DOOM's Vulkan update is now live

AMD gains about 20-25% just by switching to Vulkan, along with an additional 5-10% with async enabled. This seems to follow past trends.

Nvidia has bigger problems than async's miniscule gains.
 
Saying one supports a feature does not mean they support it at a hardware level. They mean, in the case of Nvidias current lineup, that they emulate the feature via software. There is plenty of proof all over the internet that proves Nvidia lacks true hardware support of asynchronous compute+graphics.
My point is there is no evidence it truly gimps them in any way. there are not enough games properly coded in either dx12 or vulkan to show a performance hit and many benchmarks are showing some small gains and even some large ones in doom plus my own testing showed it either ran the same or a tad better.
 
They sold over 500,000 steam machines, controllers, and links combined. I would think most of the sales were for the controller.

They've been continuously adding features, functionalites and bugfixing/tweaking the controller since it launched last November. Valve has already spoke of a second version in the past but I don't think we'll see it until they really feel they've done everything they can with the existing design. I'd imagine they are going to tie it into the Vive/SteamVR experience as well. It's doesn't exactly tell us anything but the current Vive has a special spot for the controllers dongle so at the very least it tells it was taken into account.

It's one of the, if not the best controllers I've ever used and it just keeps on getting better.
 
Saying one supports a feature does not mean they support it at a hardware level. They mean, in the case of Nvidias current lineup, that they emulate the feature via software. There is plenty of proof all over the internet that proves Nvidia lacks true hardware support of asynchronous compute+graphics.
There is plenty of proof that Nvidia does Asynchronous Compute just fine :rolleyes:, DX 12 does not require you able to do both Asynchronous Compute + GRAPHICS at the same time. AMD has a hardware feature that allows it with DX 12, which is a good feature if used properly to keep the GPU busy/more efficient. The bottom line is what performance you get in the end on a given application, performance/cost and finally perf/w.
 
Really for both AMD and Nvidia, the most performance will most likely come from the ability to use the whole cpu all the time for an application to the gpu vice having a single thread restriction on parts of the process.
 
There is plenty of proof that Nvidia does Asynchronous Compute just fine :rolleyes:, DX 12 does not require you able to do both Asynchronous Compute + GRAPHICS at the same time. AMD has a hardware feature that allows it with DX 12, which is a good feature if used properly to keep the GPU busy/more efficient. The bottom line is what performance you get in the end on a given application, performance/cost and finally perf/w.
Sorry but where is this proof? I have seen plenty of proof with DX12 showing no to negative improvement. And the discussion I was referring to had nothing to do with DX12 spec but his assertion Nvidia could do asynchronous compute and alluding to it being enabled in hardware by stating it was supported (common deflecting of facts too many here take liberties with).
 
It is enabled in hardware, it can do 32 compute operations at once. ASYNCHRONOUS COMPUTE, which really means multiple cpu cores giving out threads for compute which Nvidia can do up to 32 of those threads at once - HARDWARE LEVEL. Can it do that with graphics at the same time - no. What is your point anyways? AMD has a unique hardware feature for this, Nvidia also has unique hardware features as well. Unless it is proven as really advantageous and used they are just features.
 
Sorry but where is this proof? I have seen plenty of proof with DX12 showing no to negative improvement. And the discussion I was referring to had nothing to do with DX12 spec but his assertion Nvidia could do asynchronous compute and alluding to it being enabled in hardware by stating it was supported (common deflecting of facts too many here take liberties with).
Let us pretend for the sake of argument you are right and nvidia does not support a feature that virtually no games use (less than 1% let us say), so what? the raw performance coming out of nvidia cards more than makes up for badly optimized games and in not one game except AOTS does an AMD card show better performance than nvidias top two cards. Your retort will be "but dx12 and vulkan are the future nvidia will hurt then", and you would be right those api are very much the future but we can not predict how good or badly nvidia cards will do on future games on those api when so far the only games that use them are poorly coded and implemented after the fact.
 
Let us pretend for the sake of argument you are right and nvidia does not support a feature that virtually no games use (less than 1% let us say), so what? the raw performance coming out of nvidia cards more than makes up for badly optimized games and in not one game except AOTS does an AMD card show better performance than nvidias top two cards. Your retort will be "but dx12 and vulkan are the future nvidia will hurt then", and you would be right those api are very much the future but we can not predict how good or badly nvidia cards will do on future games on those api when so far the only games that use them are poorly coded and implemented after the fact.
Do you get paid for this? Really? I mean it. So you make the post claiming unfounded points. Then when it is obvious that you made unfounded points you resort to demeaning the basis of your original point as to make it seem meaningless. I made no claims to whether Nvidia will have or not future gains nor their current position nor any doomsday prophecies. Just made the statement that Nvidia can not do asynchronous compute in the context of Compute+graphics simultaneously/asynchronously/at the same time without context switching.
 
It is enabled in hardware, it can do 32 compute operations at once. ASYNCHRONOUS COMPUTE, which really means multiple cpu cores giving out threads for compute which Nvidia can do up to 32 of those threads at once - HARDWARE LEVEL. Can it do that with graphics at the same time - no. What is your point anyways? AMD has a unique hardware feature for this, Nvidia also has unique hardware features as well. Unless it is proven as really advantageous and used they are just features.
Now you are just being obtuse. 99% of the time when posters mention asynchronous compute they do in fact mean compute+graphics simultaneously/asynchronously/at the same time without context switching. You know this. I have seen you in these discussions. But I guess since you now have a 1070...
 
True but Nvidia context switches are much faster then AMD's in addition Nvidia saves much of the data I do believe in buffers to eliminate or reduce memory calls when the context switches back to graphics. Looking at Polaris, it gives three choices for Async Compute, Async Shader (graphics plus compute), Preemption (same as Nvidia), Quick Response (combination of the first two). Any of those methods are Asycn Compute. AMD gives more options. For the console with very week CPU's, Async compute gives some good advantages.
 
Now you are just being obtuse. 99% of the time when posters mention asynchronous compute they do in fact mean compute+graphics simultaneously/asynchronously/at the same time without context switching. You know this. I have seen you in these discussions. But I guess since you now have a 1070...
I don't assume and async compute is not compute+graphics at all. AMD just added a feature on top of that which can be useful.

So what if I have a 1070? Your argument is totally pointless. Go look up on Microsoft what Async Compute is - note it has nothing to do with async + graphics. It does mention some unique hardware capabilities which the developer now has access.
 
I have a more general question...

WTF is doom/vulkan doing in general to get such phenomenal performance across the board? Are these shooter games just easier to render? I don't ever remember any dragon age or witcher game running this well. Are those games just doing more stuff in the background? Because doom looks as good if not better. Maybe it's just more simplified/smaller environments.
 
Do you get paid for this? Really? I mean it. So you make the post claiming unfounded points. Then when it is obvious that you made unfounded points you resort to demeaning the basis of your original point as to make it seem meaningless. I made no claims to whether Nvidia will have or not future gains nor their current position nor any doomsday prophecies. Just made the statement that Nvidia can not do asynchronous compute in the context of Compute+graphics simultaneously/asynchronously/at the same time without context switching.
I wouldn't mind getting paid for posts. Kidding aside I never made unfounded points, all i said is that nvidia said they have async compute, no one can prove otherwise, and lastly there is little proof that their implementation if different matters at this point.
 
the doom vulkan patch was more exciting than any of the recent gpu releases, 1080 included. Now THAT's how you deliver more performance
 
I have a more general question...

WTF is doom/vulkan doing in general to get such phenomenal performance across the board? Are these shooter games just easier to render? I don't ever remember any dragon age or witcher game running this well. Are those games just doing more stuff in the background? Because doom looks as good if not better. Maybe it's just more simplified/smaller environments.
That's a really good question. The game is gorgeous and considering how bad id tech 5 was I was worried for this title. They really worked the magic voodoo in id tech 6.
 
  • Like
Reactions: noko
like this
6700k and 1080
Here's one of the heavier areas I know about in the game

gl.jpg

vk.jpg
 
6700k and 1080
Here's one of the heavier areas I know about in the game

gl.jpg

vk.jpg
You forgot the part where you are supposed to show the 1080 doing worse on Vulkan since that is what the internet says it does.
 
It's kinda late to fuck around more but here's the gist of my 45 minutes or so of experimenting. TLDR: It's a LARGE performance gain for free at best, about the same at worst.


Vulkan can easily chop several milliseconds off your CPU time in some cases. In heavier scenes I'm seeing nearly 3ms~ difference at times. That's the difference between 125 fps and 200. In CPU heavy situations.

On the other hand, it doesn't really seem to help the 1080 much on the raw GPU performance front. I guess AMD will have more to gain here with async compute.
 
It's kinda late to fuck around more but here's the gist of my 45 minutes or so of experimenting. TLDR: It's a LARGE performance gain for free at best, about the same at worst.


Vulkan can easily chop several milliseconds off your CPU time in some cases. In heavier scenes I'm seeing nearly 3ms~ difference at times. That's the difference between 125 fps and 200. In CPU heavy situations.

On the other hand, it doesn't really seem to help the 1080 much on the raw GPU performance front. I guess AMD will have more to gain here with async compute.
it could be some type of performance ceiling in the game engine the 1080 is hitting.
 
What OSD is that ? CPU, GPU usage with FPS counter.

The OSD is built into the game. There are multiple settings for it based on how much data you want to look at.

My i5 feels pretty weak when looking at how well all the threads on that i7 are getting used!
 
That is one hell of a good fps increase. I will have to try it out on my 7950.
 
Last edited:
AMD gains about 20-25% just by switching to Vulkan, along with an additional 5-10% with async enabled. This seems to follow past trends.

Nvidia has bigger problems than async's miniscule gains.

Pay attention that Doom is also using the GCN shader extensions, so every GCN-based Radeon use the vendor specific fast shaders, while the other GPUs use the standard SPIR-V shaders.
 
That's a really good question. The game is gorgeous and considering how bad id tech 5 was I was worried for this title. They really worked the magic voodoo in id tech 6.
What do you mean id tech 5 was terrible? That engine pushed 1080p 60fps on the Xbox 360 and PS3 and looked great doing it. Same thing running Wolfenstein New Order and that also looked great and ran great.

Been playing RAGE recently, in anticipation of this patch (which on a GTX 680 just results in the game hanging on level changes), and the game really is rather beautiful - no, it's not super high res and no it doesn't use a ton of fancy things. But it looks great, like a welldrawn, highly detailed comic book, has beautiful animations and runs at a completely smooth framerate all the time. Yes, there's some texture pop-in if you twirl really fast in newly loaded areas, but on a fast system it's almost nonexistent.
 
New Patch Brings Vulkan support to Doom + Benchmarks

Did he run this properly on Nvidia? I did not see the AA settings....
Showing no gains yet on 1070
The problem is that we do not know if it is a driver-pascal related issue, or settings including both AA and shadows, or the zone Guru3D tested.
Just a shame Guru3d did not test a 980ti along with Pascal.

Edit:
Well reading some other posts looks like 1080FE can also boost with Vulkan.
So comes down to what settings Guru3D used and comparing maps-zones, will be interesting to see what causes the difference on Nvidia.
Cheers
 
Last edited:
What do you mean id tech 5 was terrible? That engine pushed 1080p 60fps on the Xbox 360 and PS3 and looked great doing it. Same thing running Wolfenstein New Order and that also looked great and ran great.

Been playing RAGE recently, in anticipation of this patch (which on a GTX 680 just results in the game hanging on level changes), and the game really is rather beautiful - no, it's not super high res and no it doesn't use a ton of fancy things. But it looks great, like a welldrawn, highly detailed comic book, has beautiful animations and runs at a completely smooth framerate all the time. Yes, there's some texture pop-in if you twirl really fast in newly loaded areas, but on a fast system it's almost nonexistent.
Texture pop and load times, also the super texture format made everything look bland and fairly low res. Now those are certainly good games, but in the hierarchy of great id engines id tech 5 doesn't rank high IMO.
 
These low-level APIs ae just what AMD needed to stay competitive as they just don't have the resources to keep up with DX11 driver optimization.

Hopefully that means more competition and better games in the future. Assuming they don't find some other way to screw the pooch.
 
GamersNexus

DOOM Vulkan vs. OpenGL Benchmark – RX 480 Posts 30% Increase, GTX 1080 Mixed

Looks like Vulkan bumped the 480 up to where it should've been in the first place, Pascal scales 20% and lower as resolution increases. 480 stays steady. Maxwell no scaling.
Their OpenGL results for the 480 seem lower than other sites, though.

They used FXAA which disabled async.

This still puts 1080 behind what it should be in relation to 970 though, but at least it's not that far behind.
 
Nvidia has bigger problems than async's miniscule gains.

Like what?

I have a more general question...

WTF is doom/vulkan doing in general to get such phenomenal performance across the board? Are these shooter games just easier to render? I don't ever remember any dragon age or witcher game running this well. Are those games just doing more stuff in the background? Because doom looks as good if not better. Maybe it's just more simplified/smaller environments.

It's a combination of (a) OpenGL, (b) engine developers that know what they're doing, (c) developers who, from the get-go, treated the PC platform as something other than a red-headed bastard step-child, and (d) the fact that they're only using a 16-color CGA palette.
 
Tried it this morning, went back to OpenGL. Framerates look good but I am seeing an inexplicable stutter/juddering that makes the experience less smooth overall. YMMV.
 
Tried it this morning, went back to OpenGL. Framerates look good but I am seeing an inexplicable stutter/juddering that makes the experience less smooth overall. YMMV.
I was reading about an issue with latency affecting some users of the 1070/1080s. Seemed like the issue you're speaking of.
 
You forgot the part where you are supposed to show the 1080 doing worse on Vulkan since that is what the internet says it does.


I'm getting about 25% WORSE performance on the Vulkan render path vs. OGL on my 1080. I am running 4k vs. 1080p though and I think that's where the problem lies right now.
 
I wonder what's different?

I seem to only get a solid performance boost, no stuttering.
 
Tried it this morning, went back to OpenGL. Framerates look good but I am seeing an inexplicable stutter/juddering that makes the experience less smooth overall. YMMV.

I noticed a similar stuttering as well on my system. Hopefully the next Nvidia driver addresses this.
 
Back
Top