Ashes of the Singularity User Benchmarks Thread

[Despoiler]

WCCFTech has actually done some fairly sensible analysis of this for once: http://wccftech.com/async-shaders-give-amd-big-advantage-dx12-performance-oxide-games/

They conclude for people with a decent cpu the benefits of DX12 will be pretty small, which tbh is what most techies are expecting (other then fanboys looking for something else to argue about). It's not like DX10 or DX11 proved magically much faster then DX9 in the end, and I assure you there were just as many posts back in the day claiming DX10 in particular would do amazing things.

We've been over this ad nauseam. The benefits that all people get, regardless of CPU, is that the frametimes in DX12 are roughly half that of DX11, max fps goes up 10-20% and min fps will be higher still. This results in a smoother experience in the gameplay. In addition, freeing up the CPU and having better core scaling will let the devs do more in their game. Whatever CPU resources have been freed up or made more available will get used at some point. It's up to the devs to exploit it. Thinking things will remain static is naive at best.

 

[poofyhairguy]

Whatever CPU resources have been freed up or made more available will get used at some point.

Isn't there a chance that the consoles (with their super weak CPUs) will hold back demands for CPU resources in PC gaming?

I mean I could see some type of hairworks or physx type feature where you take the console game and then bolt this feature on and it eats CPU like crazy, but one has to assume the base game underneath will have assets and optimizations for the weaker consoles especially if Direct 12 is as close to the Xbox One API as we have been lead to believe.

I know that some PC-specific game types like RTSes use more CPU regardless so its not a universal thing. But it seems all the true AAA games coming out on PC now are ports, and the whole reason async compute was used on consoles first was to compensate for the weak CPUs.

What would be funny is if NVidia cranks out some sort of a software emulator to negate some of the performance hit for a lack of hardware features and in the early Directx 12 era AMD actually has lower CPU use. It would be an interesting turning of the tables for sure.

 

[stateofmind]

By your definition, Kepler can do Async Compute, because it too can process 32 queues of COMPUTE with Hyper-Q.

Why did NV say Kepler cannot do Async Compute? They even told Ryan Smith that Maxwell 2 is different to Kepler, it can process graphics + compute simultaneously.

At this point:

1. Oxide & AMD & the B3D Async Compute program is lying/wrong.
2. NV is lying.

We just have to keep in mind that these are commercial companies we are dealing with. If they could, they wouldn't have mentioned anything wrong with anything they have, the same as they knew about AC and how much DX11 is limiting in general and say/did nothing. Even from the consumption point of view - people were buying that GPU or another without knowing how much difference the AC/GCN/M1/M2 could do, and that's when we already have Mantle results for quite some time..

 

[Shivansps]

I was wondering about something, so much that has been talked about DX12 multiadapter, cant a DX12 IGP be used to do calculations in cases like this one? not sure if a memcpy is needed for that.

 

[TheELF]

But it seems all the true AAA games coming out on PC now are ports, and the whole reason async compute was used on consoles first was to compensate for the weak CPUs.

Can anyone find a list,are there more then these 3 games that actually use async compute on the consoles?

Because if there are only,for example, a couple of games each year that actually use it then it's no big deal,well not that big.

And does anyone remember what happened when thief came out and AMD was all "ha look at how great that is" and nvidia was all "ha ,we made the same improvement jump using dx11" ?

 

[Eymar]

Can anyone find a list,are there more then these 3 games that actually use async compute on the consoles?

Because if there are only,for example, a couple of games each year that actually use it then it's no big deal,well not that big.

And does anyone remember what happened when thief came out and AMD was all "ha look at how great that is" and nvidia was all "ha ,we made the same improvement jump using dx11" ?

In terms of real impact to end users in next year or two, their maybe no real impact. However, for current maxwell 2 owners or prospective 980/980ti buyers perceived value is going to be a concern. For current maxwell owners should they sell now to maximize resell value if lack of async compute is confirmed and lowers resell value because of it. For current prospective buyers do they even look at 980 or 980ti knowing future dx12 performance may not be there. Stupid as it seems, but perceived value will sometimes overrule real value.

 

[Mahigan]

Well this whole issue blew up... now it's time to calm down and re-iterate the facts to folks who are concerned.

That's the job of the Tech Journalists... not us.

On a side note, we all learned much more about GPU architectures in the process. I think that this is probably the best thing to come out of all of this

 

[Silverforce11]

Can anyone find a list,are there more then these 3 games that actually use async compute on the consoles?

These are coming soon, 02/2016.

Mirror's Edge:
http://www.vcpost.com/articles/8717...ion-technologies-glass-city.htm#ixzz3kSkxnueB

Tomb Raider:
http://gearnuke.com/rise-of-the-tom...breathtaking-volumetric-lighting-on-xbox-one/

Deus Ex & Hitman (Same Engine/Squaresoft):
http://gearnuke.com/deus-ex-mankind-divided-use-async-compute-enhance-pure-hair-simulation/

https://www.youtube.com/watch?v=D-epev7cT30

Fable (not sure about release date):
https://youtu.be/7MEgJLvoP2U?t=20m47s

Also all the new MS games will be DX12 in their initiative to combine Xbone + Win 10 ecosystem. Arma 3 and DayZ are moving to DX12 too. ARK is due for a DX12 patch, the performance is horrendous so they need it.

 

[TheELF]

These are coming soon, 02/2016.

Those are DX12 games,but we already learned that DX12 does not automatically mean use of async compute or at least not necessarily a meaningful amount of async compute.
DX12 provides quite a few thing you can do asynchronous like stream textures or what have you and this gives improvements to all cards,only the compute part gives huge improvements only to GCN.

 

[Silverforce11]

Those are DX12 games,but we already learned that DX12 does not automatically mean use of async compute or at least not necessarily a meaningful amount of async compute.
DX12 provides quite a few thing you can do asynchronous like stream textures or what have you and this gives improvements to all cards,only the compute part gives huge improvements only to GCN.

Read the links buddy.

 

[TheELF]

Read the links buddy.

I did and they all say they will use async compute for very concrete/limited things none of those games have as much stuff happening on screen at once as ashes has.
You really have to push limits to get the same amount of gain and none of the other games show that they will.
The additional ACEs only get used when you breach the limits of a normal game.

 

[Erenhardt]

You really have to push limits to get the same amount of gain and none of the other games show that they will.
The additional ACEs only get used when you breach the limits of a normal game.

False.

Also, what does number of units have to do with this?

 

[Silverforce11]

The additional ACEs only get used when you breach the limits of a normal game.

Some people don't understand the concept of moving from a serial pipeline where compute traffic delays graphics rendering, to one that's parallel, where graphics rendering is now moving freely.

Read it very slowly:

http://forums.anandtech.com/showpost.php?p=37678419&postcount=30

 

[TheELF]

False.

Also, what does number of units have to do with this?

Every unit is a separate task that the GPU must deal with,more units=more work for the ACEs that otherwise would do nothing.
In a "normal" game that has much less stuff going on there will not be enough work for the additional ACEs to make a big difference.

 

[TheELF]

Some people don't understand the concept of moving from a serial pipeline where compute traffic delays graphics rendering, to one that's parallel, where graphics rendering is now moving freely.

Read it very slowly:

http://forums.anandtech.com/showpost.php?p=37678419&postcount=30

I have read it very slowly,maybe you should do the same,the graph shows copying data and compressing textures done in parallel,that's supported on all cards is it not?
How would only that give you 100% more speed on a GCN card? (double FPS)

It is clear that work gets divided between all 8 ACEs to get this kind of boost.

 

[Silverforce11]

I have read it very slowly,maybe you should do the same,the graph shows copying data and compressing textures done in parallel,that's supported on all cards is it not?
How would only that give you 100% more speed on a GCN card? (double FPS)

It is clear that work gets divided between all 8 ACEs to get this kind of boost.

The parallel tasks are compute based. Even AA can be done with compute shaders.

And yes, the benefit will depend on the game. The more compute that is used, the higher the benefit, because it FREEs up the graphics pipeline to perform only graphics, no bottlenecks with compute blocking graphics.

As to Ashes, I don't know how much compute it uses. We have no data.

But this is what Oxide said:

Ashes uses a modest amount of it, which gave us a noticeable perf improvement. It was mostly opportunistic where we just took a few compute tasks we were already doing and made them asynchronous, Ashes really isn’t a poster-child for advanced GCN features.

Our use of Async Compute, however, pales with comparisons to some of the things which the console guys are starting to do.

 

[desprado]

Benchmark
http://forums.overclockers.co.uk/showthread.php?t=18690671

R9 290 21.2 fps on heavy batches and GTX 980 Ti is getting 39.5 fps on 1080p both.

Same guys were claiming that Fury X will be 20% faster than GTX 980 Ti so now they are saying that R9 390X will beat a GTX 980 Ti in Dx12 Async sharder benchmark.However, still some these member did not learn from fury X ;therefore, maybe in future they can learn.

 

[antihelten]

As to Ashes, I don't know how much compute it uses. We have no data.

Actually we do. The Oxide dev mentioned that their rendering pipeline consists of roughly 20% compute shaders at the moment, and that they are aiming for 50% in future builds.

 

[Silverforce11]

Actually we do. The Oxide dev mentioned that their rendering pipeline consists of roughly 20% compute shaders at the moment, and that they are aiming for 50% in future builds.

Thanks, I was not aware of that.

50% compute is pretty high. Serially, that would mean 50% bottleneck on the graphics pipeline, with Async Compute leading to a doubling of performance.

Seriously if they are aiming for 50%, it's going to make that game an exclusive title that runs like *donkey* on any GPU that doesn't support it. Not wise in the current GPU climate on PC (or are they making Ashes for consoles too?).

 

[.vodka]

I have read it very slowly,maybe you should do the same,the graph shows copying data and compressing textures done in parallel,that's supported on all cards is it not?
How would only that give you 100% more speed on a GCN card? (double FPS)

It is clear that work gets divided between all 8 ACEs to get this kind of boost.

You might want to read this too.

http://www.anandtech.com/show/9124/amd-dives-deep-on-asynchronous-shading

No, it has nothing to do with the amount of units on screen if you're going to keep repeating that in every thread.AotS has lots of units on screen, more than we've ever seen before because DX12 is a low overhead API that can go nuts on the draw calls front without performance going off a cliff and without murdering the CPU in the process, as it happens on DX11. No, DX11 multithreading as nV supports doesn't even compare to what DX12/Vulkan can do. Yes, everyone who's on DX12 benefits from this. nV, AMD and Intel, all three. You get to use all the cores in the CPU to work together with the GPU instead of one or two at most, on top of whatever the engine does with all the remaining cores. Much better now.


Now, you do not seem to understand that different kinds of workloads aren't exclusively tied to the same resources on the GPU. DX11 as explained by AMD is serial by nature, one job must finish before the other can start.DX11 is obsoleted, hard, by what DX12 brings. DX12 is a paradigm shift. Async compute is another one on top of it. You have a graphics workload, that now can run by itself using its relevant parts of the GPU, while for example you can now have a compute job running in parallel using other parts of the GPU. Compute means flexibility that the graphics pipeline doesn't give you by its very own nature, you can do whatever you want, basically. Go read any gamedev forum, you'll see the term compute *everywhere*, in lots of scenarios and new possibilities. As software catches up to the hardware, you as an end user get more out of those 2000-4000sp, hundreds of texture units, dozens of ROPs, etc you paid hard earned cash for. They all get to do some more work now, more efficiently, all at the same time. It doesn't matter what color GPU you own.


This removes inefficiency not only for AMD but also for nV and Intel. It doesn't matter how good their architectures are at DX11. It doesn't matter how good they actually are at DX11, that's history now, and DX11 is nV's trophy. We have great DX11 games we can all enjoy that won't hold a candle to future games based on what DX12/Vulkan allow. The transition period, on the other hand, is always a mess, as the people leading the innovation have to work out how to use the new tools at their disposal. Going forward this is what the new API and paradigm enables, and everyone's going to be on it next year, not to mention the following one. The console guys are all over it now (as they've ever been) and initially some of that awesomeness is coming to the PC thanks to DX12/Vulkan next year, more of it later.


Actual hardware is DX11 at its heart, it can do most DX12 things (the baseline, let's say) just because these GPUs we have in our rigs actually go much beyond what the spec allows... that doesn't mean they're suited or optimal for the job. The fun stuff begins next year with Pascal vs Arctic Islands, built from the ground up for DX12. Right now everyone has made compromises, nV/intel? (so far none could get any data on broadwell/skylake at the B3D DX12 AC thread) not supporting async compute, while AMD doesn't support ROVs and CR. That ought to change to level the playing field between all three going forward.


Again, the transition period will be handled by GCN 1.x and Maxwellv2, and it seems GCN is the most suited for that, so far. That could change. nV not doing a statement on the matter of async compute is concerning. The fact they mention async compute in one of their slides, and then seeing the train wreck that their AC results are at the B3D thread is more than concerning. Maybe a driver update fixes the feature, maybe not. It could mean another 970 fiasco... this time of a much more important magnitude.

Maybe it would have been better for them just not to mention it, then no one would've given a damn about their AC situation, and just released Pascal supporting it as it should. People wouldn't have gone investigating after finding something wasn't right, as it happened with the 970.



Embrace change, do not fight it when it's for the better in every front, for everyone.

 

[antihelten]

Thanks, I was not aware of that.

50% compute is pretty high. Serially, that would mean 50% bottleneck on the graphics pipeline, with Async Compute leading to a doubling of performance.

Seriously if they are aiming for 50%, it's going to make that game an exclusive title that runs like *donkey* on any GPU that doesn't support it. Not wise in the current GPU climate on PC (or are they making Ashes for consoles too?).

Note that the dev didn't really specify (as far as I remember) whether it was 20% of the shaders in the pipeline that were compute shaders or if 20% of the rendering time within the pipeline was taken up by compute shaders .

If it's the first one then increasing the number to 50% might not have that big of an impact (i.e 50% of all the shaders might be compute shaders, but if the compute shaders generally take up less time than the graphics shaders, then that wouldn't equate to 50% of the rendering time)

 

[3DVagabond]

Thanks, I was not aware of that.

50% compute is pretty high. Serially, that would mean 50% bottleneck on the graphics pipeline, with Async Compute leading to a doubling of performance.

Seriously if they are aiming for 50%, it's going to make that game an exclusive title that runs like *donkey* on any GPU that doesn't support it. Not wise in the current GPU climate on PC (or are they making Ashes for consoles too?).

There will be a separate path for cards not capable of async compute.

 

[Shivansps]

Actually we do. The Oxide dev mentioned that their rendering pipeline consists of roughly 20% compute shaders at the moment, and that they are aiming for 50% in future builds.

And no intention to use features that AMD cards does not supports? and then people ask why we think they are biased.

Anyway, i still wonder how much of that compute could be done on a IGP that has DX12 support.

 

[Silverforce11]

There will be a separate path for cards not capable of async compute.

Yeah the async disabled, serial DX11 path, where graphics + compute are stuck in traffic together, singing kumbaya...

I don't believe for a moment they intend to push their engine towards 50% compute. For the PC market, that does not compute!

 

[Silverforce11]

Anyway, i still wonder how much of that compute could be done on a IGP that has DX12 support.

That would be very very useful.

I hate the fact that my iGPU is doing nothing when I'm gaming. I wish Intel release an i5 or i7, without iGPU, 50% smaller core, and sell it cheaper.