Vega/Navi Rumors (Updated)

[CatMerc]

If there's so much potential in the cards that wouldn't take too much developer effort to extract, then AMD needs to ACTIVELY go around and invest time from their own engineers to get games to use them. If it's as easy as Glo is claiming, then it wouldn't take much of AMD's resources to help said developers get it up and running.

Sitting idly by won't work, even if there's potential for 3x 1080 Ti performance devs won't do it with coercion. AMD needs to send around engineers everywhere to get game developers to implement these features. And don't just go after new games, but look at the most popular games on the market right now. Overwatch, DOTA, PUBG, whatever.

 

[Glo.]

If there's so much potential in the cards that wouldn't take too much developer effort to extract, then AMD needs to ACTIVELY go around and invest time from their own engineers to get games to use them. If it's as easy as Glo is claiming, then it wouldn't take much of AMD's resources to help said developers get it up and running.

Sitting idly by won't work, even if there's potential for 3x 1080 Ti performance devs won't do it with coercion. AMD needs to send around engineers everywhere to get game developers to implement these features. And don't just go after new games, but look at the most popular games on the market right now. Overwatch, DOTA, PUBG, whatever.

The most questionable thing is this: are Primitive Shaders and Programmable Geometry Pipeline feature, just like Shader Intrinsics, are Vulkan and DX12, and potentially Metal 2 specific, or also available for DX11? Nobody who I consulted have said anything about DX11 but they were adamant that this is 100% Vulkan and DX12 feature compliant.

 

[PeterScott]

It actually allows for less work for developers, that they would have to with Fixed Function Shaders.

And this feature is base of Vega and every next generation AMD GPU architecture.

My reading is that provides more granular control, and that means lower level, and that means more developer work.

Even it was theoretically less work, it would still be more more work because developers still have to code the standard path.

So it is MORE work no matter what, and most developers won't chase more work unless they are paid to.

 

[Mopetar]

Does AMD have a compiler that can take PS4/XB1 code and automatically port it to the PC with the console GCN optimizations included, no fuss, plug and play?

I think that's missing the point. The optimizations needed to get the most out of Vega hardware do nothing for consoles because they don't (unless XB1X contains some of them) have those hardware features. Therefore they won't be made because of consoles at all, but only because some developer wants to add them for what is going to be minimal gain since Polaris cards also get nothing out of these optimizations for Vega-only features.

Perhaps we'll see some additional priority put towards these once the same hardware becomes available across AMDs product line. At some point there will probably be a Vega-based replacement for the Polaris line, and eventually Vega-based APUs will work their way into a lot of PCs and notebooks (which I think is even more overlooked than the consoles) and we'll see developers start to take advantage of its capabilities, but unless you plan on keeping an RX Vega for 2+ years, I don't think it's a good investment for someone who's gaming only (the argument changes somewhat if it's also used to mine).

 

[Crumpet]

With Volta now way off, Vega is even more of a serious consideration for me. Should be full reviews on Monday with some leaks over the weekend.

Aye what? What have I missed, link please???

 

[zinfamous]

Aye what? What have I missed, link please???

nVidia earnings report yesterday (direct comment from JHH) that GTX Volta is 1st or 2nd quarter 2018 (meaning--definitely 2nd quarter). I think "way off" is wrong, though, because the only valid information regarding Volta (meaning, from nVidia) was that it would only ever be a 2018 release (certainly for GTX, anyway--data center Volta was only ever a "possible" late Q4 2017 release).

All of the posting here and elsewhere about Volta showing up in 3 or 4 months is based on pure, wccftech-class rumors and nonsense.

 

[Phynaz]

It actually allows for less work for developers, that they would have to with Fixed Function Shaders.

Link showing it is an overall reduction in developer effort please.

 

[Arachnotronic]

nVidia earnings report yesterday (direct comment from JHH) that GTX Volta is 1st or 2nd quarter 2018 (meaning--definitely 2nd quarter). I think "way off" is wrong, though, because the only valid information regarding Volta (meaning, from nVidia) was that it would only ever be a 2018 release (certainly for GTX, anyway--data center Volta was only ever a "possible" late Q4 2017 release).

All of the posting here and elsewhere about Volta showing up in 3 or 4 months is based on pure, wccftech-class rumors and nonsense.

Um...no.

On data center Volta:

Next, data center, revenue of $416 million was up more than 2.5 times from a year ago. This growth, shared across AI, deep learning, high-performance computing, and GRID, is particularly notable, given that we announced and shipped production units of our Volta-based V100 accelerator as we transition from Pascal generation GPUs. Looking ahead, we see inferencing and video transcoding as emerging applications that are well suited for our GPUs. V100 was among the most important launches at this quarter's GPU Technology Conference [GTC]. It provides 10 times the deep learning power of its year-old predecessor, widely outpacing Moore's Law.

They shipped Volta last quarter in volume for data center.

On gaming:

Volta for gaming, we haven't announced anything. And all I can say is that our pipeline is filled with some exciting new toys for the gamers, and we have some really exciting new technology to offer them in the pipeline. But for the holiday season for the foreseeable future, I think Pascal is just unbeatable. It's just the best thing out there. And everybody who's looking forward to playing Call of Duty or Destiny 2, if they don't already have one, should run out and get themselves a Pascal.

Nothing about "Q1 or Q2 2018" here. Just an indication that Pascal will be here for the holiday season. Given the announcements we've seen around GDDR6 and the explicit mention of GDDR6 being built by SK Hynix for a high-end gaming GPU for early 2018, Q1 2018 makes sense.

 

[Malogeek]

Aye what? What have I missed, link please???

There isn't a precise timeframe, just that it's not holiday season or "foreseeable future". Since there still isn't a strong competitor to Pascal really, then why rush Volta? Likely it was scheduled for mid-2018 anyway so probably nothing has changed.

Anyway this is offtopic for this thread, discussion in the Volta thread.

 

[Ajay]

Edit: don't want the banhammer.

 

[Glo.]

Link showing it is an overall reduction in developer effort please.

Already gave you two films. Mike Mantor's discussion with Gamers Nexus, and Scott Wasson's discussion with RedGamingTech.
There are also slides, from AMD on this matter.

Those are opinions of game devs, which I have talked with about Primitive Shaders.

 

[railven]

Did JHH say this or one of the people during the call?

Volta for gaming, we haven't announced anything. And all I can say is that our pipeline is filled with some exciting new toys for the gamers, and we have some really exciting new technology to offer them in the pipeline. But for the holiday season for the foreseeable future, I think Pascal is just unbeatable. It's just the best thing out there. And everybody who's looking forward to playing Call of Duty or Destiny 2, if they don't already have one, should run out and get themselves a Pascal.

Not trying to be a cheerleader, but this is the kind of bravado/used_cars_salesmen pitches that sell products.

I never really payed attention to NV's presentations, but they are way out of AMD's league. And I know it's all marketing fluff, but so are pretty much any presentation, but outside of fireworks - JHH and his crew are so into the kool-aid, their delivery are almost perfect. I'd love to see this kind of flair coming from AMD, you can sell a questionable product if you have that fake smile and voice. Nerding out and performance numbers aside, marketing is a huge part of the machine.

 

[french toast]

Problem is consoles are going custom with own modifications, also due to backwards compatibility they are likely to be Polaris hybrids, not ideal for Devs to get an easy unified optimisation target.
Xbone x uses a custom Polaris solution sans fp 16 double pumped, ps4 pro contains this feature, desktop Polaris does not.

Thats just one thing out of a multitude of features the console manufacturers have diffentiated, this is likely to get worse.

On theflip side could AMD use r&d on the console SOCs to bring features over to desktop GPUs, for instance the custom command processor in xbone x that processes many draw calls on GPU without having to access CPU cycles, this sounds interesting but how feasible is this for desktop?

 

[Glo.]

Is this fake?

It puts it between ... GTX 1080 Ti and GTX 1080.

http://www.ashesofthesingularity.co...-details/5809922a-b7e7-4199-9692-9b8a3fec24cb

Fake?

 

[railven]

Problem is consoles are going custom with own modifications, also due to backwards compatibility they are likely to be Polaris hybrids, not ideal for Devs to get an easy unified optimisation target.
Xbone x uses a custom Polaris solution sans fp 16 double pumped, ps4 pro contains this feature, desktop Polaris does not.

Thats just one thing out of a multitude of features the console manufacturers have diffentiated, this is likely to get worse.

On theflip side could AMD use r&d on the console SOCs to bring features over to desktop GPUs, for instance the custom command processor in xbone x that processes many draw calls on GPU without having to access CPU cycles, this sounds interesting but how feasible is this for desktop?

I'm of opinion, basically based on Sony's current statements, B/C is something they have zero interest in supporting. They haven't hesitated to reselling us HD-Remasters or whatever you wanna call of it older games. And we all know the publishers haven't stopped either. It's a win-win for basically everyone at the expense of us consumers. I forgot what site did an article on it, but it was something like a small portion of Xbox One's own userbase doesn't use the B/C feature even though it was heavily requested.

Not implying consoles are moving away from AMD, but more so I don't see B/C being as important moving forward. With this generation of consoles stretched to freaking 2019 at the earliest, it's going to overstay its welcome just like last gen.

 

[raghu78]

Did JHH say this or one of the people during the call?

Not trying to be a cheerleader, but this is the kind of bravado/used_cars_salesmen pitches that sell products.

I never really payed attention to NV's presentations, but they are way out of AMD's league. And I know it's all marketing fluff, but so are pretty much any presentation, but outside of fireworks - JHH and his crew are so into the kool-aid, their delivery are almost perfect. I'd love to see this kind of flair coming from AMD, you can sell a questionable product if you have that fake smile and voice. Nerding out and performance numbers aside, marketing is a huge part of the machine.

Nvidia's strength is engineering followed by marketing. The former is what gives them market leadership . The latter cannot do anything without the former. Nvidia's confidence, flair or bravado whatever you call comes from good old fashioned solid engineering execution. One of the key strengths of Nvidia's engineering is that its primarily software driven. Nvidia has a great ecosystem driven by CUDA. That ecosystem locks companies, educational and research institutions in forever. Thats why any new challenger even if its the mighty Intel finds it difficult to make even a slight dent in Nvidia's dominance. Nvidia resembles Apple in a big way - even though both companies make great silicon, the real game changer is their software.

 

[Malogeek]

It puts it between ... GTX 1080 Ti and GTX 1080.

Only if there's another entry with a 12-core threadripper and a 1080 to compare it to.

 

[piesquared]

My reading is that provides more granular control, and that means lower level, and that means more developer work.

Even it was theoretically less work, it would still be more more work because developers still have to code the standard path.

So it is MORE work no matter what, and most developers won't chase more work unless they are paid to.

Some people keep attempting to frame this like it's a bad thing. Do you realize that this is what developers actually want? Having more control of the pipeline? Instead of a black box in which developers are largely blind in regards to the interaction with the driver, they rather have practically full control over that process. In the end, it makes developer work easier. This is known, and largely ignored, probably because it is easy for Nv online marketers to spin as a negative .

 

[Glo.]

Some people keep attempting to frame this like it's a bad thing. Do you realize that this is what developers actually want? Having more control of the pipeline? Instead of a black box in which developers are largely blind in regards to the interaction with the driver, they rather have practically full control over that process. In the end, it makes developer work easier. This is known, and largely ignored, probably because it is easy for Nv online marketers to spin as a negative .

Not entirely. If you have not so well documented architecture - yes, programming is hard for it.

If you have well documented hardware, direct control over it is what you want, without layer of asbtraction. GCN is very well documented, and AMD engineers are doing great job in extracting even more performance capabilities out of the hardware. GPUOpen.

Everything that adds more work in optimizing the software is considered wrong in game development. Primitive Shaders/Programmable Geometry Pipeline are adding different approach but simplify programming at the same time. Of course, different approach will require learning, but overall general consensus is that it is step in the right direction. Especially for VR.

 

[Phynaz]

It actually allows for less work for developers, that they would have to with Fixed Function Shaders.

None of the video you posted supports your argument. Not one. So we're down to believing your word that multiple graphics programmers have said so privately to you.

Sorry, but your reputation doesn't buy you that level of credibility.

 

[Phynaz]

Some people keep attempting to frame this like it's a bad thing. Do you realize that this is what developers actually want? Having more control of the pipeline? Instead of a black box in which developers are largely blind in regards to the interaction with the driver, they rather have practically full control over that process.

Do you believe the (lack of) popularity of DX12 and Vulcan supports this argument?

 

[PeterScott]

Some people keep attempting to frame this like it's a bad thing. Do you realize that this is what developers actually want? Having more control of the pipeline? Instead of a black box in which developers are largely blind in regards to the interaction with the driver, they rather have practically full control over that process. In the end, it makes developer work easier. This is known, and largely ignored, probably because it is easy for Nv online marketers to spin as a negative .

It doesn't make developer work easier. It makes optimizing for a specific architecture easier. Easiest is just using standard driver calls and skipping the optimizations.

It is very much like just coding using C++ and high level library calls vs dropping down into Assembly language to optimize something.

Almost no one bothers with Assembly language optimizations anymore, because it's a lot of work, and really ties you to a specific architecture.

In this case it's not just an AMD only optimization, it's a Vega only optimization. I really can't see anyone doing that with being incentivized by AMD.

 

[Glo.]

None of the video you posted supports your argument. Not one. So we're down to believing your word that multiple graphics programmers have said so privately to you.

Sorry, but your reputation doesn't buy you that position.

My reputation has nothing to do here.

The videos are showing you how it works. If you are familiar with what is GPU and application doing, you understand Primitive Shaders. And what this changes for developers. Ask by yourself any game developer, you can about this.

Learning curve - this can create a problem. But Primitive Shaders are actually saving some work connected to Fixed Function Shader programming, because you bypass two steps with just one

Do you believe the (lack of) popularity of DX12 and Vulcan supports this argument?

I genuinely say to everybody "good luck" programming games and applications for VR using DX11.

P.S. How long it took DX10/11 to be used more often? And Vulkan and DX12 are bringing bigger changes.

 

[Glo.]

It doesn't make developer work easier. It makes optimizing for a specific architecture easier.

Yup, this is pretty much it.

 

[Phynaz]

P.S. How long it took DX10/11 to be used more often?

Not the four years it's been since Mantle was announced.

I think we're done here.