First DirectX 12 game benchmarked *Update 2 More benchmarks

[Curufinwe_wins]

Thank you, updated the post.

That article is pretty damning for NVidia: Not only does the Fury X beat the 980ti, especially in MSAA, but official Oxide Game statements, debunks NVidia's claim of bugged AA, as false.

Really? I'm not being nvidia fanboy here but these results are basically a complete wash in fury x vs 980ti. Furthermore, oxides comments come out harsh but then they backtrack and claim they still have work to do on msaa optimization (which would make nvidia's comments true even if in bad taste.)

Debunked is an idiotic and unprofessional way to say that oxide countered the claim.

Example:

Person 1: the sky is blue!

Person 2: The sky is red!

News flash!!! Person 1's claim debunked by heresay!

Whether or not someone claims something else has no bearing at all on whether or not it is true.

Seriously though if you want to comment on a news article I would very much appreciate it if you worked to keep your own biases out of the story and presented the information in the most objective way possible.

[Yoinkerman]

By the look of the benchmarks Nvidia seems to be whining because they don't have as much of a jump

[Notional]

Really? I'm not being nvidia fanboy here but these results are basically a complete wash in fury x vs 980ti. Furthermore, oxides comments come out harsh but then they backtrack and claim they still have work to do on msaa optimization (which would make nvidia's comments true even if in bad taste.)

Debunked is an idiotic and unprofessional way to say that oxide countered the claim.

Example:

Person 1: the sky is blue!

Person 2: The sky is red!

News flash!!! Person 1's claim debunked by heresay!

Whether or not someone claims something else has no bearing at all on whether or not it is true.

 

Fury X and 980ti is basically neck and neck in those charts, with AMD having better scores overall.

 

Oxide claim their code has been validated. AMD has not made any accusation of bugged code in the AA, and NVidia does not seem to think so in DX11 applications (where they outperform AMD). Extremetech did not confirm any bugs either for any of the GPU vendors.

 

Optimization is inherently different to buggy. One is performance related, the other function related (as in defective). Your example makes no sense in this context.

[ACatWithThumbs]

I don't understand why Nvidia reacted that way when the benchmarks are pretty much showing that they have better DX11 drivers than AMD which they've been talking about ever since Mantle and the Star Swarm demo.
That shows even more as the DX12 results are exactly how they should be.
 

[BuckGup]

Nvidia being slower on high on dx12 while being faster on dx11 on high, looks like they may need to do some optimizations.

the r9 390x results are impressive but i'm sure amd could push it much further still.

You are looking at it in the wrong way as Nvidia optimized DX11 so much that the difference is very little. As time goes on you will see great leaps from Nvidia as for AMD it helps them but goes to show how bad they were

[Curufinwe_wins]

Fury X and 980ti is basically neck and neck in those charts, with AMD having better scores overall.

Oxide claim their code has been validated. AMD has not made any accusation of bugged code in the AA, and NVidia does not seem to think so in DX11 applications (where they outperform AMD). Extremetech did not confirm any bugs either for any of the GPU vendors.

Optimization is inherently different to buggy. One is performance related, the other function related (as in defective). Your example makes no sense in this context.

Many if not most bugs result in performance related issues. Like for example hairworks in witcher3 forcing 32x tesselation is/was a bug. It resulted in prefromance decreases beyond expectations. It did not prevent the game from functioning, and technically speaking was not "defective".

There are more than just critical error bugs and almost all optimizations are actually bug fixes (reducing/preventing hangs etc).

So no it doesn't. Feel free to say oxide disagrees or counterclaims, but saying unequivocally that an issue doesn't exist is beyond your (and the benchmarking people's) ability to determine from the small test sets.

Furthermore cherry-picking the response to leave out that they would suggest leaving it off anyways (which is a MAJOR concession for possible discrepancies) creates an impression quite different than the truth.

I have no objections to the format in which the benchmarkers opinions were presented provided you suggest that is their opinions/conclusions (which you basically do) not that their testing automatically becomes the truth.

[Yoinkerman]

I don't understand why Nvidia reacted that way when the benchmarks are pretty much showing that they have better DX11 drivers than AMD which they've been talking about ever since Mantle and the Star Swarm demo.

That shows even more as the DX12 results are exactly how they should be.

They have to try and cry foul to save PR face as this is one of the first dx12 games and it's big(ish) dx12 news. Nothing but PR.

[Curufinwe_wins]

They have to try and cry foul to save PR face as this is one of the first dx12 games and it's big(ish) dx12 news. Nothing but PR.

Funny enough if the dx12 game had the same fps but was dx11 I doubt anyone would care much.

Omg a game where the 390x beats the 980 at all but 1080p, a game where the fury x and 980ti are basically tied but the fury x wins more than it loses...

But because it's dx12 people assume that one game is indicative of all other dx12 games.

Which is may or may not be. But the assumption is still a bad one.

I don't agree with how nvidia said what they said, but the points made (taking away the pr saltiness) are important ones.

We have no idea what performance will look like when we this game goes live (I mean this is an alpha....) Or if this game is above or below par.

Hell for all we know oxide saying they have been working with nvidia for months could mean this is the best case scenario for nvidia's dx12 improvements and amd's massively increased raw power wins out. Or months of working together could mean nvidia thinks these devs are terrible programmers and should gtfo. Who knows?

[Notional]

I don't understand why Nvidia reacted that way when the benchmarks are pretty much showing that they have better DX11 drivers than AMD which they've been talking about ever since Mantle and the Star Swarm demo.

That shows even more as the DX12 results are exactly how they should be.

 

 

Because you are seeing AMD cards at much lower prices match performance of more expensive NVidia cards. NVidia's DX11 driver implementation is excellent with multithreaded support, which is the reason why NVidia is doing so well in the DX11 version of the game. But the future lies in the hands of DX12, not 11, so this does not bode well for NVidia. NVidia's higher priced hardware seems to not be able to physically outperform AMD, so with lower level API's like DX12, there is less optimization in the drivers (I would presume).

 

Many if not most bugs result in performance related issues. Like for example hairworks in witcher3 forcing 32x tesselation is/was a bug. It resulted in prefromance decreases beyond expectations. It did not prevent the game from functioning, and technically speaking was not "defective".

There are more than just critical error bugs and almost all optimizations are actually bug fixes (reducing/preventing hangs etc).

So no it doesn't. Feel free to say oxide disagrees or counterclaims, but saying unequivocally that an issue doesn't exist is beyond your (and the benchmarking people's) ability to determine from the small test sets.

Furthermore cherry-picking the response to leave out that they would suggest leaving it off anyways (which is a MAJOR concession for possible discrepancies) creates an impression quite different than the truth.

I have no objections to the format in which the benchmarkers opinions were presented provided you suggest that is their opinions/conclusions (which you basically do) not that their testing automatically becomes the truth.

 

A bug means it doesn't work like intended. That might result on poorer performance, but not necessarily.

HairWorks in Witcher 3 ran at a 64x tessellation factor, and that was not a bug, but set by NVidia as a standard factor. CDPR was not able to change that or optimize the code, as they did not have source code access.

I'd like to see a source for it being bugged, if you don't mind?

 

Well you say that, but the benchmarks do not show any major performance issue with 4x MSAA in any of the resolutions. The performance decrease seems to be fairly standard for MSAA in general, both in DX11 and DX12. I think this is more of a diplomatic approach by Oxide Games:

 

 

So what is going on then? Our analysis indicates that the any D3D12 problems are quite mundane. New API, new drivers. Some optimizations that that the drivers are doing in DX11 just aren’t working in DX12 yet. Oxide believes it has identified some of the issues with MSAA and is working to implement work arounds on our code. This in no way affects the validity of a DX12 to DX12 test, as the same exact work load gets sent to everyone’s GPUs. This type of optimizations is just the nature of brand new APIs with immature drivers.”

[That Norwegian Guy]

Not surprised, I actually called this development some time ago. Nvidia have been working around DX11 too much, getting an advantage through essentially throwing brute optimization code at specific games one at a time. Because AMD never had the budget for that.

 

Now that the API itself is efficient and lean, the wool is pulled off people's eyes and I say it's about time.

[spartaman64]

wow this makes sense amd has much more experience making drivers for low level apis like mantle 

[spartaman64]

this also makes you wonder what else was nvidia untruthful about in the past

[Briggsy]

Its very interesting just how matched the 390x and 980 are once you remove the cpu overhead. I'm reading a lot of comments (hopes/wishes) in this thread and in other forums about Nvidia DX12 drivers getting more efficient over time like they did with DX11, but the reality is that DX12 removes the need for extensive driver optimizations, which is why AMD are suddenly matching or beating Nvidia (in this example), but not a surprise given that AMD have been designing their hardware around low level API's (xbox 360, xbox one, PS4 for example) for many, many years now and have been desperately trying to shoehorn the same hardware into discrete GPU's.

 

Can't say I'm surprised at Nvidia's response, but I am slightly impressed that Ryan Shrout called them out on it without adding in any caveats to save face.

[Imakuni]

When the game leaves alpha and comes into Beta.. I'll at least remember the article. I won't really care untill it ACTUALLY comes out.

 

Still, so long as performance goes up (and not down), I'm cool with it.

[Commander Llama]

Mainly what I learned from this is Ashes of the Singularity looks feckin' sweet.

[Suika]

"The FX-8350 will be relevant when DX12 comes."

lol nope

[BuckGup]

Clearly you see something else then there bunch of results where dx11 scores better then dx12 which means dx12 ain't optimised yet at all, that nvidia has dx11 optimised better then ati thats quite obvious, still the ati results at dx12 is impressive

It is very good though

[Monarch]

AMD chose to focus on low-level APIs, while Nvidia decided it wanted to improve DX11. Funny thing is, Mantle made Nvidia reduce DX11 overhead. But what Nvidia did is more useful. When I buy a graphics card, I want it to play games NOW, not in a year or two. At the time I purchased GTX 970, we had DX11 games. And we still have DX11 games. There are only few games that support Mantle. It'll be a long time before we get DX12 games, so I'm glad I've had superior performance. What's the point of AMD GPUs performing better with DX12, when in current games their performance sucks.

 

Of course, I do expect Nvidia to fix their drivers once DX12 games start to come out. Obviously Nvidia's performance jump won't be as high as AMD's, since their DX11 driver overhead is really high, but Nvidia's cards at least shouldn't perform worse under DX12. 

[vXhanz]

It's interesting to look at, but it's only one game and it would be better to wait and see how things shake out in the next few months vs. this one test. I can see where Nvidia would be concerned, because at first glance it looks like AMD's cheaper hardware performs just as well as Nvidia's more expensive product. 

[Prysin]

I don't understand how the 980 can be so far behind the 390, but whatever, surprising results.

because in raw compute power, AMD has been ahead of Nvidia by a notable margin since the HD 7850 days. They apparently just didnt use their compute power very efficiently and brute forced performance.

[patrickjp93]

Look at the 1080p and 1600p low benchmarks (where cpu utilization is highest) there is a 5-15% difference in performance always in favor of the 6700k. That's HUGE in the world of cpu performance gaming impact.

Also @ Notional, you should put up the average and heavy % difference charts (they are very informative imho.)

That's called Amdahl's Law, one of the biggest reasons actual learned people like me have been saying multicore is only a stopgap solution for games. Learn it, understand it, cry in a corner, praise Intel for SIMD, and help me throw this Molotov Cocktail through the front door of Microsoft HQ.

Fewer, faster cores are king when it comes to central control of a machine. The more cores you have running, the more limited your scaling becomes, especially if you have synchronization overhead such as in the case of gaming.

[patrickjp93]

because in raw compute power, AMD has been ahead of Nvidia by a notable margin since the HD 7850 days. They apparently just didnt use their compute power very efficiently and brute forced performance.

Only theoretically/on paper. AMD's GPUs still can't keep up with Quadros or Teslas at all. The engineers behind the top 500 supercomputers still can't implement an OpenCL version of an algorithm and have it outperform the CUDA-based one in either single or double precision (Maxwell Quadro/Tesla excluded on DP obviously). K40 vs. S9170 has the K40 eating AMD's lunch. And when the best compute scientists can't make it work, AMD or the OpenCL runtime has a huge issue, possibly both.

This is an Nvidia driver optimization and game optimization problem.

[AlwaysFSX]

This is a blatant case that Nvidia has done their drivers so well that there was no performance gain going to an unoptimized DX12 driver. However, that's a fantastic thing, Nvidia can just throw money at the problem and continue to make drivers that work better.

 

This is nothing about Nvidia lying to consumers, just that they have more resources to fix problems.

[Curufinwe_wins]

That's called Amdahl's Law, one of the biggest reasons actual learned people like me have been saying multicore is only a stopgap solution for games. Learn it, understand it, cry in a corner, praise Intel for SIMD, and help me throw this Molotov Cocktail through the front door of HQ.

Fewer, faster cores are king when it comes to central control of a machine. The more cores you have running, the more limited your scaling becomes, especially if you have synchronization overhead such as in the case of gaming.

Of course, I cannot and do not dispute that. Anytime you spend switching/moving information from one thread to another (which by definition you must do more as you increase active threads) is clock time you are not using to compute. There is clearly optimal thread count for each and every code out there (except for Monte Carlo wherein the optimal thread count is basically, although finitely smaller than, the number of time's/particles its ran hahaha) and it is for most consumer codes today smaller than the total available resources.

 

I was mainly remarking that this game shows no scaling beyond 4 threads of any repute (and indeed the difference between the 6300 and 8370 suggests extremely low scaling beyond 6 threads.)

[bogus]

No wonder why NVIDIA insisted on trying to discredit Mantle with DX11 (funny thing it was with a similar benchmark that wiped the floor with Mantle BETA... and took NVIDIA more then 2 months of optimizations to try to achieve something)... like if that would make everything stop in time and we would have DX11 forever.

 

AMD and their partners sharing their work with Microsoft and Khronos was a great move. Thanks to it we have a leaner API that wont require game developers to be dependent on IHV to deliver the best performance for their products.

 

Mantle was a developers initiative. This is what the product makers want. They just need to deal with it.