Is Multi-GPU (for gaming) really dying? Or are people just saying that?

[moonbogg]

I say a lot of things, most of which is marvelous BS, as you guys are fondly aware. But YOU guys have been saying something this time, and I am wondering if that too is marvelous BS.
MGPU is dead, or so its been said. I wonder if next round SLI makes sense, and my wallet should dread, or will a single GPU going forward put the issue to bed? Please advise in detail, if you would, here in this thread.

 

[Glo.]

Not exactly dead. Application will have to see multiple GPUs as one big computational unit. That is why AMD touts multiple smaller dies of GPUs on one PCB.

That is one of the foundations of HSA 2.0. There are rumors that AMD is currently working on their own CUDA-like API that is based on Mantle v2, as a compute API. That is very reason why AMD brought Boltzmann Initiative and CUDA-OpenCL compilers.

 

[moonbogg]

I meant for gaming. My bad.

 

[Arachnotronic]

GPU makers have a vested interest in being able to sell users more than one GPU. Multi-GPU won't die.

 

[tential]

Once dual gpu plug-ins for dx12 become more common like the one oxide developed then you'll see more.

Right now there are tons of things eating dev attention that are new.

 

[Glo.]

It will be the same thing. Just compute will be used for graphics .

But the idea stays. You will have to see multiple GPUs on single PCB as one big unit. Internal Coherent Fabric of the PCB. Currently GPUs talk to each other through CPU. Single PCB makes it much easier. We are talking not 2 dies like dual Fiji, but 4 - 6 dies on single PCB.

 

[Kenmitch]

I doubt it's dead.

Depending on a persons resolution and thirst for eye candy there currently is no single card solution out there. Not really sure how the next generation cards will effect the current dilemma. I'm sure they'll help somewhat but not sure they'll eliminate the need for mGPU by some users.

VR at least the hardcore uses for it look to be more feasible with mGPU too me.

 

[guskline]

moonbogg, Glo makes some excellent points.
A "snarky" response to your question is that we begin to die the day we are born.

Is CF/SLI dead? NO!

Should I go to a single card instead of CF/SLI on my next upgrade? Perhaps but it needs more analysis.

In your case, you have a VERY powerful system so unless the Big Pascal outperforms your 2 GTX980TIs in SLI I doubt you have much to worry about.

Perhaps the biggest problem is "does the game have SLI/CF support?". If you don't play the game the question is irrelevant.

I'm looking down the road to "perhaps" retire my 2 R9 290s in CF for a single BIG VEGA, but only if it matches their performance. A single GTX980TI almost does and that was the reason I moved them to the 4790k rig.

VEGA is 2 generations newer than my 290s. Pascal is only 1 generation newer than your Maxwells.

 

[Pandamonia]

The current scaling says that it is dead to me. used to get 95% and now getting 38%???

Seriously thats bad

 

[antihelten]

I'm curious, but who in the world has been saying that multi GPU is dead/dying?

Either way it is of course completely wrong, seeing as the future for multi GPU has never looked brighter than now with the arrival of VR.

 

[Pandamonia]

I'm curious, but who in the world has been saying that multi GPU is dead/dying?

Either way it is of course completely wrong, seeing as the future for multi GPU has never looked brighter than now with the arrival of VR.

And VR will be a gimmick that will have a place but never overtake normal PC gaming.

Without near doubling the FPS its going downhill.

 

[antihelten]

And VR will be a gimmick that will have a place but never overtake normal PC gaming.

Without near doubling the FPS its going downhill.

Yes VR is a gimmick, the fact that everyone is investing in it (AMD, Nvidia, Intel, Sony, Microsoft, Google, Valve, HTC, Samsung, Facebook etc.), and the fact that everyone who has actually review the finished product has had nothing but praise for them, is of course just a big conspiracy. It's not like the biggest tech companies in the world nor the tech press has any clue what consumers want, unlike random internet posters

Also it's funny that you should mention a near doubling of FPS, as VR basically ensures that you will get that (plus perfect compatibility and zero microstutter)

 

[thesmokingman]

I say a lot of things, most of which is marvelous BS, as you guys are fondly aware. But YOU guys have been saying something this time, and I am wondering if that too is marvelous BS.
MGPU is dead, or so its been said. I wonder if next round SLI makes sense, and my wallet should dread, or will a single GPU going forward put the issue to bed? Please advise in detail, if you would, here in this thread.

It's not dead, but it will be on the back burner for a while as devs transition to dx12. Once they get on board and things are rolling again, dx12 will seriously empower mgpu. You just gotta wait past the teething stages. And yes, it's a whole lot easier typing that than living thru it.

 

[poofyhairguy]

VR at least the hardcore uses for it look to be more feasible with mGPU too me.

Exactly. VR will be the golden age of multi-GPU. One GPU per eye is the best use case for multi-GPU ever. I wouldn't invest in multi-GPU for regular games though, it's rarely supported day one if at all.

 

[Headfoot]

My 290 CF doesn't work all that well. Fallout 4 I played all my hours on that game before CF was available, dont even know if it is now. Same with The Division. But it does work great when it works, like when I go back and play GTA V on ultra on 5760x1080 at a nice smooth 60. I'm still moving back to a single GPU as soon as a viable 14/16nm card comes out that will be equal or greater performance to working 290 CF. It's been nice tiding me over until 14nm though.

 

[ShintaiDK]

I would say its dead in the long run. And DX12 is one of the big reasons.

The amount of people with multi GPU setups is very small.

 

[moonbogg]

If ever there was a time to skip MGPU, it should be this next round with its massive die shrink, you think that's true? I always want more power than a single card can offer, but with such a generational leap being expected, I anticipate a single card being at least equal to the usual dual, and a chance for my financial GPU drain to finally be corrected, at least for another year, or perhaps two.

 

[Pandamonia]

Yes VR is a gimmick, the fact that everyone is investing in it (AMD, Nvidia, Intel, Sony, Microsoft, Google, Valve, HTC, Samsung, Facebook etc.), and the fact that everyone who has actually review the finished product has had nothing but praise for them, is of course just a big conspiracy. It's not like the biggest tech companies in the world nor the tech press has any clue what consumers want, unlike random internet posters

Also it's funny that you should mention a near doubling of FPS, as VR basically ensures that you will get that (plus perfect compatibility and zero microstutter)

Same idiots thought smart watches were the next big thing... look how they turned out. Also 3D TVs and games.... oh and motion controllers...

Whilst im sure its fun for a while i wonder how long you can keep that thing on your head without your eyes going funny and headaches coming.

The next fad.

 

[Flapdrol1337]

Some games should still benefit, probably pretty ones.

No need to rely on the anandtech forum, just grab the most powerful gpu and look at multigpu tests, if it works get another.

 

[RussianSensation]

If ever there was a time to skip MGPU, it should be this next round with its massive die shrink, you think that's true? I always want more power than a single card can offer, but with such a generational leap being expected, I anticipate a single card being at least equal to the usual dual, and a chance for my financial GPU drain to finally be corrected, at least for another year, or perhaps two.

If you buy AAA games day 1, multi-GPU isn't great. If you buy games 6-12-15 months from release after patches, or wait for GOTY edition, multi-GPU is good. Look up Fury X review on AT. 295X2 rips everything apart.

Also, you are a huge BF player. Multi-GPU will surely work well for BF5. So for you, dual GPU is probably still worth it. I have a feeling you won't be satisfied if a single Big Pascal barely beats your 980Ti SLI.

 

[moonbogg]

Same idiots thought smart watches were the next big thing... look how they turned out. Also 3D TVs and games.... oh and motion controllers...

Whilst im sure its fun for a while i wonder how long you can keep that thing on your head without your eyes going funny and headaches coming.

The next fad.

lol yeah, that's for sure. MGPU or no MGPU, with that ridiculous clamp clutching your head you'll be begging for mercy and stick to 2D panels instead.

 

[Snarf Snarf]

Same idiots thought smart watches were the next big thing... look how they turned out. Also 3D TVs and games.... oh and motion controllers...

Whilst im sure its fun for a while i wonder how long you can keep that thing on your head without your eyes going funny and headaches coming.

The next fad.

Just out of curiosity have you personally demo'ed the consumer version of the Rift? The evolution from Dev kit 1 to the final version shipping to customers has been incredible. This isn't a technology that is going to replace gaming on a monitor by any stretch, but it is more than just a fad. The more serious side of AR and VR isn't even in the gaming world, it's in the professional space.

The current iteration of VR as we know it is in it's infancy and will absolutely be miles better in a few years time. With every increase in IQ and motion quality we will need bigger and badder hardware to run it. If 14-16nm generation is going to be as long as Lisa Su think's it is, MGPU is the only way we're going to get to that level of power.

P.S. If you haven't tried the final launch product I highly recommend you find someone that owns one with the appropriate PC. I personally tried Valkyrie and an early version of Elite Dangerous on dual Fury X and the experience completely shattered my expectations. VR is the real deal, the only thing stopping mass adoption is the cost of hardware and lack of game library.

 

[Shmee]

Its not dead, and I think it may actually be evolving with new innovations such as XDMA crossfire, mantle/vulcan, DX12 improvements, and better games. Recent games though have been kinda meh IMO, BF4 may have been one of the last great ones as far as tech and gameplay. I hope that DOOM can do well, as well as BF5.

 

[bystander36]

Same idiots thought smart watches were the next big thing... look how they turned out. Also 3D TVs and games.... oh and motion controllers...

Whilst im sure its fun for a while i wonder how long you can keep that thing on your head without your eyes going funny and headaches coming.

The next fad.

There is a difference between a gimmick and fad. 3D is not a gimmick. When properly supported, it does exactly as it is supposed to, and does it well. I'd call it a fad, but much of that was due to the poor support, and groaning from people who never even used it or used it without proper support.

The new VR headsets may follow the same path of 3D, in that it could not last long before people hate wearing the sets, but I'm sure it'll be pretty awesome as well. The question is whether people will find wearing a headset worth the advantages and whether they can get the latency issue and support straightened away.

I personally find several games to be extremely good in 3D, and many more with terrible support.

 

[JDG1980]

I think that traditional (SLI/CrossFire) multi-GPU is dying. I can't find the source at the moment, but it was posted here a while back that there are only about 10,000 such gaming rigs in the entire world. That's not the kind of thing that is going to sustain a lot of development effort in the long run, even if these are hardcore gamers with some word-of-mouth influence. DX12 explicit multi-adapter may see some use, but I think it's more likely that devs will focus on offloading minor calculations or rendering HUDs and other non-demanding stuff to iGPUs on systems with dGPUs, since a lot more users would benefit from that.

Multi-GPU setups will remain potentially useful for GPGPU applications. Apple seems to be expecting this to be the wave of the future, with the 2013 Mac Pro having two GPUs, one dedicated to rendering tasks and one reserved for compute tasks. And our current poor attempts at VR will be able to make use of two GPUs, one for each eye. (For my money, it's not real VR until we have full immersion with all five senses, preferably through a direct-to-brain interface.)

If multi-GPU is going to get any traction in conventional gaming, it must be absolutely transparent to the software. There can be no extra effort required of the developer, nor should game-specific driver-side hacks be needed. The GPUs would have to be placed together on an interposer and present themselves to the system as one bigger GPU.