DOOM with Vulkan Renderer Significantly Faster on AMD GPUs

[Dethroy]

Aside from the obvious gains on AMD's side one can observe something else as well...
According to this test done by PC Games Hardware (updated for the 4th time now) a GTX 980 Ti pulls ahead of a GTX 1070 by ~20% (average of the 4 resolutions that have been tested) thanks to Vulkan.

Vulkan really does utilize architectural advantages way better than the OpenGL implementation does. I wonder what kind of performance gains one would see with Vega...

It will be intersting to see if Pascal's faster pre-emption and its dynamic load balancing (which is Nvidia's current answer to async compute) will achieve similar results once ID - with the help of Nvidia - is done implementing it.

 

[john_]

My bet goes on AMD giving some cash to Vulkan developers so that they can boost their GPU's performance first.

Maybe AMD is also using it's huge bank account to help Google with it's financial problems. That's why Google is adopting Vulkan as the main low level API for Android.

Developers know that the future is low level APIs. And they know that Vulkan could be very crucial in the future giving them the opportunity to make a game not just for consoles and desktops, but for every device out there, including smartphones and tablets. Also Vulkan is the only option right now if you want to offer a game highly optimized in Windows 7 and Linux. Going DirectX 12 means that only users with Windows 10 will see any benefits and going DirectX 11 only is like putting your head in the sand hopping that everyone else will do the same.

 

[Landcross]

I hope that this poor result on Nvidia's GPUs are only because Pascal is still not optimised for Vulkan. It would be highly inappropriate for Khronos to favor AMD in a multiplatform API.

Khronos doesn't favor AMD (or any other manufacturer), it's just that AMD's architecture is better suited for Vulkan. The same way that Nvidia's architecure is often better suited for other APIs.

 

[bug]

Vulkan really does utilize architectural advantages way better than the OpenGL implementation does.

It doesn't, at least not inherently. What it does is it exposes the video card behind the drivers and lets the programmers use it to the fullest. id is top-notch, but other companies may very well come up with half-assed implementations that will make Vulkan slower than OpenGL. Vulkan is still an unknown quantity, imho, but this first step has been executed perfectly.

 

[john_]

Ah, AMD fanboy. About EXACTLY scrubs like you i was talking about earlier.

And compare dicks as much as you want, I wont buy new mobo or processor or ram to get 1% increased in game performance boost.

However don't forget to buy 480, because its so good and modern right? and then write in forums that AMD beats Nvidia everywhere.

Pathetic

As YOU predicted

and make a million and one excuse to argue that

Have a nice day.

 

[laszlo]

Damage Control, incoming...!!

......better quote @Tatty_One before is coming : "children are misbehaving in the nursery yet again, reply bans for this thread will be issued if it continues, followed by free holiday passes"

 

[Dethroy]

It doesn't, at least not inherently. What it does is it exposes the video card behind the drivers and lets the programmers use it to the fullest. id is top-notch, but other companies may very well come up with half-assed implementations that will make Vulkan slower than OpenGL. Vulkan is still an unknown quantity, imho, but this first step has been executed perfectly.

Funny you picked up on that as I wanted to write ID's Vulkan implementation at first instead. But I finally decided to word it differently because I wanted to emphasize that ID was able to do so because of Vulkan and that it wouldn't have been possible with OpenGL. I know that ID's programmers are to thank, but so is Vulkan

Edit: My takeaway... I skip this round of GPUs and await the next generation.

 

[R-T-B]

So, GCN cards are faster in Mantle, DirectX 12 Mantle and also, Vulkan Mantle.

Vulkan is the only one that is proven to use any Mantle code.

If something went on behind the scenes, it can only be said as conjecture. Let's not just spout conjecture as fact. It makes me pissy... and when I'm pissy, it makes me do stupid things, like break my computer.

You wouldn't cause an innocent frog to break his computer, right?

 

[cryohellinc]

As YOU predicted


Have a nice day.

Thank you very much. Likewise and all the best!

 

[bug]

Funny you picked up on that as I wanted to write ID's Vulkan implementation at first instead. But I finally decided to word it differently because I wanted to emphasize that ID was able to do so because of Vulkan and that it wouldn't have been possible with OpenGL. I know that ID's programmers are to thank, but so is Vulkan

Edit: My takeaway... I skip this round of GPUs and await the next generation.

Eh, if you look at Nvidia's results, a proper driver can already use the card to its fullest.
However, yes, Vulkan exists for a reason, it does put more flexibility into the hands of the developer.
Exciting times for sure, after being stuck with 28nm cards since forever.

Edit: I may not skip this generation, my 660Ti starts to show its age. And I say "may" because the 480 doesn't cut it for me. 1060 I think will provide enough HP, but I won't buy it at FE+ prices.

 

[LAN_deRf_HA]

I actually got a decent improvement on my 780 Ti with Vulkan on certain levels but I blame that more on all the weird performance degradation issues Doom has, I think they're related to cache. Also got rid of those late game crashes, which I'm pretty sure are also some how cache related.

 

[PP Mguire]

Vulkan is the only one that is proven to use any Mantle code.

If something went on behind the scenes, it can only be said as conjecture. Let's not just spout conjecture as fact. It makes me pissy... and when I'm pissy, it makes me do stupid things, like break my computer.

You wouldn't cause an innocent frog to break his computer, right?

The bottom line being they're all low level.

 

[Litvan]

Am I the only one wondering why there's not a single 1080 card on that list and only a 1070?

 

[RejZoR]

You didn't? And what exactly is this?



When you quote someone and you are just using his post as an opportunity to make a general comment, don't make questions that appear to be aimed at him.

Ever heard of 3rd person and the use of "you" for those situations (I don't even know what's the official term for it)? Ever thought I was addressing NVIDIA fanboys with "you" and not you directly?

 

[R-T-B]

The bottom line being they're all low level.

Yes, but low level simply means you have to target the hardware at a low level. It does not specify it will perform better on AMD (or NVIDIA), unless the coder specifically targeted AMD. That's most likely what is happening.

 

[Xzibit]

Quoted and fixed.

Does DOOM support asynchronous compute when running on the Vulkan API?

Asynchronous compute is a feature that provides additional performance gains on top of the baseline id Tech 6 Vulkan feature set.

Currently asynchronous compute is only supported on AMD GPUs and requires DOOM Vulkan supported drivers to run. We are working with NVIDIA to enable asynchronous compute preemption in Vulkan on NVIDIA GPUs. We hope to have an update soon.

 

[ShurikN]

This might give you an idea. Although results differ widely...
View attachment 76847

Those are not the same results as the original pic. To me that looks like OGL rather than Vulkan

 

[PP Mguire]

Yes, but low level simply means you have to target the hardware at a low level. It does not specify it will perform better on AMD (or NVIDIA), unless the coder specifically targeted AMD. That's most likely what is happening.

Wasn't exactly the context of his post though. The point was, GCN was made for low level from the get-go.

 

[the54thvoid]

Quoted and fixed.

That's about the size of it. This is the flip happening now. Nvidia need to code more (for Pascal) while RTG can rely on Async hardware.

Unlike DX11 where Nvidia ruled and RTG needed to optimise drivers heavily (why after time they improve performance because they need the optimisations), RTG have a clear edge.

Is the Async hardware really proprietary though? If it is, Nvidia can't do anything about DX12. Also, if Mantle, Vulcan and DX12 is based heavily on proprietary tech, is that 'allowed' under FRAND patents?

 

[ShurikN]

It is Vulkan. But tests were done by PC Games Hardware, not Computer Base.

How is the Fury X then barely faster than a 980, yet in the other bench it's easily beating a 1070. In the article that you linked, the last update was performed 15 days ago.

 

[dj-electric]

Since when TPU turned into WCCF? what the hell is going on in this thread?!

 

[R-T-B]

Wasn't exactly the context of his post though. The point was, GCN was made for low level from the get-go.

You can't "make something" for low level, at least not hardware wise. It goes against the definition of low level. You optimize low level to a platform, you target your hardware. That is what low level means.

 

[laszlo]

http://asawicki.info/news_1601_lower-level_graphics_api_-_what_does_it_mean.html

 

[Dethroy]

Those are not the same results as the original pic. To me that looks like OGL rather than Vulkan

&

How is the Fury X then barely faster than a 980, yet in the other bench it's easily beating a 1070. In the article that you linked, the last update was performed 15 days ago.

You are right. I messed up!
Was refreshing the site for further updates and clicked on a link which I thought to be an update because of the use of the word "final". My appologies!

That's about the size of it. This is the flip happening now. Nvidia need to code more (for Pascal) while RTG can rely on Async hardware.

Unlike DX11 where Nvidia ruled and RTG needed to optimise drivers heavily (why after time they improve performance because they need the optimisations), RTG have a clear edge.

Yup. I wonder if dynamic load balancing coupled with pre-emption will boast a similar performance gain as async compute does (unlikely imho).

 

[R-T-B]

http://asawicki.info/news_1601_lower-level_graphics_api_-_what_does_it_mean.html

Supports my claim. You target the hardware, you don't make a hardware product to be "low level from the getgo"

So lower-level API means just that driver could be smaller and simpler, while upper layers will have more responsibility of manually managing stuff instead of automatic facilities provided by the driver (for example, there is no more DISCARD or NOOVERWRITE flag when mapping a resource in DirectX 12). It also means API is again closer to the actual hardware. Thanks to all that, the usage of GPU can be optimized better by knowing all higher-level details about specific application on the engine level.

You need to understand the hardware you are writing code to run on. The driver no longer babysits you. AMD is doing better now because most lowlevel efforts have been AMD sponsored.