Crimson 16.9.1 reduces CPU overhead by quite a margin, several titles see boots in performance.

[Carfax83]

I'm SHOCKED that BFG retorted to the old statement that AMD did not impact DX12 development. Why do you think Mantel or DX12 started in the first place? They could have easily have used 11.3 specs and be done. The issue comes down to consoles and AMD pushing console low level work to PC's.

As BFG10K already noted, low level APIs are nothing new to the PC. The reason for low level APIs making a comeback on PC isn't because AMD have locked down the consoles. It's because there are far fewer IHVs today than there were over a decade ago, and also because the IHVs have similar enough architectures to make it feasible.

Fermi, Kepler, Maxwell and now Pascal all use the same drivers for instance, and of course GCN is already on it's fourth iteration.

 

[Carfax83]

And AMD gets credit for going first and being the trendsetter.

Which is really what this is about, giving AMD credit. The thought of AMD developing Mantle for nothing (Mantle was doomed from the get go once Microsoft announced DX12) is so noxious to AMD fans, that they will come up with any explanation, even ones that defy reason, to justify AMD's expenditure and development effort.

What is a myth is to say that DX12 was directly derived from/is Mantle. That's pretty baseless, but I don't think anyone in this thread was saying that.

It's been implied throughout this entire thread.

And no, "no direct relationship" is not the same thing as "no connection at all". Again, please stop distorting AMD's statement.

The only known connection they have, is that they are both low level APIs.. That's it.

 

[Carfax83]

http://www.extremetech.com/gaming/1...rectx-12-will-imitate-and-destroy-amds-mantle
just saying.

OK, so DX12 supposedly imitates and duplicates Mantle, yet DX12 supports many more IHVs and architectures, whilst Mantle only supports on IHV and one architecture..

Hmmmm, something isn't adding up.

People can claim anything once they say,"We've spoken to several unnamed sources with additional information on the topic," doesn't mean it's true.

That article you cited also claimed that DX12 wasn't a death knell for Mantle. Look how that turned out?

 

[master_shake_]

OK, so DX12 supposedly imitates and duplicates Mantle, yet DX12 supports many more IHVs and architectures, whilst Mantle only supports on IHV and one architecture..

Hmmmm, something isn't adding up.

People can claim anything once they say,"We've spoken to several unnamed sources with additional information on the topic," doesn't mean it's true.

That article you cited also claimed that DX12 wasn't a death knell for Mantle. Look how that turned out?

the article i quoted was from 2014 no one even knew what ms was going to call it.

can you provide anything to refute the article?

 

[Red Hawk]

Which is really what this is about, giving AMD credit. The thought of AMD developing Mantle for nothing (Mantle was doomed from the get go once Microsoft announced DX12) is so noxious to AMD fans, that they will come up with any explanation, even ones that defy reason, to justify AMD's expenditure and development effort.

Projecting much? We know that Mantle wasn't developed for nothing, it directly carried over into Vulkan (which is part of the point @richaron tried to make, and BFG just ignored when accusing him of spreading misinformation). And that pretty clearly translated into an advantage with Doom's Vulkan renderer out of the gate for AMD.

And as I've discussed, the idea that Mantle influenced DX12's development hardly "defies reason".

It's been implied throughout this entire thread.

Or you've been strawmanning it throughout this whole thread.

The only known connection they have, is that they are both low level APIs.. That's it.

Which is not the same as there being no possibility of any connection at all.

 

[nurturedhate]

See, the thing is, this conversation about DX12/Mantle isn't about discussing which came first or who influenced who. It is to derail the thread away from the OP and positive AMD news which is the fact that these new drivers improve performance.

 

[Zstream]

OK, so DX12 supposedly imitates and duplicates Mantle, yet DX12 supports many more IHVs and architectures, whilst Mantle only supports on IHV and one architecture..

Hmmmm, something isn't adding up.

People can claim anything once they say,"We've spoken to several unnamed sources with additional information on the topic," doesn't mean it's true.

That article you cited also claimed that DX12 wasn't a death knell for Mantle. Look how that turned out?

Ok mate, whatever you say. DX12 came about because Microsoft needs to bring the PC to the console. Additionally, to keep cost low, they need every % from the CPU and GPU. Mantle came about because the development was mainly done and didn't require many resources to release.

 

[ElFenix]

came in to read about a new driver, got treated to an argument instead.

i swear some of you would argue about whether pizza is any good or not.

 

[Tarvaln]

Thin crust is superior to regular crust! It has less calories!

 

[esquared]

OK folks. back on topic or I am going to lock this.

You want a which-came-first-mantle-or-DX12 thread? Make one.

Next post on that off topic conversation will be infracted.


esquared
Anandtech Forum Director

 

[Red Hawk]

Thin crust is superior to regular crust! It has less calories!

I'll have you know that thick crust is the best. Better texture and more filling.

Anyways, about the actual topic. If AMD did reduce driver CPU overhead, then great. I haven't tested much to see if it did. I've heard talk about reduced CPU overhead with past AMD updates though, and it never seemed to amount to much, so I am somewhat skeptical.

Doom running in Vulkan was mentioned in the OP, and for my part I did briefly test the driver with Doom w/Vulkan on my brother's aging Q6600/R9-270X/8 GB DDR2 RAM machine, which Vulkan certainly helps over OpenGL but is still CPU bound. One particular area it bottlenecks is the start of the Argent Tower level -- it chugs under 20-30 FPS with shadows turned just to medium, but turning shadows down to low and leaving most everything else on high lets it go up to 40-50 (running at 900p). Shadows tend to be a more CPU bound task, indicating the level really needs the CPU cycles. Carfax has talked about how the draw calls seem to be crazy in that level. Anyways, the driver update didn't really help. The game still chokes on that level with shadows on medium.

 

[Maverick177]

I'll have you know that thick crust is the best. Better texture and more filling.

Anyways, about the actual topic. If AMD did reduce driver CPU overhead, then great. I haven't tested much to see if it did. I've heard talk about reduced CPU overhead with past AMD updates though, and it never seemed to amount to much, so I am somewhat skeptical.

Doom running in Vulkan was mentioned in the OP, and for my part I did briefly test the driver with Doom w/Vulkan on my brother's aging Q6600/R9-270X/8 GB DDR2 RAM machine, which Vulkan certainly helps over OpenGL but is still CPU bound. One particular area it bottlenecks is the start of the Argent Tower level -- it chugs under 20-30 FPS with shadows turned just to medium, but turning shadows down to low and leaving most everything else on high lets it go up to 40-50 (running at 900p). Shadows tend to be a more CPU bound task, indicating the level really needs the CPU cycles. Carfax has talked about how the draw calls seem to be crazy in that level. Anyways, the driver update didn't really help. The game still chokes on that level with shadows on medium.

Try the beginning of the 2nd level where I got 10~15fps + from 16.8.3

 

[Headfoot]

You sound angry. Why is that?


Please stop spreading misinformation. DX12 was in development years before Mantle even existed.

As for "radical", the industry used to code to the metal years ago. It was called DOS and Assembler. Was DOS "radical"? How about x86 assembler?


On Windows DX12 will gain about as much traction as PhysX did, Vulkan even less so. Generally speaking only the biggest AAA engine houses will be able to afford the resources to properly code for DX12. Windows has thousands of games released for it every year, including huge numbers of indie games written using garbage-collected C#/.NET. The industry has been moving away from low level APIs for years.

Even if we take what we have so far, about half of DX12 games have been a shocking failure, running slower and/or buggier than DX11. The mantra "things will get better" doesn't hold given the hardware is ever changing, and DX12 paths will need constant app-side patching to maintain performance advantages for new hardware. Almost every single Mantle game now runs slower than DX11 on newer AMD hardware, yet it's quietly swept under the rug.

This post is imminently quotable for the inevitable day when DX12 sees vastly more adoption than GPU PhysX (which its getting close to already doing). These indie games you want to hold up are written on commercially available engines like Unity, Unreal, CryEngine, etc. And those engine developers will certainly (and already have) go to DX12 and thus anything built on them can be too. But yeah, it's really hard for indies to turn on DX12 in Unity.

DX11 overhead is an issue and will continue to be an issue because DX11 or DX11 compatibility level within DX12 (e.g. FL_11) wont be going away though it will most assuredly step down from the majority position, so it's good that this happened.

I do wonder whether the WDDM 2.1 thing has any bearing on this as previously mentioned. Part of why DX12 is locked to Windows 10 was supposed to be due to the changes in the driver model. So if this is a tangible benefit from that, then it makes me feel better about it being locked

 

[HurleyBird]

But yeah, it's really hard for indies to turn on DX12 in Unity.

It's not hard, but there's no reason to, at least at the moment as the DX12 renderer still needs a lot of work. Lower performance, lower stability, and lower compatibility. I expect that it will get better in the future though.

 

[redzo]

I do wonder whether the WDDM 2.1 thing has any bearing on this as previously mentioned. Part of why DX12 is locked to Windows 10 was supposed to be due to the changes in the driver model. So if this is a tangible benefit from that, then it makes me feel better about it being locked

DX12 is locked to W10 because Microsoft wants so. Microsoft wants to push W10. It has nothing to do with DX12 technicalities: vulkan doesn't care about our/yours windows version.

Independent/low budget studios will never release DX12/vulkan titles. It is out of their league. Supporting in engine DX12/vulkan is hard due to the nature of this lower level optimizations. Only AAA studios can afford this, and even they are late with it -> it always comes later with patches.

Unless a big high performance engines like UE4, ForstB or CryEngine drop the ball with full native DX12 support, both vulkan and DX12 are doomed to play catch-up with DX11. ... and AMD also has to do its best in DX11.

 

[dogen1]

DX12 is locked to W10 because Microsoft wants so. Microsoft wants to push W10. It has nothing to do with DX12 technicalities: vulkan doesn't care about our/yours windows version.

Yes it does. It didn't have to be WIN10 only, but there are technical advantages to it.

 

[nurturedhate]

Independent/low budget studios will never release DX12/vulkan titles.

Never? Wanna edit that?

 

[Headfoot]

I am loving all of these "Never" and "Always" statements. Got a lot of contenders for my quote slot in the sig. It'll be funny when DX12 in Unity works great (undoubtedly not a far ways off given how closely Unity has been working with MS and the IHVs) and tons of indies hit the really really hard "Use DX12" button.

Still, Kudos for reduced overhead in existing APIs. Reducing overhead is exactly what they needed to do.

 

[MajinCry]

Isn't the Direct3D 11 overhead difference due to AMD's GPUs having a hardware scheduler; less CPU burden and is better suited towards Direct3D sans command lists. Whereas NVidia uses a software scheduler, that allows for a decent gain in draw call performance at the cost of burdening the CPU more.

Though I do remember reading that NVidia had better draw call performance in Direct3D 9, but the tests were always bunk; coupling a snazzy razzle-dazzle-best-of-the-best-£1000-deluxe proccy with GPUs frae both camps.

Draw call overhead is best tested on weak CPUs, since, y'know, that's the sort of hardware where it's going to be noticeable as well as impacting.

 

[VirtualLarry]

Isn't the Direct3D 11 overhead difference due to AMD's GPUs having a hardware scheduler; less CPU burden and is better suited towards Direct3D sans command lists. Whereas NVidia uses a software scheduler, that allows for a decent gain in draw call performance at the cost of burdening the CPU more.

Except the skinny on the streets is the opposite - that NV has lower CPU overhead than AMD.

 

[IEC]

Haven't had any issues with this latest driver version so far, and the performance uplift has been nice. Now if they'd just hurry up and release Vega!

 

[ThatBuzzkiller]

DX12 is locked to W10 because Microsoft wants so. Microsoft wants to push W10. It has nothing to do with DX12 technicalities: vulkan doesn't care about our/yours windows version.

Independent/low budget studios will never release DX12/vulkan titles. It is out of their league. Supporting in engine DX12/vulkan is hard due to the nature of this lower level optimizations. Only AAA studios can afford this, and even they are late with it -> it always comes later with patches.

Unless a big high performance engines like UE4, ForstB or CryEngine drop the ball with full native DX12 support, both vulkan and DX12 are doomed to play catch-up with DX11. ... and AMD also has to do its best in DX11.

Actually, there is a technical reason why Microsoft locked DX12 behind Windows 10 and that's because it requires the new driver display model (WDDM 2.0) and there are benefits to be had for that decision like exposing GPU virtual memory and residency ...

Vulkan can not get the same benefits as DX12 if it does not run on Windows 10 ...

 

[dark zero]

Seems that AMD's way to improve DX 11 is very interesting. Maybe AMD realized that giving all the power to DX 12 won't be as great as expecting and giving some legacy support is not bad.

 

[Red Hawk]

DX12 is locked to W10 because Microsoft wants so. Microsoft wants to push W10. It has nothing to do with DX12 technicalities: vulkan doesn't care about our/yours windows version.

Well. Vulkan cares if you have Windows Vista or older.

 

[Shivansps]

Ive tested the new driver and i did not noticed ANY improvement on ANY of the game im playing right now... (W3, F4, SC(999% cpu limited)).

Call me when they fix the DX11 MT that right now is just broken, depending on 1 core is MUCH, MUCH worse than some minimal overhead removal.

Some people need to accept the fact that games are not, and probably not be drawcall limited for some time, otherwise we should be seeing MASSIVE fps gains by moving to DX12 whiout having to use stuff like Async Compute, that is not avalible to DX11 anyway.

This post is imminently quotable for the inevitable day when DX12 sees vastly more adoption than GPU PhysX (which its getting close to already doing). These indie games you want to hold up are written on commercially available engines like Unity, Unreal, CryEngine, etc. And those engine developers will certainly (and already have) go to DX12 and thus anything built on them can be too. But yeah, it's really hard for indies to turn on DX12 in Unity.

DX11 overhead is an issue and will continue to be an issue because DX11 or DX11 compatibility level within DX12 (e.g. FL_11) wont be going away though it will most assuredly step down from the majority position, so it's good that this happened.

I do wonder whether the WDDM 2.1 thing has any bearing on this as previously mentioned. Part of why DX12 is locked to Windows 10 was supposed to be due to the changes in the driver model. So if this is a tangible benefit from that, then it makes me feel better about it being locked

You have no idea of what you are talking about, DX12 is SLOWER on Unity3d and Unreal... its easy to activate it, but performance is plain CRAP.

if you are not conviced, 2 months ago i compiled a Unity demo, and enabled DX12 on it so people could tell me if anyone gained by going to DX12, the result was just a big fact 0, just massive negative gains.
Feel free to try it yourselft.
https://www.dropbox.com/s/z0jn3p2gnubcv6c/DX12Test.rar?dl=0