The next big gpu trend

[monstercameron]

After power, heat, drivers, 3.5GB ramgate, game works, etc. What do you think is next...I think the HDMI 2.0 thing is overblown and 6GB minimum for 4k is bs.

 

[tviceman]

The next big trend is rebranding / rebadging. Nvidia did it twice with Fermi and Kepler, AMD has done it at least twice (if not more) with it's current lineup. Nvidia will do it again with Maxwell before finfet cards come out.

 

[ShintaiDK]

14/16nm reuse for 4-5 generations and the dGPU dying off. After where this forum will be renamed to IGPs and Graphics.

 

[TridenT]

14/16nm reuse for 4-5 generations and the dGPU dying off. After where this forum will be renamed to IGPs and Graphics.

This. :thumbsup:

 

[LTC8K6]

I can see the bottom and low middle dGPU disappearing as the iGPU is very soon able to take on that whole role.

I don't see the middle and high end dGPU disappearing for a while yet.

 

[monstercameron]

The next big trend is rebranding / rebadging. Nvidia did it twice with Fermi and Kepler, AMD has done it at least twice (if not more) with it's current lineup. Nvidia will do it again with Maxwell before finfet cards come out.

Yeah rebadging really struck me, why are "we" so opposed to this? It's not like 28nm can deliver much more performance.

 

[ShintaiDK]

I think the opposed part is due to refusing to see reality and the world that is waiting. 5 stages of grief comes to mind.

 

[xthetenth]

It's probably going to be a slow process driven by a more rapid increase in spaces where the possibilities for an all in one product are most appealing, not just oh hey look the dGPU died.

 

[Ajay]

14/16nm reuse for 4-5 generations and the dGPU dying off. After where this forum will be renamed to IGPs and Graphics.

This is what I thought. But there are two issues that complicated it.
1) When with Intel iGPU be fast enough to play AAA games? It's not just a bandwidth problem, but also a die size problem.

2) Nvidia has such commanding marketshare in dGPU that I would expect they can afford at least a 10'nm' foundry node and likely whatever is next. It will take even longer before iGPUs have the die space & bandwidth to beat a dGPU in in UHD (which will become the next '1080p' of gaming in a few years).

 

[ShintaiDK]

This is what I thought. But there are two issues that complicated it.
1) When with Intel iGPU be fast enough to play AAA games? It's not just a bandwidth problem, but also a die size problem.

2) Nvidia has such commanding marketshare in dGPU that I would expect they can afford at least a 10'nm' foundry node and likely whatever is next. It will take even longer before iGPUs have the die space & bandwidth to beat a dGPU in in UHD (which will become the next '1080p' of gaming in a few years).

1: Its about economics, not performance. YoY dGPU shipments decreased 19.43%. At some time we will have stagnant/old dGPUs while IGPs keep increasing.

2. For 10nm the numbers seems to be that a chip design alone cost 150 million $. And need something like 1.5 billion $ revenue to make it attractive. And thats 1 chip design. So for example with nVidia, GTX980 and GTX970 would have to generate that amount. While today on 28nm you can get away with something like 375 million $ revenue.

 

[blastingcap]

1: Its about economics, not performance. YoY dGPU shipments decreased 19.43%. At some time we will have stagnant/old dGPUs while IGPs keep increasing.

2. For 10nm the numbers seems to be that a chip design alone cost 150 million $. And need something like 1.5 billion $ revenue to make it attractive. And thats 1 chip design. So for example with nVidia, GTX980 and GTX970 would have to generate that amount. While today on 28nm you can get away with something like 375 million $ revenue.

Software will have to become more efficient. Mantle/DX12, compression, better algorithms, etc.

Also the chip dev costs will have to come down. If not then just spread the products out over longer periods of time to keep economies of scale sane.

Slowdown happens to most maturing industries... what used to be a 6-month cycle for dGPUs turned into 12-month, and now 18+ month. Eventually dGPU cycles will resemble consoles... multi-year refresh cycles.

 

[happy medium]

2016/2017,2018, same crap as we have now ,just 16nm and less bottom end cards.

Future..2019/2020,2021?
I think there will be 10nm super high end Nvidia gpu's that can do holograms/virtual reality, and IGP's will become less of a factor in personal computers. Intel will be more focused on Baytrail like SOC's than normal cpu's with IGP's like a Skylake. AMD will be swallowed up by someone like SAmsung for their GPU tech who will then compete with Intel head to head in the SOC/APU business.

Then........

You will also have a complete standard computer/Android gaming console/ streaming entertainment system built into your 70 inch UHD TV packed with 2 xbox 360 like controller and all controlled by voice command. They will be playing very advanced games like 2x better than we have today on our computers but stuffed into a ultra thin standard 70" TV.
Kinda like a super Nvidia Shield 5x more powerfull in your TV. So SOC GPU's are the future here.

Xbox and Playstation will either join the party or be overtaken.

Well this is 6/7 years from now ,,mabe I'm a little ahead of myself.

 

[njdevilsfan87]

I don't think dGPU will die off if VR starts to become popular.

 

[JDG1980]

14/16nm reuse for 4-5 generations and the dGPU dying off. After where this forum will be renamed to IGPs and Graphics.

Never gonna happen.

 

[ChuckFx]

I don't think dGPU will die off if VR starts to become popular.

VR will slow down the process but it will become a luxury product quickly. Every new push in definition will be a case for dGPUs but apart from business sector, it will be caught up by iGPUs with their colossal R&D budgets and polyvalence. iGPUs have overtaken the market in such a prodigious way, it's just a matter of time.

dGPUs are like muscle cars in an era of Smarts.

 

[happy medium]

VR will slow down the process but it will become a luxury product quickly. Every new push in definition will be a case for dGPUs but apart from business sector, it will be caught up by iGPUs with their colossal R&D budgets and polyvalence. iGPUs have overtaken the market in such a prodigious way, it's just a matter of time.

dGPUs are like muscle cars in an era of Smarts.

And how about Holograms?, that tech looked pretty cool at E3.

 

[Erenhardt]

perf/mm2 - you can call me on that in 2 years.

 

[boozzer]

I am 100% sure a max of 2 more apu generations after skylake would be enough for high setting 1080p gaming. when that happens, dgpu would be only for gamers who want 4k or smililar res.

only light I see at the end of the tunnel? if VR becomes huge it will drive alot of sales. at least 10 generations before apus can do 4k vr let alone 8k. dgpus would take at least 4 to 6 more gens to reach 8k.

of course take all my predictions with a ton of salt. :thumbsup:

 

[Ajay]

1: Its about economics, not performance. YoY dGPU shipments decreased 19.43%. At some time we will have stagnant/old dGPUs while IGPs keep increasing.

At some point, yes. I originally thought it would be game over by the end of this decade - now I think I was wrong about that prediction (iGPUs are lagging in xtor growth rates it seems - though bandwidth should improve significantly with HMC). Another big part of that will be UHD monitors driving demand for gamers, IMHO. Surely, ASPs for dGPU will jump up at each new node and with decreasing unit shipments.

2. For 10nm the numbers seems to be that a chip design alone cost 150 million $. And need something like 1.5 billion $ revenue to make it attractive. And thats 1 chip design. So for example with nVidia, GTX980 and GTX970 would have to generate that amount. While today on 28nm you can get away with something like 375 million $ revenue.

I don't think it will be that expensive for dGPUs. The logic is much more regular than, for example, CPUs and top clocks are a lot lower. Validation costs will creep up as will costs for masks and (obviously) wafers. EUV and and an eventual turn to 450mm wafers** will help fend off cost increases.



**Though not as much as expected: http://www.semi.org/en/P044331

 

[frozentundra123456]

Yeah rebadging really struck me, why are "we" so opposed to this? It's not like 28nm can deliver much more performance.

nVidia improved performance per watt very significantly on 28nm with Maxwell. Despite the "power doesn't matter" mantra that some proclaim here, it does matter to me and I suspect a lot of other people who dont want to upgrade their PSU or to just have a cool, quiet compact system for mid range performance. I applaud AMD for the apparently improved efficiency of Fury, but they sorely need those improvements across the entire line-up as well.

 

[dark zero]

I can see the bottom and low middle dGPU disappearing as the iGPU is very soon able to take on that whole role.

I don't see the middle and high end dGPU disappearing for a while yet.

Unless they use HBM 1, setting a new low standard that Intel won't catch up that in time...

 

[monstercameron]

nVidia improved performance per watt very significantly on 28nm with Maxwell. Despite the "power doesn't matter" mantra that some proclaim here, it does matter to me and I suspect a lot of other people who dont want to upgrade their PSU or to just have a cool, quiet compact system for mid range performance. I applaud AMD for the apparently improved efficiency of Fury, but they sorely need those improvements across the entire line-up as well.

I respect you opinions but a few things come to mind, firstly, maxwell is more efficient gaming wise at the cost of compute and secondly if perf/w is a major factor for some as you proclaim, then both amd and nvidia make low cost, low power gpus.

What the future needs is MOAR COARS, the army, Intel, amd and even nvidia has acknowledged this.

Probably a tonnes of grammatical errors in this response.

 

[frozentundra123456]

Never gonna happen.

I would not say "never" but it will be much longer than some are claiming, I think. I thought if anyone could do this it would be intel with their superior process tech, but they seem to continually throw more die area at the gpu with modest improvements, not to mention each new node is delayed and seems to bring less benefits. First it was going to be "wait for gen 8" now it is "wait for gen 9". Granted iris pro has pretty good performance, but architecturally I dont think gen 8 did much at all for raw performance.

Now if anyone can I think it will be AMD with Zen and HBM.

But just as a reminder, I have been following the APU tests on Game.gpu. They don't test iris pro, but in Witcher 3, kaveri 7850k cant even get average, (not minumum, average), 30 fps at 800p on LOW. So for 1440p or 4k or VR, with high image quality, and as games become more demanding, there is a long, long, long way to go.

I also dont buy the economic argument. Between high end gamers and professional applications, I think there will be enough market to drive continued development. Prices may go up, but the prices of Quadro cards and Titans show there are those willing to pay it.

 

[LTC8K6]

I don't think dGPU will die off if VR starts to become popular.

There's no way very many people are going to wear those things on their heads.

No way. :whiste:

 

[LTC8K6]

Unless they use HBM 1, setting a new low standard that Intel won't catch up that in time...

Intel isn't sitting idle regarding stacked memory.

http://www.datacenterdynamics.com/i...troduce-hybrid-memory-cubes/87502.fullarticle