Ashes of the Singularity Day 1 Benchmark Preview @ [H]

[razor1]

It actually is if we stay focused. The topic started as a DX11 vs 12. My question was if the benefits of performance are due to async. Brent's observations are that it is not all due to async. If this is the case than Nvidia is not doing DX12 as well as AMD >? Someone had mentioned that you can compare it by disabling the feature in the game. Kyle was kind enough to test with slower Cpus to examine the impact of CPU power in DX12. Hitman benchmarks are the best exemples of dx12 and async at play to date. AMD is marketing this async thing heavily, but they also prove that it works well if implemented correctly. How do you want to defend the fact that Nvidia is not so hot at async. No amount of tech jargon will make a difference to the reality. The reality and core question of this topic

The whole problem this arguement started is the use of wrong words, he is correct async compute is available on Fermi and up (not even just kelpar), but what AMD calls async compute is not what nV calls async compute. AMD marketing as used async shaders in their documentation, marketing people at AMD when talking about this feature has been using async compute. And this is where the confusion started.

 

[Ieldra]

It actually is if we stay focused. The topic started as a DX11 vs 12. My question was if the benefits of performance are due to async. Brent's observations are that it is not all due to async. If this is the case than Nvidia is not doing DX12 as well as AMD >? Someone had mentioned that you can compare it by disabling the feature in the game. Kyle was kind enough to test with slower Cpus to examine the impact of CPU power in DX12. Hitman benchmarks are the best exemples of dx12 and async at play to date. AMD is marketing this async thing heavily, but they also prove that it works well if implemented correctly. How do you want to defend the fact that Nvidia is not so hot at async. No amount of tech jargon will make a difference to the reality. The reality and core question of this topic

I'm saying I have no idea the extent to which multi-engine will be beneficial to nvidia hardware. AMD have exaggerated the benefits of async immensely; it's 10% at best. And hard to tune.

What I'm trying to disprove is that async is somehow better than no async.

If my hardware performs better than yours, without needing the feature we erroneously call 'async', then my hardware performs better than yours.

I defend the fact that nvidia is not so hot at async by saying that the very same design decisions that lead to async not being 'as hot' on nvidia hardware allowed their hardware to have low execution times and high utilization throughout the dx11 generation.

The beyond3d thread with the benchmark proves asynchronous execution works, just that it is not concurrent execution

 

[JustReason]

Ok lets do this, start up another thread about async, out of all the other million out there, and lets do it. I want to see your reasoning, put your mouth where your call sign is. Instead of spouting out crap that doesnt' make any sense.

This is a specific topic about async compute, async shaders (multi engine) concurrency. I want to see that. I want to see your definitions, your explanations based on benchmarks, based on sample data from B3D's concurrent execution demo, how those graphics are correlated between the two, then we can go from there.

And what did I say that doesn't make sense? I asked an irreverent poster to prove it, NVIDIA being async capable hardware or software wise (being lenient here). I mean look at the results of this game AOTS. Maybe Nvidia will have some fix, maybe they wont. But at this juncture they don't and on the surface it is quite devastating. If it continues, and looking at Hitman it very well might, then Nvidia is in a bad spot.

 

[razor1]

Because if you want to talk about this, you have to know what you are talking about, he already showed he knows much more than you on the topic. I can have a conversation with him, Anarchist can too, Mahigan can too, but you I don't think you can, you don't have the understanding of what he is talking about because you don't care to understand it. How many times have we discussed it to no avail? And this is all because one side just doesn't want to learn anything about basic graphics and architecture designs. You couldn't even use the right terms when we first talked about it, and you still aren't. So why do you lambaste someone that people that can understand the topic with fair amount of depth, can see where he is coming from, but you can not?

I stated this a while back, to really talk about async the topic is extremely vast because the effects of async code affects many parts of the GPU.

 

[JustReason]

The whole problem this arguement started is the use of wrong words, he is correct async compute is available on Fermi and up (not even just kelpar), but what AMD calls async compute is not what nV calls async compute. AMD marketing as used async shaders in their documentation, marketing people at AMD when talking about this feature has been using async compute. And this is where the confusion started.

The wrong use of the words is what I referenced when I stated CUDA. In the terminology of DX12 as the discussion started the use of the word async was inline with DX12 and this game. It's the detractors that wish to do damage control on NVIDIA's behalf that are attempting to add the other definition to the discussion.

 

[razor1]

what is exposed in CUDA, is exposed in DirectCompute and CL, async compute, additionally CUDA can do concurrent execution, but that is not exposed in DirectCompute but is in OCL, which we don't know why that is.

Why isn't exposed in Directcompute, this is what nV has been saying they are working on. I have a feeling it might be a hardware design issue they can't work it out, but it is working in other API's so maybe they can.

 

[JustReason]

Because if you want to talk about this, you have to know what you are talking about, he already showed he knows much more than you on the topic. I can have a conversation with him, Anarchist can too, Mahigan can too, but you I don't think you can, you don't have the understanding of what he is talking about because you don't care to understand it. How many times have we discussed it to no avail? And this is all because one side just doesn't want to learn anything about basic graphics and architecture designs. You couldn't even use the right terms when we first talked about it, and you still aren't. So why do you lambaste someone that people that can understand the topic with fair amount of depth, can see where he is coming from, but you can not?

I stated this a while back, to really talk about async the topic is extremely vast because the effects of async code affects many parts of the GPU.

Prove it. Pretending you are smarter doesn't make it so. I haven't used the wrong terms. I am smart enough to recognize the async you and other NVIDIA posters are using isn't what was being discussed.

 

[razor1]

I have done it numerous times, and if you want me to do it again, as I stated I want to see something from you

Ashes of the Singularity Day 1 Benchmark Preview @ [H]

Because if you aren't willing to do the above, it is a waste of time for me to do so.

You were smart enough to after it was pointed out, that is not being smart.....

 

[Jeff Goldblum]

The gains for AMD is combination of things and not solely down to async compute.
The reduced cpu overhead will help alot and AMD never had a multithreaded DX11 driver unlike Nvidia who put alot of time and effort into it.

Async has been disabled in Nvidia drivers since the first Ashes alpha benches last year as the performance tanked with it enabled, the gpu grinds to halt if too much is queued up on the compute side and they have had 8 months so far and still have not come up with a fix or reenabled it, this is very telling.

AMD are now working on Async shaders evolved which gives a QOS to the compute queue

 

[Ieldra]

Okay, some background here; asynchronous execution is a computing concept, I don't care that DX12 uses the term in some other context. It's misleading. If something is being executed asynchronously with respect to something else, it is being dispatched/executed before the preceding function can return. If something is being executed concurrently with something else, then the execution of both happen in an overlapping time-frame. If two things are being executed in parallel, they are both being executed at the same instant.

In concurrent computation two computations both advance independently of each other. The second computation doesn't have to wait until the first is finished for it to advance. It doesn't state however, the mechanism how this is achieved. In single-core setup, suspending and alternating between threads is required (also called pre-emptive multithreading).

In parallel computation two computations both advance simultaneously - that is literally at the same time. This is not possible with single CPU and requires multi-core setup instead.

Now we've gotten that out of the way, it seems to me like AMD have improved their game immensely since the release of the Omega drivers beta in January 2015. Their DX11 driver stack was rebuilt, it's more efficient, and requires far less cpu overhead, and that's great, and that's fine.
The problem, as I see it, is that you're now conflating AMD's improved performance with Nvidia suffering from performance degradation, which is not the case.

Additionally, it is genuinely laughable to see you using Hitman and AotS to prove some kind of point regarding Async.

We have AotS where AMD performs terribly in DX11, and boasts huge gains from DX12 and 'async', and Hitman where AMD outperforms nvidia hardware even in DX11, where a 390 matches a 980Ti and a 390x competes with a Fury X. In addition to this mess, AMD hadware doesn't gain much in the transition to d3d12; the Fury X even performs worse under dx12, in Hitman.

AMD claims Hitman has the best implementation of 'async' yet, sweet, where's the huge bonus ? Why does the Fury X perform worse under DX12? Why is the huge boost seen in AotS nowhere to be found?
Why did IO Interactive claim async is stupidly hard to tune and more than likely not worth the effort, the specifically mentioned the difficulty in tuning it for a range of hardware, even produced by the same IHV.

The gains for AMD is combination of things and not solely down to async compute.
The reduced cpu overhead will help alot and AMD never had a multithreaded DX11 driver unlike Nvidia who put alot of time and effort into it.

Async has been disabled in Nvidia drivers since the first Ashes alpha benches last year as the performance tanked with it enabled, the gpu grinds to halt if too much is queued up on the compute side and they have had 8 months so far and still have not come up with a fix or reenabled it, this is very telling.

AMD are now working on Async shaders evolved which gives a QOS to the compute queue

Edit: Sorry! Confused you with the other guy!

You downplay the influence of async in the first line of your post, then claim Nvidia are screwed because of it further down.

I don't know how else to explain this to you, YOU SHOULD NOT, UNDER ANY CIRCUMSTANCES, ASSIGN LONG RUNNING SHADERS TO THE COMPUTE QUEUE ON NVIDIA HARDWARE. YOU CANNOT CODE FOR AMD AND NVIDIA IN THE SAME WAY IN THIS RESPECT. IT WILL NOT WORK WELL ON BOTH IHVS.

Everything you write points to the fact that you still think 'async' performance gains come out of thin air, there's a reason there even IS extra performance to be had, and that is suboptimal hardware utilization

I literally made up for the intrinsic advantage the Fury X has in AotS by exploiting a well known, yet not very well publicized as of late, NVIDIA feature called "overclocking headroom".

Let me put it this way. Let's assume AMD manages to score 10% performance bonus in every single DX12 game due to 'async'.

If Nvidia can guarantee 10% overclock then that advantage is rendered null. It's a question of utilization, and no matter how you spin it there's no skirting around the fact that NVIDIA will never have as much to gain from such a feature because their pipeline is well occupied as is

 

[Jeff Goldblum]

The gains for AMD come from using Async because they have the hardware ACEs and Nvidia do not.
Best estimates so far are 5-10% gain using async compute but AMD is gaining more so its a combination of factors.

At the end of the day Nvidias architecture in Maxwell is optimised for the serial nature of DX11, they also ditched compute in favour of rendering performance, even overclocking will not help much in DX12.
AMD have wanted more parallism since the start of GCN. They played a very long game but now with various partnerships etc its beginning to pay off.
I also cant see devs not using async in games just because it effects Nvidia, The whole console market is AMD based and will get gains for using this, PS4 has already been using this for some time.

 

[JustReason]

I have done it numerous times, and if you want me to do it again, as I stated I want to see something from you

Ashes of the Singularity Day 1 Benchmark Preview @ [H]

Because if you aren't willing to do the above, it is a waste of time for me to do so.

You were smart enough to after it was pointed out, that is not being smart.....

Is it too hard for you to prove your accusations? Also I already stated I am in route back from vacation so no time for a few days to delve into showing you up again. I posted verbatim proof with a plethora of links before and your response was null, except to claim again that I don't know what I am talking about.

How about you add something to this debate. AMD has a clear win here. Async DX12 implementation is showing real promise. And what most NVIDIA owners are worried about is that they may not in the foreseeable future be able to take advantage of async. Granted they still perform admirably enough in either API, but still a concern for future use when further uses may be entertained.

Now consider that same market share that gets touted daily. AMD has a few months to close the gap or take over, especially if Pascal has the same performance issues.

 

[Imhotep]

The gains for AMD is combination of things and not solely down to async compute.
The reduced cpu overhead will help alot and AMD never had a multithreaded DX11 driver unlike Nvidia who put alot of time and effort into it.

Async has been disabled in Nvidia drivers since the first Ashes alpha benches last year as the performance tanked with it enabled, the gpu grinds to halt if too much is queued up on the compute side and they have had 8 months so far and still have not come up with a fix or reenabled it, this is very telling.

AMD are now working on Async shaders evolved which gives a QOS to the compute queue

Okay, some background here; asynchronous execution is a computing concept, I don't care that DX12 uses the term in some other context. It's misleading. If something is being executed asynchronously with respect to something else, it is being dispatched/executed before the preceding function can return. If something is being executed concurrently with something else, then the execution of both happen in an overlapping time-frame. If two things are being executed in parallel, they are both being executed at the same instant.



Now we've gotten that out of the way, it seems to me like AMD have improved their game immensely since the release of the Omega drivers beta in January 2015. Their DX11 driver stack was rebuilt, it's more efficient, and requires far less cpu overhead, and that's great, and that's fine.
The problem, as I see it, is that you're now conflating AMD's improved performance with Nvidia suffering from performance degradation, which is not the case.

Additionally, it is genuinely laughable to see you using Hitman and AotS to prove some kind of point regarding Async.

We have AotS where AMD performs terribly in DX11, and boasts huge gains from DX12 and 'async', and Hitman where AMD outperforms nvidia hardware even in DX11, where a 390 matches a 980Ti and a 390x competes with a Fury X. In addition to this mess, AMD hadware doesn't gain much in the transition to d3d12; the Fury X even performs worse under dx12, in Hitman.

AMD claims Hitman has the best implementation of 'async' yet, sweet, where's the huge bonus ? Why does the Fury X perform worse under DX12? Why is the huge boost seen in AotS nowhere to be found?
Why did IO Interactive claim async is stupidly hard to tune and more than likely not worth the effort, the specifically mentioned the difficulty in tuning it for a range of hardware, even produced by the same IHV.


Edit: Sorry! Confused you with the other guy!

You downplay the influence of async in the first line of your post, then claim Nvidia are screwed because of it further down.

I don't know how else to explain this to you, YOU SHOULD NOT, UNDER ANY CIRCUMSTANCES, ASSIGN LONG RUNNING SHADERS TO THE COMPUTE QUEUE ON NVIDIA HARDWARE. YOU CANNOT CODE FOR AMD AND NVIDIA IN THE SAME WAY IN THIS RESPECT. IT WILL NOT WORK WELL ON BOTH IHVS.

Everything you write points to the fact that you still think 'async' performance gains come out of thin air, there's a reason there even IS extra performance to be had, and that is suboptimal hardware utilization

I literally made up for the intrinsic advantage the Fury X has in AotS by exploiting a well known, yet not very well publicized as of late, NVIDIA feature called "overclocking headroom".

Let me put it this way. Let's assume AMD manages to score 10% performance bonus in every single DX12 game due to 'async'.

If Nvidia can guarantee 10% overclock then that advantage is rendered null. It's a question of utilization, and no matter how you spin it there's no skirting around the fact that NVIDIA will never have as much to gain from such a feature because their pipeline is well occupied as is

This is where you loose any credibility. Overclocking

 

[Ieldra]

The division is dx11. I saw no mention of conservative rasterization in the division or dx11. And the performance hit is huge

it never worked that way to begin with. under dx11 you OBVIOUSLY couldnt just write code and it works well considering how many crappy games came out. You think dx12 is going to be this hugely different situation and I don't think so. The engine developers might have a bit of work to do, but thats what they do. differences in hardware are not new just because of dx12. I saw the GDC slides and thought people were mistaking things. Just because they say this and that should be done differently to run better on this or that hardware does not mean anything has changed that much. There would have been situations like that in dx11.

Conservative Rasterization is part of the DX11.3 feature set, and it's required for effects like HFTS. You could do it on AMD hardware, no conservative raster, but it would be impractically slow.

Under DX11 there was much more abstraction from the hardware, a lot of the optimization work was at the driver level... Come one, you all know this, I don't need to state the obvious.

And what do you mean overclocking doesn't make a difference in DX12 ??!?!? When I hear these things it's like someone driving a stake through my heart! Overclocking will always provide performance gains. in AotS (dx12 in case you didnt know) overclocking gives me 25% additional performance; from 60~ to 75fps

This is where you loose any credibility. Overclocking

This? This where? This what?

 

[razor1]

Is it too hard for you to prove your accusations? Also I already stated I am in route back from vacation so no time for a few days to delve into showing you up again. I posted verbatim proof with a plethora of links before and your response was null, except to claim again that I don't know what I am talking about.

How about you add something to this debate. AMD has a clear win here. Async DX12 implementation is showing real promise. And what most NVIDIA owners are worried about is that they may not in the foreseeable future be able to take advantage of async. Granted they still perform admirably enough in either API, but still a concern for future use when further uses may be entertained.

Now consider that same market share that gets touted daily. AMD has a few months to close the gap or take over, especially if Pascal has the same performance issues.

Start a new thread about it, and I will post all the links in this forum to your "knowledge" or lack there of about async? Fair enough? Yes I can go through this forum and find all the crap posts that you did........

AMD has a clear win with a game that even in DX11 has a sizable performance advantage over its counterparts, which we have never seen in any DX11 applications.... they should not be where they are at in DX11.

What does that tell us?

It tells me this game is optimized for AMD hardware well before DX12 or async is in the mix.

 

[Jeff Goldblum]

Let me put it this way. Let's assume AMD manages to score 10% performance bonus in every single DX12 game due to 'async'.
If Nvidia can guarantee 10% overclock then that advantage is rendered null. It's a question of utilization, and no matter how you spin it there's no skirting around the fact that NVIDIA will never have as much to gain from such a feature because their pipeline is well occupied as is

Overclocking in DX12 has already been mentioned by Stardock :

`The one negative Wardell sees is overclockers, who “will not have a good time with DX12.” But that’s just a small niche of the overall PC market.`

The issue with overclocking in DX12 is the api is about width, not how fast you can ram a queue through a single processing pipeline like in DX11

 

[Imhotep]

Why would anyone want to buy Card Y and overclock it in order to keep up with card X which costs less

 

[Ieldra]

Overclocking in DX12 has already been mentioned by Stardock :

`The one negative Wardell sees is overclockers, who “will not have a good time with DX12.” But that’s just a small niche of the overall PC market.`

The issue with overclocking in DX12 is the api is about width, not how fast you can ram a queue through a single processing pipeline like in DX11

So I've been hallucinating this whole time ? My 25% overclock didn't actually translate into a clean 25% increase in performance under DX12 in a drawcall heavy game?

Why would anyone want to buy Card Y and overclock it in order to keep up with card X which costs less

A fury X costs more than a 980Ti here, and an overclocked 980Ti outperforms it by a huge margin in most games.

 

[Jeff Goldblum]

Start a new thread about it, and I will post all the links in this forum to your "knowledge" or lack there of about async? Fair enough? Yes I can go through this forum and find all the crap posts that you did........

AMD has a clear win with a game that even in DX11 has a sizable performance advantage over its counterparts, which we have never seen in any DX11 applications.... they should not be where they are at in DX11.

What does that tell us?

It tells me this game is optimized for AMD hardware well before DX12 or async is in the mix.

This has already been covered by Oxide :

Is Ashes of the Singularity biased?

Ashes of the Singularity is the first DX12 game on the market, and the performance delta between AMD and Nvidia is going to court controversy from fans of both companies. We won’t know if its performance results are typical until we see more games in market. But is the game intrinsically biased to favor AMD? I think not — for multiple interlocking reasons.

First, there’s the fact that Oxide shares its engine source code with both AMD and Nvidia and has invited both companies to both see and suggest changes for most of the time Ashes has been in development. The company’s Reviewer’s Guide includes the following:

[W]e have created a special branch where not only can vendors see our source code, but they can even submit proposed changes. That is, if they want to suggest a change our branch gives them permission to do so…

This branch is synchronized directly from our main branch so it’s usually less than a week from our very latest internal main software development branch. IHVs are free to make their own builds, or test the intermediate drops that we give our QA.

Oxide also addresses the question of whether or not it optimizes for specific engines or graphics architectures directly.

Oxide primarily optimizes at an algorithmic level, not for any specific hardware. We also take care to avoid the proverbial known “glass jaws” which every hardware has. However, we do not write our code or tune for any specific GPU in mind. We find this is simply too time consuming, and we must run on a wide variety of GPUs. We believe our code is very typical of a reasonably optimized PC game.

 

[Jeff Goldblum]

So I've been hallucinating this whole time ? My 25% overclock didn't actually translate into a clean 25% increase in performance under DX12 in a drawcall heavy game?



.

What heavy DX12 drawcall game was it ? Hitman or Ashes as they are the only two groundup DX12 games we have

 

[razor1]

This has already been covered by Oxide :

Is Ashes of the Singularity biased?

Ashes of the Singularity is the first DX12 game on the market, and the performance delta between AMD and Nvidia is going to court controversy from fans of both companies. We won’t know if its performance results are typical until we see more games in market. But is the game intrinsically biased to favor AMD? I think not — for multiple interlocking reasons.

First, there’s the fact that Oxide shares its engine source code with both AMD and Nvidia and has invited both companies to both see and suggest changes for most of the time Ashes has been in development. The company’s Reviewer’s Guide includes the following:

[W]e have created a special branch where not only can vendors see our source code, but they can even submit proposed changes. That is, if they want to suggest a change our branch gives them permission to do so…

This branch is synchronized directly from our main branch so it’s usually less than a week from our very latest internal main software development branch. IHVs are free to make their own builds, or test the intermediate drops that we give our QA.

Oxide also addresses the question of whether or not it optimizes for specific engines or graphics architectures directly.

Oxide primarily optimizes at an algorithmic level, not for any specific hardware. We also take care to avoid the proverbial known “glass jaws” which every hardware has. However, we do not write our code or tune for any specific GPU in mind. We find this is simply too time consuming, and we must run on a wide variety of GPUs. We believe our code is very typical of a reasonably optimized PC game.

So what ever code they are doing, they are doing for all hardware if that was true, then what ever optimizations they are doing in DX11 should have came over to nV's hardware too, so why is this game so apparrently different from all the other DX11 games out there.

Just look at the 980ti vs Fury X performance in Dx11

Where do we ever see nV loose out to the fury x lower than 4k in DX 11?

And even when it did loose it was marginal loss at 4k in DX11

IF the developers statements where true then I would expect them to mimic what all other games in DX11 have shown up till now.

 

[Ieldra]

This has already been covered by Oxide :

Is Ashes of the Singularity biased?

Ashes of the Singularity is the first DX12 game on the market, and the performance delta between AMD and Nvidia is going to court controversy from fans of both companies. We won’t know if its performance results are typical until we see more games in market. But is the game intrinsically biased to favor AMD? I think not — for multiple interlocking reasons.

First, there’s the fact that Oxide shares its engine source code with both AMD and Nvidia and has invited both companies to both see and suggest changes for most of the time Ashes has been in development. The company’s Reviewer’s Guide includes the following:

[W]e have created a special branch where not only can vendors see our source code, but they can even submit proposed changes. That is, if they want to suggest a change our branch gives them permission to do so…

This branch is synchronized directly from our main branch so it’s usually less than a week from our very latest internal main software development branch. IHVs are free to make their own builds, or test the intermediate drops that we give our QA.

Oxide also addresses the question of whether or not it optimizes for specific engines or graphics architectures directly.

Oxide primarily optimizes at an algorithmic level, not for any specific hardware. We also take care to avoid the proverbial known “glass jaws” which every hardware has. However, we do not write our code or tune for any specific GPU in mind. We find this is simply too time consuming, and we must run on a wide variety of GPUs. We believe our code is very typical of a reasonably optimized PC game.

I've already referenced these statements by Oxide. In case it's still not clear, I've made the most important line in this quote BOLD and ITALIC.

What heavy DX12 drawcall game was it ? Hitman or Ashes as they are the only two groundup DX12 games we have

AotS...

The results from Hitman do not agree with the results from AotS, not by a long shot.


I understand you quoting Oxide regarding not having vendor specific code thinking it supports your view, but it doesn't. This is exactly what they are expected to do under DX12. An async implementation that works well for Hawaii may need tuning to work on Fiji, let alone to work on another IHV.

All they need is an engine built from the ground up to take advantage of AMD hardware (check) and to be unwilling to optimize specifically for different hardware (check), claiming that they made their code available to IHVs is laughable; they want to bake a delicious DX12 cake without getting their hands dirty.


Look at the difference in the latency here, tall bars means high execution time.

The critical thing is that AMD hardware, when running in 'async mode' executes the work in LESS THAN the sum of the graphics+compute times. This is because there is some overlap in the execution; hence concurrency

AMD vs NVIDIA practices in DX12

DirectX 12 Requires Different Optimization on Nvidia and AMD Cards, Lots of Details Shared | DualShockers

 

[Pieter3dnow]

When you do the full review please get a AMD FX chip for review. People are dying to see if DX12 is really an AMD cpu's saving grace or if it will still get trounced.

Take this next post link and scroll down a little

The thing about strong arming is . Microsoft

Any takers >? This sums it up nicely Hitman Benchmarks mit DirectX 12 (Seite 2)
Reddit is raving about this thing.

DirectX 12 auf einem AMD FX-8370
Alle bisherigen Tests wurden mit dem Core i7-6700K und damit einem sehr schnellen Prozessor durchgeführt. Die folgenden Testreihen setzen mit dem AMD FX-8370 hingegen eine deutlich langsamere CPU ein. Und es zeigt sich: DirectX 12 hat in Hitman durchaus einen großen Effekt.

All tests were run on Core i7-6700K and that is a very fast processor. The tests we did with the AMD FX 8370 which is considerably slower which led us to the conclusion that DX12 Hitman has a major impact on performance ....

The numbers are quite good you don't need a translation for that

 

[Jeff Goldblum]

I've already referenced these statements by Oxide. In case it's still not clear, I've made the most important line in this quote BOLD and ITALIC.


All they need is an engine built from the ground up to take advantage of AMD hardware (check) and to be unwilling to optimize specifically for different hardware (check), claiming that they made their code available to IHVs is laughable; they want to bake a delicious DX12 cake without getting their hands dirty.

You talk like Async is some kind of AMD secret weapon when its part of the API and yet its only AMDs implimenation of async that gives them the edge.

Alot of the big engines are DX12 already so wont need specific engines written from the ground up. DICE advised on the Mantle spec so Frostbyte 3 was one of the first to be DX12 ready.

 

[Ieldra]

You talk like Async is some kind of AMD secret weapon when its part of the API and yet its only AMDs implimenation of async that gives them the edge.

Alot of the big engines are DX12 already so wont need specific engines written from the ground up. DICE advised on the Mantle spec so Frostbyte 3 was one of the first to be DX12 ready.

huh? I'm literally saying it's the implementation that favors AMD, not that 'async' is anyone's secret weapon. It's a damn tool to eke more performance out of hardware, it won't work equally well on everything, you need to tune the damn thing .

That's all I'm saying. That, and that it's overhyped. There is NO REASON to use 'async' if there's no performance gain from it.

Asynchronous shader technology allows more flexibility in terms of the timing and order of execution for independent tasks. When used effectively, the result is better utilization of the GPU, faster frame times, and improved responsiveness.

This has gotten silly, I'm literally quoting official AMD documentation here

Gaming: Asynchronous Shaders Evolved | Community

 

[PanzeR-]

I'm not surprised by AMD's lead in DX12, but the DX11 results are weird. Wouldnt expect the 390x or fury x to come ahead of their nvidia counterpart in this...It's like whatever gives the lead to AMD in DX12 adds up to something already present in DX11

 

[Jeff Goldblum]

But Async allows multiengines in DX12 and its this REASON why it will be used like Oxide devs state :

`The Nitrous Engine that powers Ashes of the Singularity makes extensive use of asynchronous compute and uses it for up to 30% of a given frame’s workload. Oxide has stated that they believe this will be a common approach in future games and game engines, since DirectX 12 encourages the use of multiple engines to execute commands from separate queues in parallel.`

I mentioned the Async evolved earlier ^^^ its basically a QOS for the compute queue

 

[Pieter3dnow]

Let me put it this way. Let's assume AMD manages to score 10% performance bonus in every single DX12 game due to 'async'.

If Nvidia can guarantee 10% overclock then that advantage is rendered null. It's a question of utilization, and no matter how you spin it there's no skirting around the fact that NVIDIA will never have as much to gain from such a feature because their pipeline is well occupied as is

You just made this sound like it was catering to AMD while it is a DX12 feature that Nvidia wants to claim what you are writing here is just something as one of the previous poster said disabled in driver. Now why do we need to believe this ?

You counter the argument by saying were just have to peddle faster to make up for "Async" . Good god man get with the program "were" not asking Nvidia to walk on water ..

 

[Ieldra]

You just made this sound like it was catering to AMD while it is a DX12 feature that Nvidia wants to claim what you are writing here is just something as one of the previous poster said disabled in driver. Now why do we need to believe this ?

You counter the argument by saying were just have to peddle faster to make up for "Async" . Good god man get with the program "were" not asking Nvidia to walk on water ..

Found it hard to understand your post quite frankly, but here goes; I am not saying 'async' caters to AMD. I'm not saying DX12 is AMD biased, I'm not saying any of that.

I'm saying, if your code, specifically the part involving asynchronous concurrent execution of commands from multiple queues, runs well on AMD hardware, it will more than likely run badly on NVIDIA hardware.

I'm saying 'async' should be used much more carefully on NVIDIA hardware, and for different use cases. This isn't something I'm pulling out of my ass, this a natural consequence of the low-level nature of d3d12.

The counter argument you are referring to is my mention of overclocking I'm assuming. Okay. 'async shaders' increase gpu utilization on AMD hardware because it hides latency, latency that is inherently more pronounced in GCN than in Kepler/Maxwell BY DESIGN.

Now let's assume that AMD hw (with async in software) is 100% utilized, and NVIDIA hw (with no async software) is 100% utilized. Let's assume they are performing equally. Is one better than the other ? No.

Now let's assume the hardware in question is a Fury X and a 980Ti. A Fury X can overclock by about 10% on average, a 980Ti can do 20%. The 980Ti will be faster. That's all I'm saying. Async isn't magic, you're treating it as such.

You can overcome a lack of 'async', you cannot overcome the lack of conservative rasterization.

http://32ipi028l5q82yhj72224m8j.wpe...ogramming_Model_and_Hardware_Capabilities.pdf

 

[Yakk]

If you really want to have some fun with Ashes, add :

-modding

to the command line!

 

[DukenukemX]

I literally made up for the intrinsic advantage the Fury X has in AotS by exploiting a well known, yet not very well publicized as of late, NVIDIA feature called "overclocking headroom".

Let me put it this way. Let's assume AMD manages to score 10% performance bonus in every single DX12 game due to 'async'.

If Nvidia can guarantee 10% overclock then that advantage is rendered null. It's a question of utilization, and no matter how you spin it there's no skirting around the fact that NVIDIA will never have as much to gain from such a feature because their pipeline is well occupied as is

#1 Overclocking is not something you can rely on. Besides binning, there's also cooling and a number of factors that determine how much you can overclock if at all. Given that you're able to overclock.

#2 You can also overclock AMD cards. I know, shocking but it's a thing.

#3 10% is the difference between a 970 and a 980, so it's a pretty big deal.

 

[Ieldra]

#1 Overclocking is not something you can rely on. Besides binning, there's also cooling and a number of factors that determine how much you can overclock if at all. Given that you're able to overclock.

#2 You can also overclock AMD cards. I know, shocking but it's a thing.

#3 10% is the difference between a 970 and a 980, so it's a pretty big deal.

#1. Never said overclocking is something you can rely on. 'Async' isn't something you can rely on either.
#2. Never said anything to the contrary
#3. Never said anything to the contrary

If we look at current hardware though, there is no doubt there is a much larger overclocking headroom on Maxwell vs Hawaii and GCN.

I genuinely don't understand what your point is.

 

[DukenukemX]

What does that tell us?

It tells me this game is optimized for AMD hardware well before DX12 or async is in the mix.

Now you know how AMD GPU owners feel when developers use GameWorks. Except that this game was openly developed with both AMD and Nvidia. Where GameWorks titles tend to cut out AMD from the development process.

Why would anyone want to buy Card Y and overclock it in order to keep up with card X which costs less

Because people love a brand no matter how much it screws them over. People are just irrational. The GTX 970 has 512MB of defective ram and it's the #1 selling graphics card. Does that make sense to anyone?

 

[Ieldra]

Now you know how AMD GPU owners feel when developers use GameWorks. Except that this game was openly developed with both AMD and Nvidia. Where GameWorks titles tend to cut out AMD from the development process.



Because people love a brand no matter how much it screws them over. People are just irrational. The GTX 970 has 512MB of defective ram and it's the #1 selling graphics card. Does that make sense to anyone?

Gameworks is in the Division and it runs fine. Far Cry uses gameworks and it always ran better on AMD.

Crock of shit.

AMD made a huge deal out of Witcher 3 claiming CDPR and NVIDIA conspired against them, claimed they REQUIRED source code access to fix performance on their end, wailed on and on and on about it for days. Then released driver update and fixed the issue by reducing tessellation factor.

We should all be abhorred, NVIDIA pushing for use of features their hardware excels at. HOW DARE THEY, IT'S LIKE THEY'RE IN THE BUSINESS FOR THE MONEY.

 

[Imhotep]

Now you know how AMD GPU owners feel when developers use GameWorks. Except that this game was openly developed with both AMD and Nvidia. Where GameWorks titles tend to cut out AMD from the development process.



Because people love a brand no matter how much it screws them over. People are just irrational. The GTX 970 has 512MB of defective ram and it's the #1 selling graphics card. Does that make sense to anyone?

Now it does. Let me not mention who and why

 

[razor1]

Now you know how AMD GPU owners feel when developers use GameWorks. Except that this game was openly developed with both AMD and Nvidia. Where GameWorks titles tend to cut out AMD from the development process.



Because people love a brand no matter how much it screws them over. People are just irrational. The GTX 970 has 512MB of defective ram and it's the #1 selling graphics card. Does that make sense to anyone?

Most gameworks features run just fine, its only the ones with tesselation that scew with AMD hardware, and those features can be turned off, so if anyone that has AMD hardware that doesn't like gameworks, turn them off. I don't even think those features are turned on by default anyways.

DX12 and aysnc you need to have different paths for both IHV's, the developer stating that they only did one path, pretty much sums it up, They optimized for AMD and didn't do jack for nV. So Does this game show us anything for future DX12 products. Hard to say but by his statement along, I wouldn't think it will at this point.

 

[DukenukemX]

#1. Never said overclocking is something you can rely on. 'Async' isn't something you can rely on either.

Async Compute is a feature you can rely on, just not on Nvidia hardware. In fact on Nvidia hardware it seems reliable in that it always slows down the graphics hardware.

You can't compare a feature that's actually part of the silicon to overclocking. There's a reason why manufacturers clock a chip to certain speeds, because that's the speed they can guarantee it to run. There's no guarantee beyond that clock. That's overclocking 101.

Anyone who's ever had a Celeron 300a or Athlon XP-M Mobile 2500+ knows what I'm talking about.

If we look at current hardware though, there is no doubt there is a much larger overclocking headroom on Maxwell vs Hawaii and GCN.

Yes, Maxwell hardware does tend to overclock better than AMD, but nobody should buy hardware and expect it to overclock as well as others have. That's why overclocking is not something one should consider but Async Compute is cause it's a feature.

I genuinely don't understand what your point is.

Honestly, nothing will. Not if you're determined to not understand.

 

[Imhotep]

Gameworks is in the Division and it runs fine. Far Cry uses gameworks and it always ran better on AMD.

Crock of shit.

AMD made a huge deal out of Witcher 3 claiming CDPR and NVIDIA conspired against them, claimed they REQUIRED source code access to fix performance on their end, wailed on and on and on about it for days. Then released driver update and fixed the issue by reducing tessellation factor.

We should all be abhorred, NVIDIA pushing for use of features their hardware excels at. HOW DARE THEY, IT'S LIKE THEY'RE IN THE BUSINESS FOR THE MONEY.

Maybe the way Nvidia Does it ticks developers off. What good is it to have a great looking girlfriend and not being able to share anything with her

 

[DukenukemX]

Gameworks is in the Division and it runs fine. Far Cry uses gameworks and it always ran better on AMD.

Crock of shit.

AMD made a huge deal out of Witcher 3 claiming CDPR and NVIDIA conspired against them, claimed they REQUIRED source code access to fix performance on their end, wailed on and on and on about it for days. Then released driver update and fixed the issue by reducing tessellation factor.

We should all be abhorred, NVIDIA pushing for use of features their hardware excels at. HOW DARE THEY, IT'S LIKE THEY'RE IN THE BUSINESS FOR THE MONEY.

Most gameworks features run just fine, its only the ones with tesselation that scew with AMD hardware, and those features can be turned off, so if anyone that has AMD hardware that doesn't like gameworks, turn them off. I don't even think those features are turned on by default anyways.

DX12 and aysnc you need to have different paths for both IHV's, the developer stating that they only did one path, pretty much sums it up, They optimized for AMD and didn't do jack for nV. So Does this game show us anything for future DX12 products. Hard to say but by his statement along, I wouldn't think it will at this point.

You know what, I don't feel like explaining it. I'll let AdoredTV do it for me.

 

[Yakk]

Now let's assume that AMD hw (with async in software) is 100% utilized, and NVIDIA hw (with no async software) is 100% utilized. Let's assume they are performing equally. Is one better than the other ? No.

That would be a fallacy. What 100% would mean can lead to interpretation.

100% use of just 1 GPU resource does not equate to using 100% of the complete GPU. Async allows the parallel use of more GPU resources with minimal developer time and money spent. Without it, good luck for a game engine.

DX12 and aysnc you need to have different paths for both IHV's, the developer stating that they only did one path, pretty much sums it up, They optimized for AMD and didn't do jack for nV.

Devs worked with, and shared their source code with nvidia. That got them an attempted virtual public flogging from nvidia PR. I think that is more telling.