GeForce + Radeon Previewing DirectX 12 Multi-Adapter with Ashes of the Singularity

Calin Banc · 26 Oct 2015 at 19:15

The developer; he's doing most of the stuff now - assuming of course, the drivers are in a good enough state.

muziqaz · 26 Oct 2015 at 21:09

Latest AOTS version does not have anything multigpu related in options. Am I right on assuming that Anandtech got prerelease version with extra options to enable multigpu? Or am I being thick?

tommybhoy, this dilemma is easily solved. Take AMD GPUs, run them in multi gpu setup, record the scaling. Run nvidia multigpu setup, record the scaling. Whichever scales better that one shouldn't be blamed

Dicehunter · 26 Oct 2015 at 21:57

Mei said:
not sure what u trying to say!
sounds kinda racist!!
....like wth lol

Not racist, It's all about the butt.

Starfish.
(w)Hole
Got it yet ?

Rofflay · 26 Oct 2015 at 22:01

/Engaging sarcasm mode

Just think you could have Nvidia with AMD features like umm TrueAudio, Mantle and various others things that are either dead or used on one game,

Or wait you could have the FRTC with the useless 55-90 range limit.

/Sarcasm mode off

Now what the hell is the real point of this might i ask?

Dicehunter · 26 Oct 2015 at 22:04

Rofflay said:
/Engaging sarcasm mode

Just think you could have Nvidia with AMD features like umm TrueAudio, Mantle and various others things that are either dead or used on one game, Or wait you could have the FRTC with the useless 55-90 range limit.

/Sarcasm mode off

Now what the hell is the real point of this might i ask?

Well if it works with AMD + Nvidia then we could mix GPU's, Just say you have a 980 but want more performance, You could stick in your old 680 next to it to boost it up

tommybhoy · 26 Oct 2015 at 22:47

Calin Banc said:
The developer; he's doing most of the stuff now - assuming of course, the drivers are in a good enough state.

muziqaz said:
tommybhoy, this dilemma is easily solved. Take AMD GPUs, run them in multi gpu setup, record the scaling. Run nvidia multigpu setup, record the scaling. Whichever scales better that one shouldn't be blamed

Twas simply a joke.

FredFlint · 26 Oct 2015 at 22:56

D.P. said:
Very interesting and surprising results.

It open up a whole world of possible issues though. For example, Maxwell supports DX12 features that Fiji doesn't like order independent transparency/raster order views and conservative rasterization, conversely Fiji supports more advanced resource binding.

Having different GPUs will presumably mean that feature support has to drop down to a common denominator.

Think the lead card will be the main feature set and the secondary card will only be used for things it supports, i may be completely wrong though as i have not looked into the implementation details.

Kaapstad · 27 Oct 2015 at 01:19

muziqaz said:
Latest AOTS version does not have anything multigpu related in options. Am I right on assuming that Anandtech got prerelease version with extra options to enable multigpu? Or am I being thick?

No multi GPU option in the latest build.

On a multi GPU NVidia setup I have got to turn off all the other cards on the motherboard dip switches to even get the primary card to run DX12 lol.

At least with the AMD cards all I have to do is disable C/F to get a single card DX12 run.

Also as I have said above in other posts the bench is seriously flawed @1080p as the TitanX, Fury X or GTX 980 Ti should all be able to reach 80fps using the bench thread settings but none of them can even get get close. It is as if DX12 is behaving like DX11 lol. Until the game devs get 1080p sorted for both brands using single GPUs any other results should be take with a pinch of salt.

Dygaza · 27 Oct 2015 at 01:30

I agree that there should be more scaling between resolutions, but 80fps sounds too high in such compute heavy application. Not everything gpu does scale linearly with amount of pixels.

Kaapstad · 27 Oct 2015 at 03:36

Dygaza said:
I agree that there should be more scaling between resolutions, but 80fps sounds too high in such compute heavy application. Not everything gpu does scale linearly with amount of pixels.

The problem is even if you run high end graphics cards with a 5960X (which is a beast on compute heavy applications) you still don't get the performance you should do.

I think people are reading too much into the AOTS bench/game but unfortunately at the moment is is very flawed.

Another odd thing is if you use max settings (higher than the bench thread settings) @2160p a TitanX will run it easy and will only use just over 3gb of VRAM yet a Fury X will tank and really crawl and try to use 5gb of VRAM. There should not be that much difference between the two cards with VRAM usage.

The devs have got a lot of work to do before this game is right.

Noctifer · 27 Oct 2015 at 08:31

https://twitter.com/draginol/status/658679124098228225

We expect to make the build we shared with @anandtech available to the public by Beta 1 of Ashes of the Singularity next month.

mahius · 27 Oct 2015 at 10:01

Nice idea, I can kinda understand why Kaap is disappointed, but this is early days. We need more DX12 games to test this with. Oh and 980ti + Titan X is disappointing. If this scenario is to be believed, then a cheap old gen card will do barely anything. Otherwise, I'm more excited about having both an NV card and an AMD card in the same machine. If a game is more optimised to AMD, then use the AMD card as single GPU and vice versa. And if people want a bit more FPS, then turn both cards on, granted these numbers improve. Otherwise I'm still in the camp that we shouldn't bother with more than 1 GPU, even SLI and CF have their issues, but at least they do scale well on some games.

nashathedog · 27 Oct 2015 at 10:08

shankly1985 said:
I would like this to be come a reality. User based wins if anything though I can see nvidia blocking it in there drivers.

Wouldn't surprise me,

Would it also mean we can get a Nvidia card and use it for Phys-x or Nvidia works with an AMD card?

Would it also mena we would have the ability to use completely different card ranges together such as a Fury with a 380?
Or a 970 with a Ti?

nashathedog · 27 Oct 2015 at 10:18

Gregster said:
The AMD card is the bottleneck....I'll get me coat

an optimized SLI/CF setup can get better than 80% gains – but overall the data here confirms our earlier raw results: we’re actually seeing a significant uptake in performance with the mixed GPU setups. R9 Fury X + GTX 980 Ti is some 75% faster than a single R9 Fury X while GTX 980 Ti + R9 Fury X is 64% faster than a single GTX 980 Ti. Meanwhile the dual AMD setup sees a 66% performance gain, followed by the dual NVIDIA setup at only 46%.

Marine-RX179 · 27 Oct 2015 at 10:26

nashathedog said:

So conclusion is don't mix two different Nvidia graphic cards together, as that combination offer the worst performance gain

pgi947 · 27 Oct 2015 at 10:42

nashathedog said:
Wouldn't surprise me,

Would it also mean we can get a Nvidia card and use it for Phys-x or Nvidia works with an AMD card?

Would it also mena we would have the ability to use completely different card ranges together such as a Fury with a 380?
Or a 970 with a Ti?

I think the latter is the more likely outcome. It's no bad thing, I know everyone's creaming of the prospect of two vendors running in unison in a single system, but personally I think just the prospect of having two driver sets running in the same system is asking for trouble.

Different architectures from the same vendor though, bring it on

Mauller · 27 Oct 2015 at 10:46

Another major point with scaling in this MultiGPU setup, is how they are performing their SFR.

Depending on the implementation it can make a vast difference in the overall performance.

I think they are using a Split screen method, wethere it is a fixed or variable split is unknown. I would assume they would have some kind of variable split with the engine determining the workload and trying to equalise it as much as possible. Split screen should work better in an RTS game considering the majority of the top half of the screen is not looking at sky like with an FPS or RPG etc.

SFR Supertiling should still be the best overall. Better scaling and same latency as single card.

Noctifer · 27 Oct 2015 at 11:09

Mauller said:
Another major point with scaling in this MultiGPU setup, is how they are performing their SFR.

Depending on the implementation it can make a vast difference in the overall performance.

I think they are using a Split screen method, wethere it is a fixed or variable split is unknown. I would assume they would have some kind of variable split with the engine determining the workload and trying to equalise it as much as possible. Split screen should work better in an RTS game considering the majority of the top half of the screen is not looking at sky like with an FPS or RPG etc.

SFR Supertiling should still be the best overall. Better scaling and same latency as single card.

Ashes uses AFR not SFR. Read the article.

D.P. · 27 Oct 2015 at 11:16

FredFlint said:
Think the lead card will be the main feature set and the secondary card will only be used for things it supports, i may be completely wrong though as i have not looked into the implementation details.

But there is no way that will work for AfR, you will get one frame renders one way and the other frame will end up with different effects. So the only way to do this would be to drop feature set down to the lowest denominator.

Noctifer · 27 Oct 2015 at 11:22

D.P. said:
But there is no way that will work for AfR, you will get one frame renders one way and the other frame will end up with different effects. So the only way to do this would be to drop feature set down to the lowest denominator.

The feature has a lot of flexibility according to the article. A single GPU doesn't need to work on the whole frame, tasks can be split. 90% of frame 1 workload could be done by GPU1 and the other10% by GPU2, while those 10% contain features that GPU1 can't do. This is a implementation that is a lot more complex, but I guess it could be possible.