IHV-specific Enhancements in Games

Well, for the next few months the GCN-iest GCN thing i'll have is a 7730M, but that's reasonably feature complete as far as a GCN implementation goes so, if time allows it, everything is possible (if not necessarily probable)

 

So, basically what the OP is saying is that it's OK for Nvidia to do IHV specific things (Hardware accelerated PhysX, for example) but not for AMD to release something that works across all IHV hardware?

Granted the game isn't out yet. So there's the entire possibility for AMD to follow in Nvidia's footsteps and block this feature (Nvidia with AA in Batman: AA). But I can't see them doing that as they've never done it in the past that I'm aware of.

I don't see Nvidia attempting to optimize anything for AMD hardware. So I'm not sure why AMD should try to optimize for Nvidia hardware. That stuff would be up to the developer to do if they so chose. Assuming, of course, that it's even possible (IE - some hardware limitation prevents a new technology feature from performing as fast on the competition).

So, AMD is now bad for doing what Nvidia has been doing for the past 5-10 years?

Nvidia should be the only one applauded for trying to push graphics technology forward (and at times keeping it proprietary to themselves), while AMD should be booed for doing the same (and making it available for everyone)?

I really don't get the logic for some people.

Regards,
SB

 

Seriously.

Though I am glad Nvidia's learned some lessons of the past and didn't try to make Crysis 2's Tessellation an Nvidia only feature...

 

Just tried HK2207, Mecha and Leo on my 680 and they all work fine. Strange that nobody has thought to benchmark these before.

 

HK2207 is over 2 years old and Nvidia went out of their way to bash and mock it when it was released, google it. Leo could not run on Fermi and hell would freeze over the day Nvidia acknowledges an AMD tech demo much less use it as a benchmark.

 

Neither of those things prevent an independent reviewer from using them as a benchmark.

 

Most people won't complain about high levels of tessellation if they're necessary to improve a silhouette. People complain when a flat surface is heavily tessellated when it would have been better served by a simple normal map.

You also have a complaint about the effect always being in frame. That's a great place for any good effect to be. If a game developer, whether pushed by Nvidia or not, wants to tessellate a main character like Lara I bet most reasonable people would be fine with it assuming the tessellation is not overdone past the point where there's no possible gain in image quality.

I bet most cases of character tessellation won't require extremely high levels of tessellation in order to achieve pixel sized triangles and they will perform reasonably on AMD hardware. I say there won't be high tessellation levels because I think you'll need a good number of patches for a character pre-tessellation. Whether the character looks or performs better than a non-tessellated mesh on either IHV's hardware is the question.

 

That's what I said, basically. Pitcairn XT does what, 2.5 TFLOPS? GTX 680 is just over 3. So, in my books that's approximately the same ballpark - give or take a bit depending on test.

edit:
Used their percentage function in the diagrams...

GTX 680 vs. HD 7870 (100%):
Fluid3D: 69%
Fluid2D: 93%
Mandel vector: 94%
Mandel scalar: 106%
QJulia: 89%
CS Raytracing (8x8): 136%
CS Raytracing (16x16): 143%
DX SDK FluidCS11 64k particles, Grid+Sort: 81%
DX SDK FluidCS11 16k particles, shared memory: 151%
--
Avg: 106,9% (962/9)
Compute: 107,77% (3,09/2,86 TFLOPS)


GTX 680 vs. HD 7870 (100%):
Fluid3D: 92%
Fluid2D: 114%
Mandel vector: 100%
Mandel scalar: 109%
QJulia: 106%
CS Raytracing (8x8): 151%
CS Raytracing (16x16): 152%
DX SDK FluidCS11 64k particles, Grid+Sort: 93%
DX SDK FluidCS11 16k particles, shared memory: 126%
--
Avg: 115,8% (1043/9)
Compute: 120,7% (3,09/2,56 TFLOPS)

 

OK, what I mean is that it is out of line with the rest of its "game" positioning.

 

Ahh, yes - now I get it.

 

This is my thought when reading the OP. At least AMD believe in IHV-independent feature sets for achieving these things. It's the whole purpose of DirectCompute with DX11 and the reason why DirectX exists.

Create the best hardware you can, looking forward to future focuses of features without trying to tie those features to your hardware only and I will buy your products. It's why I haven't bought Nvidia for years.

 

I'm loathe to beat this decaying horse carcass but what IHV independent API do you imagine nVidia should have used for hardware accelerated physics back then?

It's far more interesting when everyone can play the game though. I wonder whether UE4 will push the envelope on the upcoming consoles. Looks like they're sticking with PhysX. Frostbite and Cryengine are presumably going to remain CPU based.

 

I think the issue for PhysX is that Nvidia never opened it up, not that they own it. the only way they would have let AMD have it is if AMD made their shit CUDA compatible and that, well, wasn't gonna happen.

Edit: Mild fact correction. Damn brainfarts...

 

That's not the issue. Microsoft's DirectX is not open either and AMD has no problem supporting it. It's just a matter of not wanting to get in bed with the competition.

After years of haggling over this PhysX nonsense it's very telling that there still isn't a CPU based physics engine that handles fluid and particle interaction particularly well. Obviously GPU acceleration is a requirement but somebody has to step up to the plate with an OpenCL or Compute Shader interactive physics engine. TressFX is just a start.

 

there's Bullet & DMM2, both accelerated via OpenCL and/or DirectCompute. Why they're not being used practicly at all is a curious thing.

 

Possibly because they are quite new

 

Is it possible that PhysX may have stunted the potential for cross-hardware physics and developer uptake due to the fact that it didn't work on possibly half of their potential purchasers? The titles that do tout PhysX were basically funded by Nvidia to showcase their IP they obtained.

Sure Nvidia wanted to make money off their investment by selling more cards, I just don't think it has really been worth it for them, and definitely not for us, the gamers.

 

The new 3dmark uses Bullet, 3dmark 11 too I think. The result was less than impressive though.

 

Well nothing is stopping people from rolling their own engines. DICE and Crytek have done exactly that. PhysX can't hold back progress because it's just one of several options out there.

You could flip the question around and ask who is trying to push hardware agnostic physics acceleration? The stuff AMD is doing with hair in Tomb Raider is the first and only example I can think of.