DOOM updated with Vulkan support

[Maverick177]

Well, ROTR game developers wrote an update for it.

Seems odd to write an update for something that doesn't exist or work?



https://steamcommunity.com/app/391220/discussions/0/358417008720278827/

Maxwell is only 1 year and 9 months old, where is it?

http://i.imgur.com/uY1lCak.jpg

Pascal actually lost FPS switching to Async Compute Patch, while Fiji gains some, why is that?

I don't know, why don't you direct these questions to NVIDIA? Please do that and come back with answers for us forum goers

I did, but with no response. When I can't find a quote from nVidia, I go the forums to lurk.

 

[Arachnotronic]

was that why performance was lowered across the board for that game?

I got a nearly 30% speedup in RoTR on my 1080 SLI setup with the latest DX12 patch relative to the DX11 codepath. I have a pretty fast CPU, too, so I don't know if it's all just "removal of CPU bottleneck."

 

[Bacon1]

Well, ROTR game developers wrote an update for it.

Seems odd to write an update for something that doesn't exist or work?



https://steamcommunity.com/app/391220/discussions/0/358417008720278827/

Sure, but even on Hawaii the gain from enabling async compute is 1% or margin of error. No one has tested Async Compute enabled vs disabled on Pascal even though I've asked them to test it.

Nvidia said that Maxwell supported Async Compute for a long time, and now they've tried to ignore the issue and hope everyone forgets. I wouldn't doubt we see the same issue again with Pascal.

I honestly hope to see Pascal AND Maxwell support it, because it is FREE performance for everyone and more developers spend time developing for it.

However, until Nvidia can prove that they have it and stop faking support: https://youtu.be/UKkFqG77-x4

I'm going to doubt that they do truly support it.

 

[Bacon1]

I got a nearly 30% speedup in RoTR on my 1080 SLI setup with the latest DX12 patch relative to the DX11 codepath. I have a pretty fast CPU, too, so I don't know if it's all just "removal of CPU bottleneck."

Test it with Async Compute enabled and test with it disabled. Its simple 10 minute test.

 

[PhonakV30]

That was February, though?

Well , Why 1 years ? Just tell me WHY? Nvidia had 1 years yet still no Async Compute?

 

[LTC8K6]

https://www.computerbase.de/2016-05...ramm-ashes-of-the-singularity-async-compute_2

There is AOTS with some comparisons from a little while ago.

 

[garagisti]

They are pushing it hard, see Twitter. Also, in the Bethesda PR for this patch, there is a link to an AMD Youtube video explaining Async Compute.

AMD's fingers are all over this, good marketing move from them.

Well yes, they developed mantle, which is what Vulkan is based on and so is DX12. I hope you're not going to equate developing new standards to something as shady as Gameworks. I say this, as a lot of people do like to muddy waters.

 

[LTC8K6]

Well , Why 1 years ? Just tell me WHY? Nvidia had 1 years yet still no Async Compute?

I'm not the board NV rep...

My favorite NV card is the GDDR5 GT730.

 

[Maverick177]

https://www.computerbase.de/2016-05...ramm-ashes-of-the-singularity-async-compute_2

There is AOTS with some comparisons from a little while ago.

It's 3% at best scenario. I certainly hope it's not within margin of errors.

 

[Bacon1]

https://www.computerbase.de/2016-05...ramm-ashes-of-the-singularity-async-compute_2

There is AOTS with some comparisons from a little while ago.

Yes, @ 4k Pascal is slower while fury x gains 10%. 1080 is 1% (margin of error) difference.

None of the gains or loss are anything more than margin of error since the test runs differently each time.

 

[garagisti]

https://twitter.com/Xathian1/status/752951596074565632

lol, this guy is rubbing it in. However, 7970(Hawaii) was a compute monster anyways.

 

[LTC8K6]

It's 3% at best scenario. I certainly hope it's not within margin of errors.

I'm sure there will be improvements as drivers mature and games mature.

 

[LTC8K6]

Yes, @ 4k Pascal is slower while fury x gains 10%. 1080 is 1% (margin of error) difference.

None of the gains or loss are anything more than margin of error since the test runs differently each time.

Looks faster to me.

At any rate, that test was two months ago.

 

[krumme]

https://twitter.com/Xathian1/status/752951596074565632

lol, this guy is rubbing it in. However, 7970(Hawaii) was a compute monster anyways.

An 7970 have 2 ace. And 480 have 4 - or something similar to 4, as I recall.
I have a hard time accepting asynch is the sole reason for the jump but surely if the game uses the ace and compute power an 7970 will walk over an 980 ti. Now an Fury will just beat 1070. And that's just what is happening.

 

[Bacon1]

Looks faster to me.

At any rate, that test was two months ago.

Again, 3% faster, 3% slower, 1% faster... not what I'd really call working well if at all considering the test isn't synthetic.

I want async compute to work on all cards so more developers take the time to utilize it so everyone gets more fps.

 

[Grooveriding]

I'm not buying it from nvidia. They're not ones to let AMD have a standing lead or better showing if they can help it. There was supposed to be async compute for Maxwell and a driver delivered to enable it over a year ago. That never happened, one of several marketing lies from nvidia.

A Fury X is performing like a GTX 1080 using Vulkan in Doom

I doubt very much we will ever see that mythical async compute driver for Maxwell, but we might see something for Pascal, maybe. I don't expect it will deliver anywhere near the gains AMD is somehow getting though. Nvidia has a lot more resources than AMD, whether that be drivers or to deal with game developers, and they don't have anything ready for this release ? The chance of them realizing the sort of gains AMD seems to be getting out of Vulkan and DX12 are slim to none I think.

They are probably continuing to run with simply having the faster processor like they always have and using that to maintain their leads. If AMD had actually released a decently sized flagship GPU they'd have likely left the current Pascal GPUs in the dust for Doom. As it is a Fury X, on 28nm compared to 16nm FF, is delivering GTX 1080 performance in the game... That is really impressive.

https://www.computerbase.de/2016-07...md-nvidia/#diagramm-doom-mit-vulkan-2560-1440

 

[Yakk]

Fury X

/Beastmode ON

 

[Yakk]

https://twitter.com/Xathian1/status/752951596074565632

lol, this guy is rubbing it in. However, 7970(Hawaii) was a compute monster anyways.

Tahiti, 1 generation older than Hawaii.

And... ouch! To that tweet!

 

[Silverforce11]

https://www.computerbase.de/2016-07...md-nvidia/#diagramm-doom-mit-vulkan-3840-2160

What's happening? If DX12/Vulkan is the future at this rate the RX 470 will be the one competing with a GTX 1060 and it won't even be close between the GTX 1060 and RX 480.

Surely, Nvidia is sandbagging right? I cannot believe that their cards are making no meaningful (and sometimes worse off) performance increases with DX12 and Vulkan. AMD is running riot meanwhile.

Why do you find it hard to believe?

https://www.khronos.org/assets/uploads/developers/library/2015-gdc/Khronos-Vulkan-GDC_Mar15.pdf

This is from a presentation at SIGGRAPH 2015. This specific presentation was given by Tim Foley (NV Corp).

Pay attention.

Now if that wasn't enough, remember these?

Basically DX12 and Vulkan (& OSX's Metal) are so close, and since Vulkan is built upon Mantle, next-gen API is AMD code with a new brand/logo. You are all running AMD code if you are playing DX12 or Vulkan games.

I told you guys years ago when Mantle was announced, that in a few years, you would all be running AMD code whether you like it or not. Some people refused to believe it, but yet here we are, on the cusp of the DX12/Vulkan era!

 

[Silverforce11]

AMD doesn't have a lot of money for R&D, but they compete fiercely to setup a future where their intrinsic hardware design choices will shine. It's been years in the making, but as soon as the next-gen console APUs were being developed, it provided AMD a huge opportunity to change the course of PC gaming.

How best to get the ecosystem to benefit their hardware? By building the foundation of next-gen APIs and giving it away for free. First to Apple with Metal, then to Kronos, this forced MS into action because these are the biggest threat to the Windows ecosystem (DX). MS rushed DX12 so much the programming guide was lifted, entire paragraphs, from the Mantle programming guide that AMD wrote. They had to get it out first and not come behind Vulkan, but it was possible to come from behind, since Vulkan is cross-platform, with a lot more partners all with an interest so it's development took longer than DX12.

Yet here we are today, and I hope Vulkan is the future, it is open, cross-platform and no UWP restrictions. It needs big players to champion it, get into Unity, Unreal Engine, the popular indie engines. But it looks like Vulkan will lose to DX12, with most of the AAA titles upcoming confirmed DX12 rather than using Vulkan.

Since Polaris is confirmed to be in next-gen consoles, anytime somebody ask for advice on future proofing, you have to be completely ignorant to even suggest that NV GPUs are more future proof at this point. Certainly after knowing more about DX12/Vulkan and it's roots.

 

[RampantAndroid]

You are all running AMD code if you are playing DX12 or Vulkan games.

This keeps coming up...

I believe that it was during the developer talks for DX12 two years ago that it was said that in fact, DX12 was about bringing some of the custom work done for the XBox One to the PC (and phones, if anyone cares about windows phone 10.) The Xbox itself was released in 2013, the same year Mantle's development started.

AMD and Microsoft probably worked pretty closely on the XBox, and the learnings from working with the XBox's GCN setup probably helped shape DX12 a fair bit (and helped get AMD started on the path that lead to Mantle.)

But to suggest that DX12 is AMD code is...well, not supportable by evidence that I've ever seen.

If you state that "Microsoft started with the XBox's GCN architecture, and that might have started them on the path to DX12, and so DX12 was designed with GCN in mind" you would be making a far better statement.

 

[geoxile]

This keeps coming up...

I believe that it was during the developer talks for DX12 two years ago that it was said that in fact, DX12 was about bringing some of the custom work done for the XBox One to the PC (and phones, if anyone cares about windows phone 10.) The Xbox itself was released in 2013, the same year Mantle's development started.

AMD and Microsoft probably worked pretty closely on the XBox, and the learnings from working with the XBox's GCN setup probably helped shape DX12 a fair bit (and helped get AMD started on the path that lead to Mantle.)

But to suggest that DX12 is AMD code is...well, not supportable by evidence that I've ever seen.

If you state that "Microsoft started with the XBox's GCN architecture, and that might have started them on the path to DX12, and so DX12 was designed with GCN in mind" you would be making a far better statement.

Johan Andersson from EA DICE is the designer of Mantle. He started working on it in 2008 long before any of this console-like API on PC paradigm was considered by MS. AMD came later and gave it a platform and according to AMD, they were sharing the details with MS every step of the way. DX12 on the X1 didn't even come until much much later and the Metro 2033 devs complained that MS had basically provided them with a mutant DX11 API while Sony was giving them a low-level API for the PS4 (some sort of low-level OpenGL based API IIRC) and the difference could be orders of magnitudes in some cases.

It's pretty clear DX12 is Johan's API which he worked hard to promote.

http://www.heise.de/newsticker/meld...ber-AMDs-3D-Schnittstelle-Mantle-2045398.html
From 2013

c't: The Community and even journalists were surprised by the announcement of the 3D interface Mantle end of September, 2013. When work began on Mantle?

Andersson: About five years ago we started to think about it. Back then we had already gained experience with consoles and also wanted for the PC, a similar type of access and programmability. We talked with several companies, including Intel and Nvidia.

c't: So you have virtually laid the foundation for Mantle?

Andersson: Of course not I alone, but I was probably the one who has placed the most from this stuff. Because we develop high-end games for the PC, we have a certain influence in the industry. Mantle meant for us at DICE to create a completely different rendering backend. For AMD it was build a driver team that puts its resources and time in the development of an alternative 3D interface. These are all great steps and I have to pay respect to AMD that they have implemented my suggestions. Because none of the other manufacturers did.

 

[Silverforce11]

If you state that "Microsoft started with the XBox's GCN architecture, and that might have started them on the path to DX12, and so DX12 was designed with GCN in mind" you would be making a far better statement.

Go to SIGGRAPH's site, read a few presentations on Vulkan and DX12, some of them go indepth to the actual code examples and it's the same. We know for a fact that Vulkan is Mantle based, if it's highly similar to DX12, as major devs claim, then DX12 is running Mantle-like code. Whether it's directly from AMD, or in collaboration with MS via Xbox, is semantics. The end result is the same: next-gen APIs are built for AMD's GCN architecture to shine.

ps. Credit to Johan Andersson ofc.

 

[RampantAndroid]

Go to SIGGRAPH's site, read a few presentations on Vulkan and DX12, some of them go indepth to the actual code examples and it's the same. We know for a fact that Vulkan is Mantle based, if it's highly similar to DX12, as major devs claim, then DX12 is running Mantle-like code. Whether it's directly from AMD, or in collaboration with MS via Xbox, is semantics. The end result is the same: next-gen APIs are built for AMD's GCN architecture to shine.

Except it isn't semantics, as you claimed that running DX12 meant you were running code from AMD.

Johan Andersson from EA DICE is the designer of Mantle. He started working on it in 2008 long before any of this console-like API on PC paradigm was considered by MS. AMD came later and gave it a platform and according to AMD, they were sharing the details with MS every step of the way. DX12 on the X1 didn't even come until much much later and the Metro 2033 devs complained that MS had basically provided them with a mutant DX11 API while Sony was giving them a low-level API for the PS4 (some sort of low-level OpenGL based API IIRC) and the difference could be orders of magnitudes in some cases.

It's pretty clear DX12 is Johan's API which he worked hard to promote.

http://www.heise.de/newsticker/meld...ber-AMDs-3D-Schnittstelle-Mantle-2045398.html
From 2013

This article seems pertinent: http://techreport.com/review/26239/a-closer-look-at-directx-12

He started on a path that resulted in Mantle, yes. But before Mantle existed, he was just talking about where AMD & NVidia should be going (and possibly Microsoft as well, but I haven't seen that stated clearly.) None of this proves that DX12 is in fact a copy/paste of Mantle. To be clear - they do have much of the same lineage. It seems AMD played a part in that quite prominently, and that is paying off quite well for them right now. I'm not trying to argue some silly AMD versus Nvidia angle, nor am I trying to discredit Andersson or downplay his involvement. I look forward to the future and having proper competition in the GPU space again.

I wish there was a rx490 right now, I'd be seriously thinking about getting one...

 

[dogen1]

Sure, but even on Hawaii the gain from enabling async compute is 1% or margin of error. No one has tested Async Compute enabled vs disabled on Pascal even though I've asked them to test it.

Nvidia said that Maxwell supported Async Compute for a long time, and now they've tried to ignore the issue and hope everyone forgets. I wouldn't doubt we see the same issue again with Pascal.

I honestly hope to see Pascal AND Maxwell support it, because it is FREE performance for everyone and more developers spend time developing for it.

However, until Nvidia can prove that they have it and stop faking support:

I'm going to doubt that they do truly support it.

Do we know that's fake? This is possible in DX11 as well, if that's what you're going off of.