DOOM with Vulkan Renderer Significantly Faster on AMD GPUs

[xkm1948]

I just wish more VR applications employ the new Vulkan API. VR is the future.

I am also looking into improvement using AS for scientific computing. It would be awesome if OpenCL can benefit from these as well.

 

[R-T-B]

That's just flat-out not true. There is a difference between optimizing a game to be good at things like tesselation/texturing/lighting, and just straight up not putting hardware like RAM/ACE's/SP's in a card.

From the perspective of the API, no there is not. AMD is severely constrained in tesselation. You could counter your arguement by saying NVIDIA hardware is ready for future games featuring massive tesselation, and AMD just lacks the hardware to compete, etc

That AMD has better compute (they do) does not make them better in low level apis. It makes them better at games that ufilize compute heavily. The API? The API just facilitates access to the hardware. It has no brand loyalties.

 

[ur1manphantom]

Salty nvidia users

 

[the54thvoid]

I guess I hear champagne pops coming from the red camp

Because their top range cards beat an over priced smaller mode process mid range card? This is Nvidia's fault for pricing their cards where they do.

If the GTX1080 was priced to replace the 980 it architecturally replaces, it wouldn't be such a surprise a card with Fiji's raw power does so well. Fury X has a whole shed load of processing power and finally we're seeing it being used. On paper it's more than a match for what Nvidia has at it's top end for now. The problem still remains - where is gaming going? Is it DX12, is it Vulkan? How long is DX11 staying around?

Undoubtedly the improvements shown in this Vulkan game show AMD to be fantastic. The GTX 1080 does actually put in a very good show and that is the 980 replacement. Also recall that Fury X is still (where available) a >£500 card.

What people should still realise (and so few do due to their differing views) is that nothing has changed on the generic landscape yet. If we all knew that Vulkan was the way forward and it was going to be used in everything, shit, we'd all go and buy AMD tomorrow. But we don't know what's happening on the gaming dev front. I could go out tomorrow and try and buy a Fury X and then find out some of the new games have went with an Nvidia sponsored DX12 title and lose out to an optimised Maxwell/Pascal.
Ditto with staying Nvidia. It's shit that instead of one playing field (DX11), there's now possibly 3 (or more). This is not the best environment to purchase in. We need some stability from dev's to decide which is the likely way forward

Salty nvidia users

Well that's okay because salt is not bitter. Unlike others.

 

[cdawall]

Well this thread turned into a shit storm. Why are people upset that the API improved performance on AMD cards? Everyone knew this was coming since mantle/DX12/vulkan was known to exist. If you purchased NV and planned to have the card forever, plan to have mid-range GCN products perform better.

 

[ShurikN]

Well this thread turned into a shit storm. Why are people upset that the API improved performance on AMD cards? Everyone knew this was coming since mantle/DX12/vulkan was known to exist. If you purchased NV and planned to have the card forever, plan to have mid-range GCN products perform better.

If you pay close attention it's only NV fanboys that are upset. bust shhhh dont tell them, they might get agitated

 

[laszlo]

Has anyone tried it on HD7XXX cards?

a asked also..thought someone here will try and let us know...

i find the answer however... in steam forum ..:

"I'm using an older HD7970 and it's running like a dream so far with Vulkan at 1080p with a mixture of high to ultra settings (and medium shadows, but I haven't tweaked the graphics settings much from default to be honest). Well impressed that I'm able to run games like this at 60-80 fps consistently on a 5-year-old card. "


seems is working like a charm and older cgn stuff still good stuff (to a certain point) 2 bad my actual card is not cgn..waiting custom 480 from aib....

 

[Slizzo]

Those gains are probably due to Vulkan handling CPU bottlenecks much better.

No doubt. People aren't really understanding that DX12 and Vulkan go a long way towards helping along bottlenecks that are present in systems far more than if you used the same APIs on a brand new, top of the line, very powerful system. There isn't much in the way of a bottleneck in that case, and these new APIs won't return too much performance on an already streamlined system.

 

[FordGT90Concept]

The only problem with HD7XXX cards is they have two ACEs instead of eight. Not sure how much of a performance impact that would be but HD7XXX will no doubt benefit from Vulkan because it is still GCN.

 

[efikkan]

A benchmark with temporal antialiasing? Who cares? Run proper AA and see how things actually scales.

Not sure why there is so much talk here about async shaders. It's not clear that Doom even uses them.

Async shaders isn't implemented in Vulkan yet, so clearly it's not used.
But as usual, clueless people are the quickest to make assumptions.

This is mostly about Vulkan being faster than OpenGL (which we knew) and, in AMD's case, Vulkan is a lot faster than OpenGL because AMD never had the best support of OpenGL in the first place.

To this date AMD has never bothered to implement proper OpenGL support, so clearly they can get higher "gains" from Vulkan.

 

[TheoneandonlyMrK]

A benchmark with temporal antialiasing? Who cares? Run proper AA and see how things actually scales.


Async shaders isn't implemented in Vulkan yet, so clearly it's not used.
But as usual, clueless people are the quickest to make assumptions.


To this date AMD has never bothered to implement proper OpenGL support, so clearly they can get higher "gains" from Vulkan.

You sure about that , I copied this from Bethesda doom faq.


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 in Vulkan on NVIDIA GPUs. We hope to have an update soon.

Soooo yeah but not if you're running the wrong setup for it.

It's also clear that if they weren't paying out enough love to open Gl they are now.

And I personally think vulkan has got bite and backing ,ie apple ,Google , PC and I'd expect its implementations to work well on console

Have to admit I've chatted some poop over the years re Amd on tpu but its nice to see some of it is as I said it would be.

Damn this pone

 

[RejZoR]

We aren't even comparing Radeons to Radeons here, OGL vs Vulkan. We are comparing NVIDIA to AMD and AMD takes a huge edge here even with R9 Fury which as you know wasn't a real competitor against GTX 980Ti. But now it goes against latest GTX 1070. Which is around GTX 980Ti level. Not bad I'd say. And per your word, not even using Async...

 

[FordGT90Concept]

We need more titles of the DX12/Vulkan flavor before we draw any broad conclusions. That said, I don't think I ever seen benchmarks of Talos Principle running Vulkan and it was the first Vulkan title.

Damn this pone

 

[Dethroy]

We need more titles of the DX12/Vulkan flavor before we draw any broad conclusions.

And I'd also like to see if Pascal's Dynamic Load Balancing and Pre-Emption translate to real world performance gains and if these gains are anywhere close to what Async Compute & Shaders do for GCN.

Having said that, I really like the direction we are heading to (eliminating bottlenecks & general performance gains thanks to better utilization of hardware architectures/features) and everyone -AMD & Nvidia fanboys alike - should be happy about said direction. Certainly, interesting and exciting times lie ahead of us!

 

[$ReaPeR$]

holy crap those gains!!!! finally the future is here.

 

[the54thvoid]

Hmmm.

The weird thing is there a host of youtube vids with 1080's at 1440p nightmare gfx settings on Vulkan running at around 150fps.

We'll need to get the 1070/1080 owners thread to run the benchmark. See what the overpriced mid tier 980 replacement does.

 

[Slizzo]

Hmmm.

The weird thing is there a host of youtube vids with 1080's at 1440p nightmare gfx settings on Vulkan running at around 150fps.

We'll need to get the 1070/1080 owners thread to run the benchmark. See what the overpriced mid tier 980 replacement does.

Certain areas I definitely make around that in framerate.

 

[Dethroy]

You are late to the party

Interesting read &
Another interesting video...

 

[nem..]

Amd -> <- Nvidia

 

[PP Mguire]

If someone wrote a low-level api game that utilized tons of nvidia friendly tessellation, nvidia would suddenly appear to be awesomely optimized for low level as well.

The fact is you don't gear an architecture to be low level. You gear your game for an architecture.

Still not really the context of the original post and quote, but who cares really? I have shitty old Maxwell and it still plays Doom damn good at 4k. More FPS for cheaper cards is a win/win in my books. I would rather see this on other games that slam GPUs, like Witcher 3 or the devs for ARK properly implement Vulkan or DX12 so performance isn't so shit. A guy can dream though

 

[Aquinus]

So, TSSAA (8x) plus Ultra and 16x AF, I'm seeing no lower than 90FPS at 1080p with the average being over 100 but under 115. The GPU is constantly pegged out and my 390 is running a lot hotter than it normally does at full load, normally I'll see ~70*C under load but, I'm seeing about 80*C with this. I would say, that Vulkan is doing a pretty good job at tapping those unused resources. My only complaint is that TSSAA, even at 8x is kind of meh. Either way, I'm impressed with what I'm seeing. I think I'm going to see how eyefinity runs. I'll report back.

Edit: I'm noticing some flickering at far distances. It could be the TSSAA, I guess.
Edit 2: Flickering went away when I turned off AA. Funny thing is that frame-rate didn't change much after I disabled it.
Edit 3: It appears that TSSAA (8x) seems to be the only mode that causes flickering.
Edit 4: Same scene does about 39-60FPS on 5760x1080 which is far better than I was seeing before but, it isn't what I would consider acceptable. This is with the same settings as at 1080p, Ultra + 16x AF + 8x TSSAA. This is also with no overclock on my GPU with the stock 1040/1500(6000).

 

[Jism]

There's alot of 'compute' potential on AMD GPU's in general, this is why they are favoured in for example the Bitcoin mining thing. They provide a much higher hashrate then nvidia cards. What you are seeing here is exploit of full potential of AMD gpu's, so if your avg AMD GPU is running 10 degrees hotter then usual then it means you are fully utilizing it.

Vulkan should be adopted in ANY game, since it offers so much more then the standard DX11/12 and / or OpenGL. It means that we get higher FPS from even 2 generations of cards before, which contain the GCN featureset.

AMD was'nt stupid when it developped the Async stuff into their stuff. AMD was'nt stupid either that the future contains multiple cores in both CPU and GPU camps and it is betting huge. By doing so and pushing Mantle for what it was, it just opened up a new feature. Remember almost any generation of console now and future, will proberly be driven by AMD GPU's. The vulkan thing is something that we will see more often.

 

[R-T-B]

Still not really the context of the original post and quote, but who cares really?

I don't honestly. I just have a better understanding of what "low level" means being a programmer, and am trying to share my knowledge. Async != low level. I will leave it at that for now.

 

[Jism]

Low level is without any abstract(s) such as DX as a layer in between. This offers extra performance and spares CPU-cycles from which gains can be mostly seen at lower-end systems.

If anyone remembers the Amiga 1200,

That thing had less then 2MB of ram, and not even a GPU but an AGA capable graphics chip with up to 256 colors, lol. But still programmers managed to sqeeze graphics and tech demo's out of that peace of antique like never before. It was because they understood coding to the metal. Same as the PS2, which is RISC based and on some games really shined and where way ahead of other games, with only 4MB of video-ram and 32MB of total system ram.

Even Nasa builds hardware that's based upon 8086 processors and head into space. It's because they understand the logic(s) and power of a chip. The PS3 contains a complete different CPU platform, but still manages to produce the very best graphics best in it's time. Just straight coding to the metal.

 

[looncraz]

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

Also known as the future of PC gaming. ..