 at 12:58

CAT-THE-FIFTH said:
At least the sub £300 cards are more like 18 months to 24 months.

sub £300
sub £300
sub £300
sub £300

With Brexit and Nvidia's pricing structure, you honestly think we are going to ever see sub £300 cards from Nvidia, with the possible exception of the lowest "50" model in the range?

Griffildur · 26 Jul 2016 at 13:02

CAT-THE-FIFTH said:
Well for a £1100+ card to be outdated that quickly lol.

It makes me wonder if people with a GTX980TI should just skip the new Titan X especially considering that Nvidia is supposedly supplying agressive rebates to clear Maxwell stock.

Volta is meant to be a new uarch which is more compute focussed.

At least the sub £300 cards are more like 18 months to 24 months.

If I had a 980 TI I would have skipped this generation completely

. Came from a 970 though.

akarypid · 26 Jul 2016 at 13:07

Jono8 said:
You say that as if you are surprised

NVIDIA Graphics card get out of date very quickly.

...there, fixed that for you!

Gregster · 26 Jul 2016 at 13:18

A rock and a hard place really. AMD cards gain performance over time but we have completed those games that they get performance in but NVidia don't improve over time and force you to upgrade to get new performance levels

eddyr · 26 Jul 2016 at 13:45

Seanspeed said:
No, it hasn't so far because current architectures can only scale so much on a given die. If they can improve that with an architecture that works well in both TDP and performance at larger ranges, they can get far more products out of a single die.

This would indeed be a 'scalable' architecture.

I wonder whether their scalability refers to a modular design based around a particular sized memory interface. The subsequent performance/utilisation across the range of products/module combinations would be supported by async. This doesn't require having multiple gpu dies composed of the basic modular unit on an interposer to scale performance however. I think that would be asking for trouble by relying so heavily on 3rd parties responsible for interposer+assembly to finalise your product stack.

Orangey · 26 Jul 2016 at 13:57

Foveated rendering is bullcrap. It was mocked when people first tried to push it and it should be mocked now. Let's trash the IQ of 2/3 the screen just to say "look we can do 4K90!".

AllBodies · 26 Jul 2016 at 15:14

Orangey said:
Foveated rendering is bullcrap. It was mocked when people first tried to push it and it should be mocked now. Let's trash the IQ of 2/3 the screen just to say "look we can do 4K90!".

It has to be combined with pupil tracking to work properly. i.e. for you to not realise it's happening.

Your eyes cannot consciously see the entire VR screen at once, even with the current only ~100 FOV. And as the FOV increases your conscious vision will be even less as a % of the full image, and that also means the gains from foveated rendering increase.

If you think foveated rendering (with pupil tracking) is stupid, then all I can suggest is you look up how the eye works.

Geck0 · 27 Jul 2016 at 11:09

Seanspeed said:
No, it hasn't so far because current architectures can only scale so much on a given die. If they can improve that with an architecture that works well in both TDP and performance at larger ranges, they can get far more products out of a single die.

This would indeed be a 'scalable' architecture.

Anyways, thanks for answering my question - you are simply guessing and have no source.

It was a proof of concept of what? Using an interposer? That does not mean that stackable graphics processors are do-able. Maybe they are, I'd just like a stronger argument than pointing to a product that stacked memory. This is not splitting hairs by any means. It is crucial as to whether or not it's something we can expect.

You made the claim, it's you who needs to support your assertion.

I've been keeping up plenty. Multi-GPU in the new API's is made harder due to not having the handy abstraction handling this in on the driver level. Notice how few DX12/Vulkan games support it? Obviously it can be done, but there is absolutely nothing to suggest it is going to be more supported than before, much less become an outright paradigm shift.

What are you talking about? I never claimed to have the answers to these things. :/

Anyways, I've learned very quickly you are not worth ever responding to again. You're rude, condescending, have no intention of supporting your arguments with anything meaningful, and also admitted you are financially invested in AMD. A cocktail for an absolutely insufferable discussion...

I really hope you dont talk like this to people in real life man.

I'm financially invested, not emotionally - I'll say my bit and go on, I don't invest time in splitting hairs with people that build a straw man then run away shrieking about how nasty I'm being. This is the internet, get over it.

I work long hours in the insurance industry arguing and litigating over liability & quantum of loss, I say my piece here in my down time and if I come across as rude then I make no apology and if I get banned I go on to another forum. I'm not here to kiss your butt or hold your hand and be your mate.

Yes, I do speak like this in real life, it's too short to waste on ****heads.

Carry on, it's not like it's the end of the world and if proven wrong, just remind me and I'll make a public apology and correction. I'm all for it.

JanM · 27 Jul 2016 at 11:32

AllBodies said:
It has to be combined with pupil tracking to work properly. i.e. for you to not realise it's happening.

Your eyes cannot consciously see the entire VR screen at once, even with the current only ~100 FOV. And as the FOV increases your conscious vision will be even less as a % of the full image, and that also means the gains from foveated rendering increase.

If you think foveated rendering (with pupil tracking) is stupid, then all I can suggest is you look up how the eye works.

Yeah combined with some sort of eye tracking system foveated rendering will work great. I don't see a reason not to use it for VR as you'd get huge performance gains for nothing and as a result be able to get more from your video card. It does have to be implemented properly.

For those that don't like it there's always the option to switch it off.

nashathedog · 27 Jul 2016 at 12:26

Seanspeed said:
I've not seen any paradigm shift towards multi-GPU with the new API's. Quite the opposite.

+1

I think it's slowly on the way out, We are getting gpu's with enough performance to negate the need for more than one gpu, I think that holds true for 99% of people and the companies would much rather sell these hugely price range toppers than cheaper cards x 2 which then require additional support and use up additional resources they are paying to maintain.

Greebo · 27 Jul 2016 at 12:39

nashathedog said:
+1

I think it's slowly on the way out, We are getting gpu's with enough performance to negate the need for more than one gpu, I think that holds true for 99% of people and the companies would much rather sell these hugely price range toppers than cheaper cards x 2 which then require additional support and use up additional resources they are paying to maintain.

depends. consoles are mgpu and if VR becomes massive there is a lot to be said for having two cards, one doing each eye rather than one monster card trying to do both eyes.

Griffildur · 27 Jul 2016 at 14:31

VR is not going anywhere, there's too much money invested and a lot of high profile companies fighting over a piece of the pie. Things will have to progress much quicker than usual now, to meet the demands of VR. Give it another 2-3 years, I think things will change massively.

Rroff · 27 Jul 2016 at 14:37

Griffildur said:
VR is not going anywhere, there's too much money invested and a lot of high profile companies fighting over a piece of the pie.

That doesn't really guarantee anything though - wouldn't be the first time a lot of money and high profile companies have jumped on some tech only for it to fall by the wayside.

VR needs to be as convenient as putting on and taking off a regular pair of glasses and be as seamless to control as a keyboard and mouse before it is going to truly take over.

Greebo · 27 Jul 2016 at 14:41

Griffildur said:
VR is not going anywhere, there's too much money invested and a lot of high profile companies fighting over a piece of the pie. Things will have to progress much quicker than usual now, to meet the demands of VR. Give it another 2-3 years, I think things will change massively.

A bit like 3d you mean?

Griffildur · 27 Jul 2016 at 15:03

Greebo said:
A bit like 3d you mean?

3D never got to the point VR is at now.

You have Microsoft, Facebook, Google, Samsung, Intel and quite a few others behind it. That never happened with 3D. Plus there are great real world applications for VR not just gaming. So yeah, I think it will catch. What we see now is rough, but second gen should be much better.

D.P. · 27 Jul 2016 at 15:08

Greebo said:
depends. consoles are mgpu and if VR becomes massive there is a lot to be said for having two cards, one doing each eye rather than one monster card trying to do both eyes.

Consoles aren't currently mgpu, and VR is far mroe complex than simply having 1 PGU render each eye. You coudl do that, but it would be icnredibly wasteful

Simultaneous multi-projection on a single GPU could be far more efficient resource wise. Pascal sees a 70% performance increase from doing that,which would make a single more powerful GPU much more attractive. Simultaneous multi-projection could work with mgpu using a alternate frame rendering but then latency increases again which is more important than absolute FPS.

Mei · 27 Jul 2016 at 15:17

things happening right now might hurt VR popularity but it will never die like 3D did, it will be around with our kids

and some of the stuff they work on helps fix problems way outside too so its great

Griffildur · 27 Jul 2016 at 15:20

Rroff said:
That doesn't really guarantee anything though - wouldn't be the first time a lot of money and high profile companies have jumped on some tech only for it to fall by the wayside.

VR needs to be as convenient as putting on and taking off a regular pair of glasses and be as seamless to control as a keyboard and mouse before it is going to truly take over.

agreed and personally I think the future of VR is not what we see now, for me VR will be a thing of beauty, no cables, no crazy weights on your face. We will get there eventually

Seanspeed · 27 Jul 2016 at 15:29

Rroff said:
That doesn't really guarantee anything though - wouldn't be the first time a lot of money and high profile companies have jumped on some tech only for it to fall by the wayside.

VR needs to be as convenient as putting on and taking off a regular pair of glasses and be as seamless to control as a keyboard and mouse before it is going to truly take over.

VR is not going anywhere. It is here, it works, it has value already for many people.

How big it actually gets is debatable, though. Worst case scenario it fizzles out into something that is supported largely by racing/flight sim communities and other niche enthusiasts.

Also, there are plenty of things that are less convenient than putting on a pair of glasses that we do all the time because it is worth doing. Interest in 3D dropped not because it was too inconvenient, but because the costs of doing everything, spending extra on 3D Bluray player, extra on 3D Blurays, having to buy 3D glasses, and of course buying a whole new TV if yours doesn't support it - simply isn't worth the payoff for most people. I like 3D actually and I still never got the urge to really want to splash out on a home 3D setup. VR brings a whole lot more to the table. It is currently inconvenient and it is currently expensive, but it also does something far more magical than even the best 3D experiences.

Griffildur · 27 Jul 2016 at 15:32

Seanspeed said:
VR is not going anywhere. It is here, it works, it has value already for many people.

How big it actually gets is debatable, though. Worst case scenario it fizzles out into something that is supported largely by racing/flight sim communities and other niche enthusiasts.

Also, there are plenty of things that are less convenient than putting on a pair of glasses that we do all the time because it is worth doing. Interest in 3D dropped not because it was too inconvenient, but because the costs of doing everything, spending extra on 3D Bluray player, extra on 3D Blurays and having to buy 3D glasses - simply isn't worth the payoff for most people. I like 3D actually and I still never got the urge to really want to splash out on a home 3D setup. VR brings a whole lot more to the table. It is currently inconvenient and it is currently expensive, but it also does something far more magical than even the best 3D experiences.

3D is amazing, I have to say. I have the 3D setup, kit, monitor etc. there are a few games which support it.

I can even run normal blu rays and enable 3D. I've seen the entire star wars collection in 3D even though the disks were not actually 3D and it was good. It was expensive, but worth it. Too bad it didn't catch up.