Exclusive: DirectX 12 Will Allow Multi-GPU Between GeForce And Radeon

[eroshima]

Why does it sound exactly like Lucid tech??? ... Nice to see it being enable on software instead of hardware. Well... kinnda xD

 

[RedJaron]

is SFR really new thing? i thought SFR was there along with AFR since the beginning. also i heard that SFR did not really take off because the poor scaling beat the purpose of going for multi gpu config in the first place

Read the whole article. It specifically says SFR has been around a while, just largely ignored.

Is it just me or does anyone else wonder why they have an utterly useless redundant GPU embedded in their Intel Core processor, which never gets utilized for games or calculations, since 99% of PC enthusiasts have a Video Card.

Because computer enthusiasts make up a very small portion of computer users in general. For every enthusiast or gamer with a discrete cards, tens of normal home and office PCs simply run off the IGPU. Notice that the IGPU is not on any of the -E series chips. It would make some sense to lose the IGPU on the K-series chips, since those are rarely ( if ever, ) used without a discrete card, but that would require making another lithography pattern and production run. The IGPU may be "wasted" space on the K chips, but it does keep production simple and therefore cheaper. And also remember that the QuickSync function introduced in Sandy Bridge was tied to the IGPU, so it was still useful, even on the K chips.

Why does it sound exactly like Lucid tech??? ... Nice to see it being enable on software instead of hardware. Well... kinnda xD

I'm not sure the idea of pooling all graphics resources together was first introduced by Lucid's Hydra. But yes, offering it in software instead of hardware is nice.

 

[Icepilot]

So 4, relatively low cost GPUs, each supplying 30 FPS, 4K resolution means a visor can provide low lag, 60 FPS 3D.

 

[animalosity]

It's about time. DX 11 sucks. Good features and looks pretty and yada yada yada....Here we are in 2015 and Developers still can't figure out how to make Crossfire or SLI work in some games....

 

[RedJaron]

It's about time. DX 11 sucks.

Huh?

Here we are in 2015 and Developers still can't figure out how to make Crossfire or SLI work in some games....

Because most of the time it's not worth it. Devs are under huge pressure by distributors and publishers to get their games out as soon as possible. When on such a tight schedule, why spend so much time, effort, and money to support a feature a very small fraction of the gaming community even runs? Especially since most games can be played at very high detail settings with a single GPU. Those games that boast high graphic detail are typically the ones that support multi-GPUs reasonably well.

 

[TechyInAZ]

You know, people could now build the worlds first 6 way Gaming rig with 6 graphics cards or 3 Dual GPU cards.

 

[turkey3_scratch]

You know, people could now build the worlds first 6 way Gaming rig with 6 graphics cards or 3 Dual GPU cards.

I wonder how Pixar and movie rendering facilities do it.

 

[renz496]

You know, people could now build the worlds first 6 way Gaming rig with 6 graphics cards or 3 Dual GPU cards.

I wonder how Pixar and movie rendering facilities do it.

Render farm? just like how HPC machines have thousands of GPU working together to do massive calculation.

 

[renz496]

It's about time. DX 11 sucks. Good features and looks pretty and yada yada yada....Here we are in 2015 and Developers still can't figure out how to make Crossfire or SLI work in some games....

because the idea of multi gpu is coming from GPU vendor and not game dev themselves. outside PC there is no such thing as multi gpu config. with Mantle AMD want to give developer more control over the gpu including crossfire support. but still the reality is how many people actually have SLI/CF setup among PC gamer? probably that's why Sniper Elite 3 doesn't have crossfire support in Mantle because game dev probably think that it is not worth the effort.

 

[renz496]

And yes Nvidia should also market their GPUs as a Physx card to work alongside an AMD GPU.. Its one way to grab more market share, from people that already own an AMD card. I personally would add an Nvidia card as Physx.

nvidia doesn't need to do that to grab market share from AMD. the report from JPR showing that nvidia is eating AMD market share on discrete GPU with alarming rate for the past two quarters. also the amount of games that really takes advantage of gpu PhysX might not really worth to get dedicated PhysX card. even without those fancy effect it does not change the gameplay at all. and the way i see it some of the effect was overdone to the point it might obscure in game vision. i think only the arkham series and Metro Last Light use gpu PhysX well. well in the manner it is not overdone. game dev probably better use tech that is applicable to everyone regardless hardware. like the stuff they used in Alien Isolation. some of the effect were GPU accelerated but it was able to work on both AMD and Nvidia gpu. the only problem is some game dev or publisher might not going to use this PC only effect if there is no "initiative" from other party to encourage them to use it

 

[Kewlx25]

Just as the implementation of these different distributed rendering techniques will be left up to the developers, won't the use of mixed GPU's, especially cross-vender, be up to the GPU manufacturer? I doubt AMD and Nvidia will allow such a configuration. Suddenly the less expensive AMD flagship combined with an inexpensive Nvidia Cuda / PhysX card would gain a lot more traction I would imagine.

In order to be DX12 compatible your device must be able to compute. DX12 will treat all computer engines and a resource pool and distribute work among them. It doesn't matter if they're AMD, Nvidia, Intel, or whatever, as long as the device can support DX12.

That doesn't make sense to me, can you explain more? I understand the concept, but I don't see DX12 not needing drivers which will lock the resources, as they do currently, if a card of a competing vendor is present.

DX12 GPUs are going to be more similar to a CPU for some basic features; They will support context switching and protected mode.

GPUs will no longer be "video cards", they will computing resources.

 

[MThorne]

If DX12 certification is on the line, NVIDIA might be obligated to play ball here.

 

[renz496]

If DX12 certification is on the line, NVIDIA might be obligated to play ball here.

still i don't see cross hardware multi gpu in the future. probably not even AMD like the idea. imagine if AMD have to make sure that their driver have to work correctly with every nvidia cards out there? not just maxwell but also Kepler and Fermi which is DX12 capable. as it is they don't even support their 5k and 6k series for DX12 despite both card are DX11 cards just like Fermi. the same goes to nvidia as well. so do we hope game developer to make this right instead? even they will have their hands full with after launch patch needed to address the problem that probably arise from their own game.

 

[Shin-san]

The Geforce 750 Ti just became REALLY interesting. If you have a smaller PSU but have the extra capacity to drive 60W more, it could give you a nice boost

 

[Squeesher]

I can sense millions of eBay and Amazon private auctions/sales being cancelled right now.

 

[Djibrille]

I can see everybody has high hopes that they will be able to use their old graphic cards in pseudo-crossfire/sli fashion but I highly doubt that. There are very few instances(if any) where hardware manufacturers allowed a new technology to enable their users to postpone upgrading. So I don't think we will be able to use our older cards (except maybe a few of the more recent cards like gtx 980 - gtx 970 or R9's) in this fashion. And I could bet my salary for many months that there's no way an integrated Intel GPU will work together with a discreet one.
But maybe I'm just being pessimistic and I will be able to pair my R9 270x with a cheap gtx 970 and have great 4k gaming on my new 4K+IPS+FreeSync+120Hz - 100$ monitor.

 

[Bobaslaw]

Interesting thought. Do I pool 2 980's via DX12 or use them in SLI.

My main question regarding performance comparison of pooling 2 gtx 980 NOT in SLI behind the dx12 api vs SLI-ing 2 x GTX 980.

Using DX12 SFR with a pool of 2 x gtx980's will have the advantage of additive VRAM = 8GB, while SLI will result in 4GB as is stated

How does the rest of the performance compare!?

Is it a possibility that pooling 2 of the same cards individually that could otherwise be SLI-ed or Crossfired produce comparable performance to SLIing/Crossfiring WHILE doubling the available VRAM.

I am very interested to know if DX12 pooling will challenge SLI/Crossfire in this way.

 

[renz496]

Interesting thought. Do I pool 2 980's via DX12 or use them in SLI.

My main question regarding performance comparison of pooling 2 gtx 980 NOT in SLI behind the dx12 api vs SLI-ing 2 x GTX 980.

Using DX12 SFR with a pool of 2 x gtx980's will have the advantage of additive VRAM = 8GB, while SLI will result in 4GB as is stated

How does the rest of the performance compare!?

Is it a possibility that pooling 2 of the same cards individually that could otherwise be SLI-ed or Crossfired produce comparable performance to SLIing/Crossfiring WHILE doubling the available VRAM.

I am very interested to know if DX12 pooling will challenge SLI/Crossfire in this way.

better wait for more info to comes out. they said the VRAM can be combined but they never talk beyond that. if the solution really that easy AMD and nvidia probably have the solution for this problem long ago. the same goes with this cross hardware multi gpu setup. certain things are much more confusing right now. DX12 was supposed to be low level. which means the API should be tweak specifically to the architecture itself like console low level API. and now were are talking about an API that should work with two different architecture at once. isn't that going to defeat the purpose of going low level?

 

[lcp1109]

If Nvidia or AMD try to block theirs GPUs from working with "Asynchronous Multi-GPU Capabilities" they can expect many complaints and probably an investigation by the European Commission for possible violation of competition law,

 

[renz496]

If Nvidia or AMD try to block theirs GPUs from working with "Asynchronous Multi-GPU Capabilities" they can expect many complaints and probably an investigation by the European Commission for possible violation of competition law,

that assumes if the tech really did successful? the thing is to make this kind of thing really possible it needs directly to be supported by AMD and nvidia themselves. the idea of mixing AMD and nvidia card (not in hybtrid physx manner) is not actually new. MSI try to make such thing possible with their big bang fuzion motherboard but it really goes nowhere because there is no support from nvidia nor AMD. if left to game developer then we probably won't see it being used in one or two triple A titles.

 

[lcp1109]

If Nvidia or AMD try to block theirs GPUs from working with "Asynchronous Multi-GPU Capabilities" they can expect many complaints and probably an investigation by the European Commission for possible violation of competition law,

that assumes if the tech really did successful? the thing is to make this kind of thing really possible it needs directly to be supported by AMD and nvidia themselves. the idea of mixing AMD and nvidia card (not in hybtrid physx manner) is not actually new. MSI try to make such thing possible with their big bang fuzion motherboard but it really goes nowhere because there is no support from nvidia nor AMD. if left to game developer then we probably won't see it being used in one or two triple A titles.

I might be wrong, but I think they can be even forced to release infos that can help developers improve compatibility between the new directx performance and graphic cards from both companies.

 

[renz496]

If Nvidia or AMD try to block theirs GPUs from working with "Asynchronous Multi-GPU Capabilities" they can expect many complaints and probably an investigation by the European Commission for possible violation of competition law,

that assumes if the tech really did successful? the thing is to make this kind of thing really possible it needs directly to be supported by AMD and nvidia themselves. the idea of mixing AMD and nvidia card (not in hybtrid physx manner) is not actually new. MSI try to make such thing possible with their big bang fuzion motherboard but it really goes nowhere because there is no support from nvidia nor AMD. if left to game developer then we probably won't see it being used in one or two triple A titles.

I might be wrong, but I think they can be even forced to release infos that can help developers improve compatibility between the new directx performance and graphic cards from both companies.

amd and nvidia can release the info but does game developer have the means to make it right? but for sure they can't force both nvidia and AMD to do it for them. they probably able to make 980 and 290X work together. but will they willing to spent resource to make sure 980 work with other R9 series such as 290, 285, 280X, 280, 270X, 270, 265, 260X, 260, 250X, 250 and 240? and you can't force AMD/nvidia/intel to support all the feature of direct x. AMD for example did not support DX11 DCL. it could be one of the reason why nvidia DX11 drivers can almost match AMD Mantle performance in certain cases.

 

[thisguy12345]

can i run dual graphics on my a10 7850k with somthing other than an r7250 after dx12?

 

[renz496]

can i run dual graphics on my a10 7850k with somthing other than an r7250 after dx12?

probably. but it seems we need game developer to do that and not AMD. and we don't know if game developer will be interested with it or not.