The next big gpu trend

[RaistlinZ]

I think the future will have Companion GPU's or Sub-Processors on the same PCB.

Think of it as a smaller processor that just handles certain processes, freeing up the main GPU to handle the majority of rendering. Maybe the Sub-P could just handle compute, or just handle PhysX - while not exactly being a "multi-GPU" card.

 

[monstercameron]

I think the future will have Companion GPU's or Sub-Processors on the same PCB.

Think of it as a smaller processor that just handles certain processes, freeing up the main GPU to handle the majority of rendering. Maybe the Sub-P could just handle compute, or just handle PhysX - while not exactly being a "multi-GPU" card.

sounds like async shaders with amds ace units!

 

[Pariah]

Anyone predicting the end of dGPU's after the next die shrink is predicting the end of PC gaming. Something people have been idiotically predicting for years, like Linux taking over the desktop, which obviously never happened. There is no way iGPU's will be capable of rendering AAA titles at 4k in 3 or 4 years. And 4k is not going to be the highest standard for nearly as long as 1080p was. We already have 5k monitors for sale, and 8k displays are in the demonstration phase. As long as top end monitors are too much for iGPU's, which will be the case for the forseeable future, we will continue to see dGPU's.

The OP was clearly trolling Nvidia users when he posted this, though he denied it when the mods called him out. To answer his question anyway, nothing unexpected is likely to change in the next 3 or 4 years. The constant progression of technology will trudge on, RAM will transition from GDDR5 to HBM, DirectX 11 to DirectX 12, HDMI to HDMI2. I don't see water cooling going mainstream, I don't even see Nvidia implementing it. Maybe optionally on a Titan sequel, but absolutely not as the only cooling solution on a mainstream card like AMD has done with the Fury X.

 

[cbn]

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

1.) iGPU will be limited by power unless Intel raises the TDP of the mainstream socket past 95 watts.

2.) How much are these large Intel iGPUs going to cost?

 

[Ajay]

Long term, the question is how long it will be until Intel cuts off PCI Express from everything but server/high end workstation models. At that point, most of the sales will be fanless chips that also include the PCH, memory and disk. The corpse of AMD will be long gone from x86 by then, if not out of business. nVidia's best chance might be to make some sort of ARM solution happen but that won't be easy.

Well, that won't be happening anytime soon - it is patently anti-competitive. If Intel wanted to do that, they would have to offer/agree on another interconnect standard.

 

[3DVagabond]

Well, that won't be happening anytime soon - it is patently anti-competitive. If Intel wanted to do that, they would have to offer/agree on another interconnect standard.

How is it anti competitive?

 

[Ajay]

How is it anti competitive?

Killing off your competitors by removing a piece of technology that allowed them to compete (vs iGPU) - pretty sure regulators would turn that smile upside down. NV won a suit against Intel already over bus access.

 

[monstercameron]

How is it anti competitive?

wasn't the stipulation that intel had to support pcie for x amount of years?

 

[ShintaiDK]

1.) iGPU will be limited by power unless Intel raises the TDP of the mainstream socket past 95 watts.

2.) How much are these large Intel iGPUs going to cost?

1. Irrelevant in terms of the dGPU future. Its all about the dGPU ROI.

2. Same as they cost now.

 

[3DVagabond]

wasn't the stipulation that intel had to support pcie for x amount of years?

And then what?

 

[VirtualLarry]

And then what?

And then NV dies, in the x86 space. Why do you think they're jumping onboard the Power bandwagon, and introducing NVLink. PCI-E isn't going to be around forever.

 

[imaheadcase]

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.

None of those was a trend or a big deal.

 

[imaheadcase]

And then NV dies, in the x86 space. Why do you think they're jumping onboard the Power bandwagon, and introducing NVLink. PCI-E isn't going to be around forever.

Nor will video cards as we see it. Current trend is going to have all devices on one device. Video gamers and a small % of big business only rely on the horsepower we have right now.

 

[3DVagabond]

And then NV dies, in the x86 space. Why do you think they're jumping onboard the Power bandwagon, and introducing NVLink. PCI-E isn't going to be around forever.

Or possibly AMD continues to support it with their CPU's?

 

[ShintaiDK]

PCIe isnt going away.

And for those with the tinfoils, its not exactly graphics that is the sole user of PCIe.

 

[Kippa]

I think the next big thing is 3D layered logic circuits going like stacked memory has gone. Also the possibility of rather having 1 fat gpu die on a pcb that there maybe several smaller gpus interlinked (not talking about sli) working together. Also the possibility of photonic circuits rather than electron based ones.

Also the possibility of having an external box outside the PC to house specialised hardware for the graphics instead of inside the PC. Rather than the gfx card being limited to the power and space within the PC case being 1 or 2 PCI E slots in a PC case, there could be a card that goes into a PCI E slot and fast cable going from the card to an external box which houses special hardware that replaces the gfx card. Lets face it Nvidia and AMD have severe limitations on space, power and heat with regards to putting a power full gfx card inside the PC which to be honest is relatively small space even if they do use 14nm based gpu dies. What if you could have the gfx hardware outside of the PC in its own case? You wouldn't be limited with regards to the size of the PCB, the wattage or Heat generated. Shifting the graphics hardware in a specialised hardware OUTSIDE of the PC case in a GFX case of its could the next big thing. You could have an "external gfx box" which could house the hardware equivalent of 10 gfx cards in one box. Rather than designing the gfx hardware to what is a relatively small real-estate inside a pc, outside in its own box you could REALLY expand the hardware with no restrictions on space, wattage or heat limits. Hell you could even use shit loads of 28nm tech stuff and massive amounts of ram and not have to worry about the space it takes on a pcb. An external "gfx box" with the equivalent of 10 x 28nm based gpus spread over a couple of PCBs and run 4K at 144Mhz running Crysis at 144fps.......... NICE......
For the love of god AMD and NVIDIA please consider shifting your gfx hardware outside the PC in a nice fat GFX box. You can do it with hard drives, soundcards et al why not do it with gfx technology and get rid of ALL the restrictions you have with having to put the gfx tech physically inside the PC.

Also vram limited to 640kb as that is more than enough for future needs.

 

[Namisecond]

I respect you opinions but a few things come to mind, firstly, maxwell is more efficient gaming wise at the cost of compute

In the matter of consumer graphics, I don't think lack of compute is a very big issue. We are talking about gaming here.

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.

I think I understand where he's coming from. It's not a metric of perf/w it's more a matter of perf/comfort of mind. I used to be all into the "moar cores" After a while I realized, I just wanted to play the game. Without worrying about cooling issues. Without excessive fan and coil noise. Without having to figure extravagant power supplies into a build or upgrade. For me, it's right around where the GTX 960 is. It's why the R9 380 is at a disadvantage in my eyes. That extra 6-pin connector and 80W of thermal dissipation makes me think twice.

 

[Namisecond]

Nor will video cards as we see it. Current trend is going to have all devices on one device.

The whole "convergence theory"? where everything will be shoehorned into something the size of a smart phone, where you attach a keyboard and monitor if you need a different input method?

 

[sm625]

And then NV dies, in the x86 space. Why do you think they're jumping onboard the Power bandwagon, and introducing NVLink. PCI-E isn't going to be around forever.

Nvidia isnt going to die because of PCIe. In another thread I speculated on the use of embedded flash similar to HBM. With 32-64GB of HBF you no longer need a particularly high bandwidth interface between CPU and GPU, since most or all of the game assets can be stored in the onboard flash. And it would only add perhaps 5-10% to the cost of a high midrange video card. Because of the low added cost and high utility, I expect both companies to begin using it by their 3rd generation of HBM products. It is of particular importance to AMD becasue it allows them to sell a high performance APU with enough high performance embedded flash to contain an OS and at least one game all on one SoC, which could get them design wins otherwise not possible.