Rumor: AMD cancel RX Vega Primitive Shaders from drivers

After CES2018, AMD told people in a breakout session, that they cancelled the implicit driver path for Primitive Shaders, they are now offering only explicit control to developers who wishes to implement it. So Primitive Shaders will only be ever implemented if a developer properly codes for it. The driver portion has been cancelled.

This comes directly from Marc Sauter (y33H@), editor at the german IT site Golem.de.

Source: https://www.forum-3dcenter.org/vbulletin/showthread.php?p=11610696#post11610696

Source 2: https://www.forum-3dcenter.org/vbulletin/showthread.php?p=11611522#post11611522

Source came from: https://www.reddit.com/r/Amd/comments/7sedpq/amd_cancels_the_driver_path_for_primitive_shaders/

If this true Lawsuit in coming from those who own RX Vega GPU.

 

Idk, it is shaddy no denying that..

But then again Nvidia promoted async compute with maxwell and you saw how that turned out..

 

Well no performance loss, but there will be no performance gain either. Vega owners are ****** as fury owners did after years. A wannabe enthusiast GPU "FINEWINE" edition. Many promises and useless features all over the place. AMD is a superior hype race officially. Neither Intel nor Nvidia lied so much ever before. Look at ryzen and vega slides. Then look at real performance numbers and satisfaction values. AMD chose the wrong way. They are gonna lose all RX and Ryzen users faster than the period took to acquire them. RIP.

 

Well the devs won't implement them, as AMD hardly have a Market Share, if any at all now.

 

Maxwell supports async compute, it's implementation just wasn't as robust as AMDs specific definition, so I'm not sure how that relates.

 

People have the right to get mad when companies do things like this behind customers back. like what Apple did to slow down your older phone without letting people know.

AMD had 6 months to let everyone know but failed to do so.

 

Well, here I was already thinking "another fanboy bashing for nothing" (sry @NvidiaFreak650 ).
Then I read up on what primitive shaders even are, and what they are supposed to do.

Then I don't understand, why are they cancelling it? Didn't it work before? In theory that's something really useful...

 

They aren't canceling it - the developer can still manually support it but initially AMD's driver was supposed to automagically replace the geometry shaders when it felt it could provide a benefit to geometry performance. It probably led to flickering/image artifact issues that they couldn't resolve in the automated process. So now the developer has to decide when to use it - which becomes a question of support, if the PS4 Pro and Xbox One X support primitive shaders then it's possible we will see some games with it. If it's just desktop vega, I doubt we'll see much of anything - that being said is geometry performance really that much of a bottleneck with Vega? I genuinely don't know.

 

Again, thanks for explaining.

 

Disabling it specifically for Maxwell or to have negative performance when enabled sounds about right.

My point was it's nothing new about promoting certain stuff then ditch it for inefficiency or compatibility reasons..

 

no API support, no party. Hopefully in DirectX 13 ('pardon, DirectX 14) and Vulkan 2.0 there will be a unification of the geometry pipeline....

 

There are quite a few wrappers/injectors for DX games. I guess those people can put this flag and override developer's choice.
Or even enable it for older games.

 

Didn't AMD also say they will enable something called "Draw Stream Binning Rasterizer" after Vega launch and improve performance?

https://overclock3d.net/news/gpu_displays/amd_vega_gpu_architectural_analysis/3

 

DSBR is active on Vega FE but only utilized in some pro applications - I haven't really heard about it's status on RX Vega cards.

DSBR isn't really a catch-all performance/power increase/decrease (respectively) like lots of people believe. Nvidia's Tom Peterson was interviewed on PC Perspective about it and basically said they enable it on a profile basis - it actually decreases shader performance, so it's only useful when memory bandwidth is a bottleneck and not shader performance. It's also known to cause graphical issues in some games/applications.

 

Why should them? Developers will never care about such huge difference in the shader pipeline just for one architecture (Vega) on one platform (PC).
Developers care more about features that are available on consoles and that could or will available on many architectures on PC like the features coming with Shader Model 6.1 and 6.2.

 

I am not writing about game developers, but about people behind software like re-shade or other wrappers/injectors which can override anything developer of game implements or does not.
I have seen wrappers adding shader code to old DX7 games, making them effectively DX8.1.

 

Intel better keep Koduri on a tight leash so that future Intel iGPU's aren't burdened with features that won't work and need to be quietly forgotten behind the scenes.

 

To be fair these features, both DSBR and Primitive Shaders require a lot of software development work, testing, etc. We have no idea what went on behind the scenes as far as RTG's budget and what got allocated towards the Zen project in the last few years. Typically it's about 2-3 years from initial design of a GPU architecture to the final shipping product. So Raja most likely planned these features with an expectation that he would have a software team capable of delivering them and then who knows what happened.

I'm not even sure what RTG's way forward is at this point - I think AMD needs to be firing on all cylinders for it's CPU division to keep Zen competitive, especially now that Intel is tripping up. On the GPU side, I don't think they have a chance at beating Nvidia. I think their polaris strategy of just targeting the masses with mid/low tier cards at extremely competitive price points is really the only way forward because it's extremely safe. I know people are hoping for a crazy MCM setup on the GPU side but I think the engineering effort both software/hardware is not something AMD can afford right now, if anything goes wrong and the product is a failure it's game over.

It's going to be interesting to see how it all plays out.

 

It aint over till the fat lady sings,

that being said this is not entirely surprising,
implementing them implicitly would be an insane amount of work many things can go wrong.