Nvidia’s GameWorks program usurps power from developers, end-users, and AMD

[CAT-THE-FIFTH]

More details here:

http://www.extremetech.com/extreme/1...-users-and-amd

The effects of this programme were already seen with Batman:Arkham Origins!



I posted that on OcUK a while back in a discussion about the latest Batman game.

The chap is the editor at Hardware.fr,the biggest hardware review website in France.

[CAT-THE-FIFTH]

[herulach]

Interesting, although not that surprising. To be fair to work though, this does seem like one of those things that is genuinely a hardware issue, as I read that extreme tech article they demonstrated a performance improvement by turning down image quality.

Relative performance of mantle on Cuda will be an interesting counterpoint.
I appreciate that's an open api, but I'd be very surprised if it isn't designed with the way gcn cores run in mind

[jigger]

WTF

Can't they see that pulling this kind of crap is hurting Nvidia customers... Nvidia is run by halfwits.

[KeyboardDemon]

Interesting, although not that surprising. To be fair to work though, this does seem like one of those things that is genuinely a hardware issue, as I read that extreme tech article they demonstrated a performance improvement by turning down image quality.

Relative performance of mantle on Cuda will be an interesting counterpoint.
I appreciate that's an open api, but I'd be very surprised if it isn't designed with the way gcn cores run in mind

I think that by having it as an open standard means that if nVidia wanted to update their drivers or optimise them to work with Mantle they could as they would have access to the code behind Mantle while at the same time they are not giving AMD the same level of access in order for AMD to optimise their drivers to work with GameWorks. Or maybe I have just misunderstood it all, but I would hate to see a day when I decide on wether or not I want to buy a game based on the GPU I have in my PC.

[kalniel]

Interesting, although not that surprising. To be fair to work though, this does seem like one of those things that is genuinely a hardware issue

I don't know if refusing to allow AMD to code improvements is a hardware issue. Seems quite unlikely.

[Biscuit]

Typical nVidia anti competitive behaviour :-(

[herulach]

I don't know if refusing to allow AMD to code improvements is a hardware issue. Seems quite unlikely.

From the article (emphasis mine):

There’s a second reason to be dubious of Arkham Origins: it pulls the same tricks with tessellation that Nvidia has been playing since Fermi launched. One of the differences between AMD and Nvidia hardware is that Nvidia has a stronger tessellation engine. In most games, this doesn’t matter, but Nvidia has periodically backed games and benchmarks that include huge amounts of tessellation to no discernible purpose. Arkham Origins is one such title.

Here’s what the ground looks like in Arkham Origins: First, the actual model, and second, the wireframe in DX11. Pouring triangles into these surfaces can make them look subtly more realistic, but it’s also a cheap way to disadvantage a competitor. Deep performance inspection reveals that the R9 290X takes 30-40% more time in tessellation per frame than an equivalent Nvidia card.

I've no idea whether the increased level of tesselation in Arkham Origins is the result of nvidia meddling in the design to hurt performance on AMD or not, however, I suspect its more likely to be lazy programming (made all the more likely when you consider the slowdowns in some scenes due to the ridiculous number of physics objects - the scarecrow bits for example).

TBH I'd give the whole article a lot more credence if the conclusions matched the rest of the article - essentially what they're saying is that one studio hasn't included improvements AMD built in, and has used a suboptimal means of making something look good. They then go on to demonstrate that if you force the image quality down, you can improve performance - derp.

You could level the exact same criticisms at any dev not building anything from scratch. What if unreal engine doesn't work on Maxwell, what if Speedtree is faster on Kaveri than an i7?

I;m not for a second suggesting its a good situation, just that its understandable that if the entire engine is built on nvidia's libraries, I'd imagine its difficult to pull part out an replace it with AMDs. The opposite will be equally true.

I think that by having it as an open standard means that if nVidia wanted to update their drivers or optimise them to work with Mantle they could as they would have access to the code behind Mantle while at the same time they are not giving AMD the same level of access in order for AMD to optimise their drivers to work with GameWorks. Or maybe I have just misunderstood it all, but I would hate to see a day when I decide on wether or not I want to buy a game based on the GPU I have in my PC.

Mantle is an open standard, so they can optimise drivers yes, but if for example Mantle is heavily dependent on bandwidth in a particular part, or performance of a particular hardware path then its going to be at best expensive to optimise for it in 8XX series cards, if not impossible.

[kalniel]

Also from the article:

AMD attempted to provide Warner Bros. Montreal with code to improve Arkham Origins performance in tessellation, as well as to fix certain multi-GPU problems with the game. The studio turned down both.

[herulach]

Also from the article:

Exactly, but as i said, that doesn't neccessarily mean its anything to do with gameworks. Without knowing the reason why that was declined its very circumstantial - if it happens again its a pattern, but until then, there might have been a perfectly valid reason for doing it.

[scaryjim]

... there might have been a perfectly valid reason for doing it.

A valid reason for a game publisher refusing to accept code that will improve the experience of ~ 1/3 of their user base (based on Steam's hardware survey)? I can't think of many, and certainly no good ones...

[shaithis]

A valid reason for a game publisher refusing to accept code that will improve the experience of ~ 1/3 of their user base (based on Steam's hardware survey)? I can't think of many, and certainly no good ones...

It broke something
It had undesirable effects
It required a lot of man hours to implement

I could go on........

It does seem like the GPU market is seeing a lot of changes at the moment, I just hope nVidia and AMD start to play nice with each other and we can all benefit, although I can't see it happening any time soon :S

[herulach]

A valid reason for a game publisher refusing to accept code that will improve the experience of ~ 1/3 of their user base (based on Steam's hardware survey)? I can't think of many, and certainly no good ones...

Valid probably not the best word - but non nvidia related is certainly feasible. Publishers pushing deadlines and not wanting to redo testing seeming to be the obvious two, but you could also choose laziness, ambivalence, incompatibility with other code, adversly impacting the performance on nvidia hardware, wanting the cape/snow to react in a particular way, odd interactions with the PhysX code etc theres any number of reasons that aren't neccessarily good, but have nothing to do with nvidia.

For clarity, I'm not a fanboy (you'll note not too dissimilar comments in the mantle thread) and I don't think any of them are good reasons, but nothing in that article proves the headline. If another developer does the same thing, then maybe, but in the meantime I'm much more open to believing in lazy developers than nvidia doing something anti competitive in such a blatant fashion (I'd fully believe they'd do something, I just think they'd be more subtle about it. Paying people to make something perform worse on other hardware smacks of leaving themselves open to lawsuits).

[scaryjim]

... Publishers pushing deadlines and not wanting to redo testing seeming to be the obvious two ...

I'd buy that, if the twitter exchange CAT posted didn't clearly indicate that a) AMD met WB's own deadlines, and b) WB outright refused the code. It shows a blatant lack of respect for both AMD, and any potential customer who's running AMD hardware. They could easily accept the code but give no guarantee as to when it'll be patched in to the game; to just refuse the code looks childish at best.

I think the more pertinent point in the article is that GameWorks locks out developers from understanding the shader code - yes, the tool makes development easier, but ultimately it stops being a genuine DX11 game and becomes a GameWorks game: AMD and Intel can't optimise their drivers for GameWorks, because NV are being closed about the internal workings. Neither side is entirely guilt-free in dev relations - DIRT: Showdown was a classic example of what single-vendor code optimisation can do - but NVs cheap tesselation shots just smell of desperation: they do nothing to improve in-game experience and everything to screw over a competing graphics vendor.

Releasing dev tools that allow them to control the performance of competing cards? Nothing to do with supporting PC gaming, everything to do with supporting their own bottom line. At least Mantle will be an open API - AFAICT GameWorks turns a theoretically open API into an nvidia proprietary one. Can't see how that could ever be a good thing.

[shaithis]

Another interesting point in this general area is that, AFAIK, Batman:AO actively blocks AMD/nVidia combos that have been hacked together. Whether that's a side-effect of using newer GameWorks libraries or whether nVidia are turning the screw on the developers they have in their pocket, I am not sure.

[herulach]

I'd buy that, if the twitter exchange CAT posted didn't clearly indicate that a) AMD met WB's own deadlines, and b) WB outright refused the code. It shows a blatant lack of respect for both AMD, and any potential customer who's running AMD hardware. They could easily accept the code but give no guarantee as to when it'll be patched in to the game; to just refuse the code looks childish at best.

I think the more pertinent point in the article is that GameWorks locks out developers from understanding the shader code - yes, the tool makes development easier, but ultimately it stops being a genuine DX11 game and becomes a GameWorks game: AMD and Intel can't optimise their drivers for GameWorks, because NV are being closed about the internal workings. Neither side is entirely guilt-free in dev relations - DIRT: Showdown was a classic example of what single-vendor code optimisation can do - but NVs cheap tesselation shots just smell of desperation: they do nothing to improve in-game experience and everything to screw over a competing graphics vendor.

Releasing dev tools that allow them to control the performance of competing cards? Nothing to do with supporting PC gaming, everything to do with supporting their own bottom line. At least Mantle will be an open API - AFAICT GameWorks turns a theoretically open API into an nvidia proprietary one. Can't see how that could ever be a good thing.

I'm not suggesting for a second that it is. My issue is that the article clearly tries to give the impression that this was some nefarious intervention by nvidia, when actually all there is evidence of is lazy development practices.

I'd still make the point that mantle whilst an open API will be designed to give a competitive advantage on AMD hardware if it takes off (for example, SSE which required hardware support took amd 2 years to implement). If it isn't then AMD are daft.