Anandtech Review Summerize Up Pascal's "Async Compute" Truth

[jabbadap]

RX 480 gaining what, 30% in Doom (Vulkan). I wouldn't call that "small".

Well that gain is not all because of async compute(gain from async on/off with vulkan is about 5-15% depending on system and place of the game). Doom is the first game to take advantage of gcn shader instrinsic(close to metal=console like) which really give them performance boost. Nvidia has no alternative as it uses glsl shaders for vulkan.

 

[Steevo]

http://www.hardocp.com/article/2016...x_1060_founders_edition_review/5#.V5D1JDW1UhQ

DX12 and Vulkan makes a RX480 perform like a 980Ti, which is essentially a GTX1070.

 

[FordGT90Concept]

Well that gain is not all because of async compute(gain from async on/off with vulkan is about 5-15% depending on system and place of the game). Doom is the first game to take advantage of gcn shader instrinsic(close to metal=console like) which really give them performance boost. Nvidia has no alternative as it uses glsl shaders for vulkan.

The boost in Vulkan comes from AMD's long frame render time using OpenGL. Vulkan brings AMD cards up where they should be when running OpenGL (we're talking something massive like 20-30% here). OpenGL to Vulkan should only get about a 5-10% performance boost. Async compute is also 5-10% on AMD cards depending on how much of the GPU is idle.

Vulkan fixes problems AMD never bothered to (or couldn't) address in OpenGL.

 

[the54thvoid]

http://www.hardocp.com/article/2016...x_1060_founders_edition_review/5#.V5D1JDW1UhQ

DX12 and Vulkan makes a RX480 perform like a 980Ti, which is essentially a GTX1070.

We know that.

But that's not the point of the thread. But still, this thread is going to degenerate into another 'my band's bigger than yours' thread.

THE OP was talking about why Nvidia abandoned compute style hardware and AMD didn't. It's not about which is better - it's about why. And you're long enough in the tooth to know how much we AMD (sorry ATI) owners, back in the day slagged off Fermi for it's power consumption. Nvidia reacted to that, dropped a lot of compute and DP and went for frequency and efficiency over sheer hardware. AMD plugged away at hardware, (look at the transistor and shader counts) and you have a situation where yes, the RX 480 under Vulkan does well but it consumes nearly the same power as a GTX1070.
Moreover - that's one title. ONE title. Nvidia gambled on DX11 being around for a while longer.

All DX12 is doing is leveling the current hardware playing field (which is great). Choose AMD for DX12 specialisation (developer dependent) or choose Nvidia for the bulk of current DX11 (and future DX11 games). Hell, Pascal is fast enough that it still works with DX12 (just a fraction under the RX480 in Hitman - an AMD Async heavy game).

So please, let's not do a 'this card is better than that card' debate. It'd be great if it got more technical - like why @theoneandonlymrk should be folding on 390X's.

 

[yogurt_21]

you've piqued my interest.

why is Fiji so bad at folding compared with Hawaii and Tahiti?

 

[TheoneandonlyMrK]

We know that.

But that's not the point of the thread. But still, this thread is going to degenerate into another 'my band's bigger than yours' thread.

THE OP was talking about why Nvidia abandoned compute style hardware and AMD didn't. It's not about which is better - it's about why. And you're long enough in the tooth to know how much we AMD (sorry ATI) owners, back in the day slagged off Fermi for it's power consumption. Nvidia reacted to that, dropped a lot of compute and DP and went for frequency and efficiency over sheer hardware. AMD plugged away at hardware, (look at the transistor and shader counts) and you have a situation where yes, the RX 480 under Vulkan does well but it consumes nearly the same power as a GTX1070.
Moreover - that's one title. ONE title. Nvidia gambled on DX11 being around for a while longer.

All DX12 is doing is leveling the current hardware playing field (which is great). Choose AMD for DX12 specialisation (developer dependent) or choose Nvidia for the bulk of current DX11 (and future DX11 games). Hell, Pascal is fast enough that it still works with DX12 (just a fraction under the RX480 in Hitman - an AMD Async heavy game).

So please, let's not do a 'this card is better than that card' debate. It'd be great if it got more technical - like why @theoneandonlymrk should be folding on 390X's.

No I shouldn't and I would have if 700 watts applied to folding was what I was after, my rigs made for 3 gpus and so it Will be 1 day ,and it will use less than 700 watts for 3 cards doing more for less.
The 480 is on the same par in actual folding performance terms as a 390X using half the power though I'd agree on paper the 390X looks good ,folding @home isn't solely double precision Intense it seams.
Off topic though I apologise but I was mearly answering your question,its clear though that my cards lack of Aces has a cost.
That cost applies to nvidia more as they have no Ace equivalent, and their polymorph engine clearly is purely optimized for graphics work, nvidia stripped all the allegedly unnecessary bits out of its low to high end gpus to fit less but more robust shader arrays in place then clock the snot out of them, fair enough and for five years its not hurt them but times change and the next five will see a different approach from them I'm sure.
No dev in their right mind fully ignores compliant hardware across multiple formats so dx12 and vulkan Will gain support relatively quickly compared to dx11 imho as dx11 was sole platform based yet did get used eventually.
As for some comments regarding nvidia catching up to Amds tesselation performance,that depends on your outlook I suppose because to catchup to Amds 4th gen arch tesselation performance nvidia had to over egg it hence some of the Gtx480-580 heat ie nvidia use much more silicon real estate to get the job done then Amd ever will and the same will apply to async compute, nvidias next new architecture is going to knock some sideways with its lack of efficiency gains imho.

 

[jabbadap]

you've piqued my interest.

why is Fiji so bad at folding compared with Hawaii and Tahiti?

It's a double precision test. Fiji fp32/fp64 ratio is 16:1, Hawaii has ratio of 8:1 and Tahiti 4:1(for radeons, not firepros obviously). Thus the double precision numbers for the cards are Fiji:2*1.05*4096/16=537.6GFlops, Hawaii:2*1*2816/8=704Gflops and Tahiti:2*0.925*2048/4=947.4Gflops. So with looking pure shader power Tahiti should be on top on that test, but peak compute power is always hard to reach in real world applications.

 

[little cat]

Who is using double precision for gaming !?

 

[jabbadap]

The boost in Vulkan comes from AMD's long frame render time using OpenGL. Vulkan brings AMD cards up where they should be when running OpenGL (we're talking something massive like 20-30% here). OpenGL to Vulkan should only get about a 5-10% performance boost. Async compute is also 5-10% on AMD cards depending on how much of the GPU is idle.

Vulkan fixes problems AMD never bothered to (or couldn't) address in OpenGL.

Well yeah of course you are right about that.

Spoiler: Uhh maybe we are going a bit off topic here

But Vulkan is low level API: You can do simple port which work with all and get that lower cpu overhead benefits(or in bad ports not). But the thing is you can still get much more out of it using vendor specific extensions(AMD:VK_AMD_rasterization_order, VK_AMD_shader_trinary_minmax, VK_AMD_shader_explicit_vertex_parameter, VK_AMD_gcn_shader Nvidia: VK_NV_glsl_shader, VK_NV_dedicated_allocation). Now amd gives you low level extensions of it's own architecture while nvidia gives you just it's higher level OpenGL abstraction. Until nvidia opens up it's own architecture, I would say amd will have upper hand on vulkan(if game developer uses amds extensions).

iD will reveal this years siggraph2016, which will start next sunday, more about doom vulkan and usage of gcn specific extensions:

The devil is in the details: idTech 666

Abstract: A behind-the-scenes look into the latest renderer technology powering the critically acclaimed DOOM. The lecture will cover how technology was designed for balancing a good visual quality and performance ratio. Numerous topics will be covered, among them details about the lighting solution, techniques for decoupling costs frequency and GCN specific approaches.

Presenters:
Tiago Sousa (id Software), Jean Geffroy (id Software)

 

[GhostRyder]

Reference from Original Review article:
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/9

Regarding Maxwell2 Gen cards(980/970)


Regarding Pascal's improvement in terms of Async Compute



So yes, Nvidia did try to improve the Async compute from Maxwell2's non-existent state to a better condition. However under careful anaylsis it is predicted such a small change won't provide much help during DX12/Vulkan age.

Regarding AMD:




In the end it boils down to how soon Nvidia have chosen to make the shift from old ways to unified ALUs.


Then what is pre-emption?




Nvidia tried to cater to current gen and older games by implementing more traditional(that is suitable for DX11) way of GPU design to ensure max performance gain with minimum cost of R&D as well as production. At the same time Nvidia introduces some new workaround for DX12/Vulkan titles that are compute heavy. In this way you get happy customers and great return on revenue for the company. People will buy new cards down the road for DX12/Vuklan titles any way, If I were to work for Nvidia this will be a safe and sound strategy for GPU development for sure.

The generation that comes after Pascal will probably see drastic design change comparing to Maswell-Maxwell2-Pascal. Nvidia will shift heavily towards hardware level async compute and comes out with a super robust design. And maybe by that time the entire Tessellation situation will repeat itself again.

AMD on the other hands have been shifting towards async compute for a long time. I strongly doubt this is related to console development. After all what SONY or MS want is a hardware they can milk on for a long time without worrying the graphic/Visual lacking behind PC too much. DX12/Vulkan helps unlock max potential of hardwares which will great for console development.

My verdict:
Nvidia focus more on PC Consumer experience, which always aims for best performance during the GPU's usefulness life span.

AMD focus more on Console. PC GPU design may just be tagging along. Sometime their assumed "futuristic design" may help in new games. However as soon as Nvidia as shifts towards the direction AMD's design becomes obsolete.

Completely agree, its Nvidias focus vs AMD's focus and quite frankly I do like the idea of the "Future" more at times because I don't replace my cards every year. Unfortunately, one if very obviously winning even if there are big benefits to each idea. But that idea of them always focusing on the "Future" is not the biggest problem this round.

The problem though this round still comes down to one major thing, without a card in the higher areas of the market it does not matter where their focus is whether Async Compute or not because your only pandering to the middle ground and lower (Granted thats a huge majority area). Right now, since there is nothing competing against the 1070/80 and probably won't be for the foreseeable future (Least before December unless something drastic changes) there is nothing that is going to help their reputation and ever shrinking market share especially because the GTX 1060 competes very well in most areas against it (or better). Even with Async Compute and Vulkan getting big gains to AMD even if it say put thcard right at a GTX 1070, thats still not going to be every game running like that which still puts it to far down the list to be something truly special.

When it comes down to it, all that has happened is a waiting game to see what else can happen. Frankly with this situation, I don't have high hopes anymore for anything changing unless either a miracle occurs for AMD or Radeon Graphics get bought. This is unfortunate, because it makes things boring since we already know what our choices are going to be...Least on the high end.

 

[R-T-B]

Who is using double precision for gaming !?

We were talking folding, not gaming.

 

[Frag_Maniac]

I still think most of the problems lie on the game development, not the hardware end. Right now most developers are getting endorsement from Nvidia, and that means they probably also have better insight on what hardware features are going to be needed in the near future. Dx12 so far has been an insignificant factor, as much so as 6 or 8 core CPUs.

 

[Deleted member 50521]

Completely agree, its Nvidias focus vs AMD's focus and quite frankly I do like the idea of the "Future" more at times because I don't replace my cards every year. Unfortunately, one if very obviously winning even if there are big benefits to each idea. But that idea of them always focusing on the "Future" is not the biggest problem this round.

The problem though this round still comes down to one major thing, without a card in the higher areas of the market it does not matter where their focus is whether Async Compute or not because your only pandering to the middle ground and lower (Granted thats a huge majority area). Right now, since there is nothing competing against the 1070/80 and probably won't be for the foreseeable future (Least before December unless something drastic changes) there is nothing that is going to help their reputation and ever shrinking market share especially because the GTX 1060 competes very well in most areas against it (or better). Even with Async Compute and Vulkan getting big gains to AMD even if it say put thcard right at a GTX 1070, thats still not going to be every game running like that which still puts it to far down the list to be something truly special.

When it comes down to it, all that has happened is a waiting game to see what else can happen. Frankly with this situation, I don't have high hopes anymore for anything changing unless either a miracle occurs for AMD or Radeon Graphics get bought. This is unfortunate, because it makes things boring since we already know what our choices are going to be...Least on the high end.

With the way GCN is designed there is no way for Vega to compete against pascal for current gen games. AMD can stuff up to 8192ALU into vega and make it run at 2GHz. However without developer support to properly utilize the massive amount of ALUs with async engine, AMD is still gonna fail.

 

[little cat]

I still think most of the problems lie on the game development, not the hardware end. Right now most developers are getting endorsement from Nvidia, and that means they probably also have better insight on what hardware features are going to be needed in the near future. Dx12 so far has been an insignificant factor, as much so as 6 or 8 core CPUs.

i dealt with other software not games but if a core is dedicated to the OS , another to the player and the third to the enemies it gives 3 cores for parallel processing !

 

[Frag_Maniac]

i dealt with other software not games but if a core is dedicated to the OS , another to the player and the third to the enemies it gives 3 cores for parallel processing !

It only matters what IS being done in game development, not what CAN be done. At the look of the current state of things, it's questionable if PC gaming will EVER be up to the the capability of the hardware. Sadly it's a very political thing, controlled mostly by console systems manufacturers.

 

[Deleted member 50521]

Take home message of my entire discussion:

If I were to run nvidia, I would totally NOT implement async even for Pascal.

 

[FordGT90Concept]

Pascal can't because the hardware can't handle it. They need a new architecture with something like GCN's Asynchronous Compute Engine: something that finds the idle shaders and puts them to work on something else.