Particle Effects On Amd Graphics Cards

[Sacia]

I was curious if there was any intention of including particle effects for those of us who have AMD GPU's on our PC.  I know warframe is being developed for the PS4 and Xbox One (Both of which are using AMD cards). Is this something we can be looking forward to in the future?

 

Share the sparkly love with all of us!

[SkacikPLz]

Well, the thing with particle effect:

A)they have simplistic version of it (available without physx)

b)they have fancy version of it (available only with physx)

 

The main thing with fancy particles is that they're rely heavily on GPU, unlike standard particles.

If DE wanted to have similar feature for both GPUs they'd have to re-write their particle system entirely and then fine tune it for both Nvidia GPUs and AMD gpus.

 

THEN, they would have to fine tune it for each of the popular GPUs from both vendors.

 

Also, thing with PhysX is that it's fairly simple and easy to implement/use.

Nvidia supplies developers with more or less ready to implement binaries which are easy to hook up to any engine.

Additionally they provide software and plugins for creating PhysX objects like Clothing and Particles, whereas if any developer would wish to create similar middleware on their own, they would have to create all workflow software/plugins on their own.

 

Doable? Yes.

Reasonable to do when you already have PhysX? No.

 

//

Also, even if Consloles had Nvidia and physx ready GPUs. Believe me, one Nvidia GPU from mainstream power shelf would NOT be enough to run the game in 1080, high detail AND PhysX particles.

 

Next gens are much better than the old ones, but by no means they hold up to decent modern PC

Edited September 17, 2013 by SkacikPLz

[Saenol]

I would not count on additional PhysX particles for users of non-NVIDIA graphics cards.

 

It's never been an issue of hardware, but one of licensing.

 

Microsoft and Sony are not NVIDIA competitors and had no reason not to enter into a licensing agreement with NVIDIA.

 

AMD is a direct NVIDIA competitor and would be committing suicide if they tied themselves to a competitor's proprietary products.

 

Reasonable to do when you already have PhysX? No.

 

There is nothing about PhysX that prevents it's effects from running in full on the CPU, or even on other graphics hardware.

 

There is utterly no NVIDIA hardware in the XBox One or PlayStation 4 (indeed, they both have AMD CPUs and GPUs), yet they should have full PhysX support.

 

NVIDIA wants to push their proprietary middleware, while AMD wisely wishes to continue pushing OpenCL and avoid paying royalties or allowing themselves to be tied to a competitor.

Edited September 17, 2013 by Saenol

[BoomyGordo]

Physx relys on many cores rendering the particles. AMD CPUs while powerful only have about 4-8 cores  which is great but cant handle calculating the floating points needed to render hundreds or thousands of particles and the fluid interaction of those particles with the envoironment. Nvidia does this by having usualy upwards of 20+ cores on their cards. i wont pretend to know specifics but usualy its only 4-8 cores dedicated for rendering graphics and then all the rest are for physx alone

[SkacikPLz]

 

There is nothing about PhysX that prevents it's effects from running in full on the CPU, or even on other graphics hardware.

This is true, however the way PhysX is made makes sure that it'll bring your FPS to crawl if it's run on anything else than Nvidia GPU.

Trying to run it on CPU or AMD GPU will not end well for your gameplay experience.

[Saenol]

Physx relys on many cores rendering the particles. AMD CPUs while powerful only have about 4-8 cores  which is great but cant handle calculating the floating points needed to render hundreds or thousands of particles and the fluid interaction of those particles with the envoironment.

 

There is no question that GPUs are better at handling this sort of workload, but even a modern mid-range CPU can handle many thousands of PhysX particles without issue.

 

You can test this yourself with any of a number of PhysX benchmarks out there.

 

usualy its only 4-8 cores dedicated for rendering graphics and then all the rest are for physx alone

 

PhysX is generally no where near as demanding as rendering graphics in most games, and I do not believe GPUs split up work like this anyway.

 

 

This is true, however the way PhysX is made makes sure that it'll bring your FPS to crawl if it's run on anything else than Nvidia GPU.

Trying to run it on CPU or AMD GPU will not end well for your gameplay experience.

 

You can't run it on an AMD GPU in a PC. However this is due to licensing issues, not the hardware not being up to task.

 

NVIDIA writes PhysX in CUDA and while it could likely be ported over to OpenCL or D3D11 without difficulty, and with equivalent performance, no one can legally do this except NVIDIA, and their incentive to do so is nil.

Edited September 17, 2013 by Saenol

[Letter13]

Another one of these threads. 

 

Nope.

 

You can't run it on an AMD GPU in a PC. However this is due to licensing issues, not the hardware not being up to task.

 

Actually that's not true. AMD GPUs do not use CUDA architecture; GPU based PhysX uses specialized hardware instructions that can only be executed on CUDA or hardware with an architecture that supports the custom PhysX instructions. AMD GPUs do NOT support these hardware instructions, and literally cannot execute the PhysX instruction set because the hardware does not support it.

 

the way PhysX is made makes sure that it'll bring your FPS to crawl if it's run on anything else than Nvidia GPU.

Trying to run it on CPU or AMD GPU will not end well for your gameplay experience.

 

Sort of. The issue here is that CPUs are designed to handle the majority of hardware instructions and code efficiently; PhysX (and most Physics) calculations are floating point multiplies and divisions; these hardware instructions are INCREDIBLY demanding when it comes to CPU cycles, and floating point calculations aren't executed that often on CPUs so they're not wholly optimized for it.

 

Graphics and 3D Rendering are another type of computational task that requires a significant amount of floating point calculations; as such GPUs have been optimized to kingdom come for executing floating point instructions (GPUs have a lot of cores. Like a ton of 'em. This is great for floating point). 

 

When Nvidia bought Aegia and the PhysX platform, they began to dabble in CUDA technology (which effectively allowed Nvidia GPUs to be tapped into for their incredible floating point calculating ability by normal software). By allowing complex floating point calculations to be handed off from the CPU to the GPU, the speedup of physics based calculations was astronomical, while allowing the CPU to remain free enough to provide a smooth experience to the user. It's the same for GPU based encoding/decoding of video, and hardware acceleration in programs like Adobe Photoshop.

 

While ATI has a similar technology, it's nowhere near as robust or as widely used as Nvidia's CUDA tech.

[OTTERMATIC]

The way PhysX is designed (that's what's behind the particles), is that it relies heavily on Nvidia's card structure. AMD uses a completely different architecture, and you just can't get it to run on it, at least, not without a vast amount of work. It's just not worth the effort.

[GottFaust]

Little known fact: all physics implementations in this game use PhysX from the ragdolls, to the "non-physx" particles, to effects like Vauban's vortex. It's all done using the standard CPU implementation.

The advanced features required for the "PhysX" particles require Nvidia hardware to work without absolutely murdering your frame-rate however. Hence them being Nvidia locked. It's not a question of power, but one of proprietary hardware. GPU Accelerated PhysX requires CUDA cores to run. ATI would have to liscense the use of this core architecture to be able to utalize it. This is assuming that Nvidia is even willing to let them.

The PS4 implementation is using a special hybrid setup proprietary to the PS4 itself and still results in the loss of most of the complex movement and a large particle count reduction according to Steve.

[CrunkTanium]

People posting in this thread are clueless,

 

#1 Cuda is not needed for PhysX not do you need an Nvidia card as even ARM processors are capable according to Nvidia.

#2 Games are configured to block the effects of PhysX when AMD cards are detected which can often be enabled (not sure about WF) via manual settings or driver hacks.

#3 AMD/ATI cards are perfectly capable and AMD actually has it's own CUDA implementation but Nvidia does not play nice with others.

#4 While Nvidia has been dropping the ball lately with drivers ATI has been pushing out trash drivers for over a decade.

#5 Only an idiot buys an AMD card for gaming knowing full well that nearly every game developer is in bed with Nvidia.

 

Contrary to some headlines, the Nvidia PhysX SDK actually offers multi-core support for CPUs. When used correctly, it even comes dangerously close to the performance of a single-card, GPU-based solution. Despite this, however, there's still a catch. PhysX automatically handles thread distribution, moving the load away from the CPU and onto the GPU when a compatible graphics card is active. Game developers need to shift some of the load back to the CPU.

 

http://www.tomshardware.com/reviews/nvidia-physx-hack-amd-radeon,2764.html

http://www.tomshardware.com/news/nvidia-physx-apple-iphone,7313.html

[GottFaust]

People posting in this thread are clueless,

 

#1 Cuda is not needed for PhysX not do you need an Nvidia card as even ARM processors are capable according to Nvidia.

#2 Games are configured to block the effects of PhysX when AMD cards are detected which can often be enabled (not sure about WF) via manual settings or driver hacks.

#3 AMD/ATI cards are perfectly capable and AMD actually has it's own CUDA implementation but Nvidia does not play nice with others.

#4 While Nvidia has been dropping the ball lately with drivers ATI has been pushing out trash drivers for over a decade.

#5 Only an idiot buys an AMD card for gaming knowing full well that nearly every game developer is in bed with Nvidia.

 

Contrary to some headlines, the Nvidia PhysX SDK actually offers multi-core support for CPUs. When used correctly, it even comes dangerously close to the performance of a single-card, GPU-based solution. Despite this, however, there's still a catch. PhysX automatically handles thread distribution, moving the load away from the CPU and onto the GPU when a compatible graphics card is active. Game developers need to shift some of the load back to the CPU.

 

http://www.tomshardware.com/reviews/nvidia-physx-hack-amd-radeon,2764.html

http://www.tomshardware.com/news/nvidia-physx-apple-iphone,7313.html

You fail to see the part where they are already using CPU PhysX to the best of its capabilities. All of those rag-doll, "non PhysX" particles, and effects like Vortex are entirely done via CPU PhysX. The "PhysX" option you see there is to turn on GPU accelerated PhysX and all the benefits that it brings. Regardless of what you may think, GPU Accelerated PhysX requires CUDA cores which are proprietary Nvidia hardware. ATI will never have those unless they license them from Nvidia directly.

Here's a quote from DE:

Te remove the confusion, the game uses CPU PhysX.

 

However, some advanced effects use NVidia propietary technology. This is what you can activate in the graphics options in Warframe. Halas, this technology wont work with ATI cards and there's nothing we can do about it. You can always email your concerns to NVidia.

 

We made sure that these effects do not interfere with the gameplay.

Source: https://forums.warframe.com/index.php?/topic/82785-physx-support/page-2#entry1085874