Where is the GK110 successor?

[Galatian]

I think nvidia really got theirself into a situation here with the GK110 cards. I believe they intended Titan to be a halo product and use to sell of unused Tesla chips. When they figured out TSMCs 20nm tech would be (yet again) delayed, they took the bullet and released an "entire series" of GK110 cards. It's probably even the same reasoning for AMD.

Of course now both companies have to a)wait before 20 nm finally materializes and b)their top-cards will be in an awful position to compete with the big chip GK110/Hawaii. If rumors are true and 20 nm will actually not bring much on the table, I could see both AMD and Nvidia in a position where they either have to release big chips from the get go or they might not even see a performance increase at all. I mean you never know if your planned architecture is actually better (see Ivybridge/Haswell).

 

[BallaTheFeared]

I don't think but a handful of people bought a $1000 GPU for gaming, before or after the 780 or Ti.

 

[SPBHM]

hopefully the next gen is going to start with a GK110 not a GK104 successor for the high end.

 

[Zanovar]

hopefully the next gen is going to start with a GK110 not a GK104 successor for the high end.

I highly doubt that,why would they?they made a killing im guessing.

 

[BrightCandle]

I don't really want to see the mainstream gk104 as the primary release either. This round of GPUs those that upgraded for the new node lost out because none of the early chips (7000 or 600) were really the high end but they came with high end prices. Typically in the past I have factored refreshes to be at most 30% performance but these high end replacements were more than double that because both were mid range cards sold as top.

But I think based on the move to push more mobile we might see it this way again, the process change will come with smaller chips that are less risky and expensive to make and only once the process matures will we see true high end parts. It makes business sense when you think about it.

 

[Pandamonia]

I don't really want to see the mainstream gk104 as the primary release either. This round of GPUs those that upgraded for the new node lost out because none of the early chips (7000 or 600) were really the high end but they came with high end prices. Typically in the past I have factored refreshes to be at most 30% performance but these high end replacements were more than double that because both were mid range cards sold as top.

But I think based on the move to push more mobile we might see it this way again, the process change will come with smaller chips that are less risky and expensive to make and only once the process matures will we see true high end parts. It makes business sense when you think about it.

25% faster per generation is enough to get the gamers frothing at the mouth for MOAR POWA!

 

[Lonbjerg]

This GPU is old by any standards..

So when and where is its replacement? I thought 20nm was due this year?

Prices are insane for GPU tech this old..

Sods law says that soon as i buy 2x 780ti's something new comes out which makes the £1000 SLI look slow.

So you are telling me that the value of my Titan is getting better and better? :awe:
Besides HIGH-end has never been about price, but all about preformance at a premium.

Your premise is false, and you are blaming a GPU for the abscense of 20nm on top of that :whiste:

 

[Techhog]

I don't really want to see the mainstream gk104 as the primary release either. This round of GPUs those that upgraded for the new node lost out because none of the early chips (7000 or 600) were really the high end but they came with high end prices. Typically in the past I have factored refreshes to be at most 30% performance but these high end replacements were more than double that because both were mid range cards sold as top.

But I think based on the move to push more mobile we might see it this way again, the process change will come with smaller chips that are less risky and expensive to make and only once the process matures will we see true high end parts. It makes business sense when you think about it.

Not true. Tahiti was intended to be the 28nm flagship, and 28nm Hawaii only exists due to the delay and the fact that GK110 completely slaughtered Tahiti by every single metric.

 

[Pandamonia]

Not true. Tahiti was intended to be the 28nm flagship, and 28nm Hawaii only exists due to the delay and the fact that GK110 completely slaughtered Tahiti by every single metric.

Its clearly not hard to scale up a modular GPU on a mature process. Its not like AMD had a panic attack when the £1000 titan launched.

 

[Lonbjerg]

Its clearly not hard to scale up a modular GPU on a mature process. Its not like AMD had a panic attack when the £1000 titan launched.

You want a bigger die than GK110...and you whine about price? :biggrin:

 

[Pandamonia]

You want a bigger die than GK110...and you whine about price? :biggrin:

I never said i wanted a bigger one no?

I just want its replacement on 20nm

 

[Lonbjerg]

I never said i wanted a bigger one no?

I just want its replacement on 20nm

Then ask TMSC speed up their process?
And this hans nothing to do with GK110...or Hawaii XT.
Like I said, your premise is wrong, your arguments are "cart before horse"...and you make no sense.

 

[skipsneeky2]

Not true, you still see aliasing.

But i am willing to bet 4k without AA on a 24-28'' screen is gonna look years better then 1440 with AA in the majority of games right?I would hope so.

 

[Pandamonia]

But i am willing to bet 4k without AA on a 24-28'' screen is gonna look years better then 1440 with AA in the majority of games right?I would hope so.

Yes but you need about £1100 of GPU just to see that.

 

[NTMBK]

TSMC cancelled their high performance 20nm process, and only have 20nm SoC. I doubt you will see a GK110 replacement before 16nm FinFET.

 

[Pandamonia]

TSMC cancelled their high performance 20nm process, and only have 20nm SoC. I doubt you will see a GK110 replacement before 16nm FinFET.

i dont see it personally.

They would lose a fortune.

 

[Grooveriding]

But i am willing to bet 4k without AA on a 24-28'' screen is gonna look years better then 1440 with AA in the majority of games right?I would hope so.

I don't know, you'd have to quantify what 'years better' means in terms of IQ

Higher resolution is always better in my book, it adds fidelity and really raises the bar in overall IQ of games. AA is its own metric that affects IQ by eliminating aliasing on edges. So an apples and oranges comparison imo, and I found I want both, there is always going to be different levels of acceptable subjective compromise.

It's best to have both. I saw aliasing in any game on a 32" 4K screen where AA was disabled and had to enable it, which crushed framerates. It's not just AA that kills it though. Some games even without AA are just too much for dual high end GPUs to handle without dialing back other settings besides AA to get a good framerate. I can deal with anything 50fps+, below that and I don't like the experience. You can't hold 50+fps @ 4K all the time with dual Titans in some of the newer games with decent settings without AA, never mind with it.

My wife has a Yoga 2 Pro laptop which has a 3200x1800 resolution screen in a 13.3" size. Still see aliasing in WoW on that machine without AA. 4K is not the saviour to eliminating aliasing without AA, you still need AA if you don't like the appearance of jaggies.

 

[Morbus]

Damn this is getting expensive

Yeah, there's a simple fix: don't go for high resolutions... 1080p is already very high resolution anyway. 1366x768 is where the reasonably priced gaming is at. I held off upgrading to 1080p because I knew I'd have to upgrade my card. But it's never going to as cheap as with my old 1152x864 CRT @ 85Hz.

Man, those were the days...

Or not, I'm very happy with my 1080p IPS, thank you very much. And besides, as times goes by, graphic cards won't "notice" the increase in resolution that much anyway, due to the increased bus-widths and whatnot. My old HD4890 1GB dropped less than 10% performance when going from 1366x768 to 1080p. Back in the years of my FX5600, I'd see performance almost double when dropping from 1024x764 to 800x600...

Time will come when 1440p isn't expensive. But right now, even 1080p is.

 

[BrightCandle]

Fudzilla thinks its starting bottom up with Maxwell and its coming to notebooks before the desktop: http://www.fudzilla.com/home/item/33628-nvidia-maxwell-coming-in-q1-is-gm117

 

[DominionSeraph]

I think nvidia really got theirself into a situation here with the GK110 cards. I believe they intended Titan to be a halo product and use to sell of unused Tesla chips. When they figured out TSMCs 20nm tech would be (yet again) delayed, they took the bullet and released an "entire series" of GK110 cards. It's probably even the same reasoning for AMD.

LOL, what? Do you think the GTX 480/470 and 580/570 were also never meant to be gaming cards?

GK104 was a midrange chip that performed beyond expectations allowing them to sell it as a high-end product. The low-end versions are STILL selling at GTX 560 Ti (GF114)'s release price nearly two years after release, and in the GTX 770 it's still at where the Big Fermi GTX 470 and 570 were.

Nvidia held onto GK110 because they could. If the market meaningfully tops out at $500 and your 294mm^2 product warrants the $500 spot compared to the competition, why release a 551mm^2 part?

GK110 is selling for halo prices. We don't have a version at $350 like the GTX 470 and 570 were, the cut-down version goes for $500.

Having your midrange chip still at the mid-high end and your high-end chip in the halo doesn't sound like a painful position to me. Their profits seem to agree.
http://www.bit-tech.net/news/hardware/2013/05/10/nvidia-profits/1

 

[NTMBK]

i dont see it personally.

They would lose a fortune.

Why? If 20nm SoC doesn't provide noticeably higher performance than 28nm HP, and the higher cost of a 20nm wafer balances out the cost reduction from a smaller die, then why bother?

Why do you think that AMD just brought out the massive Hawaii die? They saw that 20nm SoC was going to be nothing to write home about (for GPUs), so they just went all out on 28nm so that they have something to keep them going until 16nm is ready.

 

[Pandamonia]

Why? If 20nm SoC doesn't provide noticeably higher performance than 28nm HP, and the higher cost of a 20nm wafer balances out the cost reduction from a smaller die, then why bother?

Why do you think that AMD just brought out the massive Hawaii die? They saw that 20nm SoC was going to be nothing to write home about (for GPUs), so they just went all out on 28nm so that they have something to keep them going until 16nm is ready.

if you add more transistors then you add performance.

As long as you can reach the same clock speed. Thats how GPU's work.

No one knows any facts about 20nm thats the problem

 

[Pandamonia]

Yeah, there's a simple fix: don't go for high resolutions... 1080p is already very high resolution anyway. 1366x768 is where the reasonably priced gaming is at. I held off upgrading to 1080p because I knew I'd have to upgrade my card. But it's never going to as cheap as with my old 1152x864 CRT @ 85Hz.

Man, those were the days...

Or not, I'm very happy with my 1080p IPS, thank you very much. And besides, as times goes by, graphic cards won't "notice" the increase in resolution that much anyway, due to the increased bus-widths and whatnot. My old HD4890 1GB dropped less than 10% performance when going from 1366x768 to 1080p. Back in the years of my FX5600, I'd see performance almost double when dropping from 1024x764 to 800x600...

Time will come when 1440p isn't expensive. But right now, even 1080p is.

i had 1900x1200 for years. Now i have 1440p IPS

Its a nice upgrade but its going to be pricey. The old parts are going to be worth maybe £400 max. So i want the best GPU for the future but thats hard given mantle and other stuff coming out.

 

[Lonbjerg]

if you add more transistors then you add performance.


As long as you can reach the same clock speed. Thats how GPU's work.

No one knows any facts about 20nm thats the problem

Paradox detected ()

 

[Pandamonia]

Paradox detected ()

Thats the way they have worked forever?