Widescreen Gaming Forum

I don't understand why ATI does not bring this to their cards. I actually really like the built in Physx on Nvidia cards I was using it on my gtx 260 but just got a 5870 for eyefinity. I really wish it supported Physx I don't feel like getting another card though just for Physx I am using my GTX 260 now in my HTPC so.......

I don't understand why ATI does not bring this to their cards. I actually really like the built in Physx on Nvidia cards I was using it on my gtx 260 but just got a 5870 for eyefinity. I really wish it supported Physx I don't feel like getting another card though just for Physx I am using my GTX 260 now in my HTPC so.......

It's for purely political reasons why it's not done. PhysX is a nVidia property. ATI and nVidia don't play well together. It's not that they can't, it's that they won't.

_________________
Brad Hawthorne
Product Manager
Nthusim Pty. Ltd. | www.nthusim.com

[quote]I don't understand why ATI does not bring this to their cards. I actually really like the built in Physx on Nvidia cards I was using it on my gtx 260 but just got a 5870 for eyefinity. I really wish it supported Physx I don't feel like getting another card though just for Physx I am using my GTX 260 now in my HTPC so.......

It's for purely political reasons why it's not done. PhysX is a nVidia property. ATI and nVidia don't play well together. It's not that they can't, it's that they won't.

I think you just mean the Word or logo Physx they could name it something else Havoc is another type of Physics. Soo... Why the hell wouldn't they I am just going to throw it out there and say I don't think they can. I know it would pull more people in on there cards. Same reason Nvidia doesn't want people using there card with a ATI card so now people have to use a hacked driver to get Nvidia Physx to work along side a ATI card.

ATI knows that more and more games are going to support this, and near future people are going to want this. Why wouldn't someone want physx enabled in games. People are just lying to them selves saying they don't care for it.

DirectCompute is something that can be used on DX10-capable cards.

If you have the DX SDK, there are several DirectCompute samples included. One of them is this beautiful ocean simulator. When you run it, you can open the task manager and look at CPU usage, and it uses 0-5% CPU. If needed, you can turn on wireframe mode just to see how complex the ocean simulator really is.

The DirectCompute samples included with the DX SDK are indeed using DX11 headers, so it's a DX11 feature that just works on existing video cards (which is good). I'll grab a screenshot of it later.


Anyway, as far as PhysX not being on ATI, I don't think it's a big deal. Almost no games used hardware PhysX, but plenty use software PhysX, Havok, or their own implementation. PhysX isn't a hardware-only library. There are plenty of games out there that use software PhysX just fine.

Using hardware PhysX does have an impact on FPS if you only have one video card, but you can actually pick up a <$100 USD nVidia 9800 card just for PhysX.

As a developer, I'm more concerned about ATI not having CUDA support, but again, that's something that doesn't matter to the rest of the people, and with OpenCL available now, it's even less of a problem.

_________________
Widescreen Fixer - https://www.widescreenfixer.org/

Widescreen Fixer Twitter - https://twitter.com/widescreenfixer
Personal Twitter - https://twitter.com/davidrudie

I never quite understood why it was decided that offloading physics processing from the CPU to the GPU (or PhysX card) was a good idea. My CPU, memory, and buses involved are more than capable of handling the workload (my CPU never seems to exceed 50% usage). Offloading onto the GPU just seems like a recipe for disaster with modern games bringing the top of the line video cards to their knees. Perhaps someone could explain this concept.

I never quite understood why it was decided that offloading physics processing from the CPU to the GPU (or PhysX card) was a good idea. My CPU, memory, and buses involved are more than capable of handling the workload (my CPU never seems to exceed 50% usage). Offloading onto the GPU just seems like a recipe for disaster with modern games bringing the top of the line video cards to their knees. Perhaps someone could explain this concept.

That's the idea of having a dedicated card for it.

The GPU is capable of processing physics better than a CPU due to it's high core and processing unit count. You can spawn a lot of threads to handle all the individual particles and the GPU will love it because it's designed for that.

A CPU processing the same data that hardware PhysX handles will result in a much slower game than if you ran them on the GPU. However, keeping your CPU and GPU free by installing another video card just for handling PhysX, and you have little to no loss in performance at all.

_________________
Widescreen Fixer - https://www.widescreenfixer.org/

Widescreen Fixer Twitter - https://twitter.com/widescreenfixer
Personal Twitter - https://twitter.com/davidrudie

Here are those shots of the OceanCS Direct Compute sample:






Here's some links to a video I captured of it:
Flash Player Version
HTML5 Version (Requires Chrome or Safari)

_________________
Widescreen Fixer - https://www.widescreenfixer.org/

Widescreen Fixer Twitter - https://twitter.com/widescreenfixer
Personal Twitter - https://twitter.com/davidrudie

I never quite understood why it was decided that offloading physics processing from the CPU to the GPU (or PhysX card) was a good idea. My CPU, memory, and buses involved are more than capable of handling the workload (my CPU never seems to exceed 50% usage). Offloading onto the GPU just seems like a recipe for disaster with modern games bringing the top of the line video cards to their knees. Perhaps someone could explain this concept.

Think of hardware physics not as additional load that would fit in idle CPU cycles, but as many many loads.
Offloading physics to a dedicated card is a "good idea" for processing massively parallel stuff - even though it's not what physics always are. It's not necessary for physics acceleration per se, but it's an interesting route for "advanced" physics (breakable environnements, realistic cloth & liquid simulation) and a necessity for "extreme" physics (objects that break in persistent debris which are in turn affected by physics, in a swimming pool, surrounded by flags, under falling hail). It's thanks to the offloading that you can get the insane amount of bricks and glass shards piling up and flying about in duly destroyed UT3 PhysX maps. This is an extreme, tech-demo-ish example of course but in a world with CPU-bound physics, no one would've bothered designing stuff like that.
Just my non-dev way of seeing things anyway.

PhysX?

:roll:

Physx on CPU's even i7's have lagged from some reports, it sounds like it is only single threaded on the cpu and 1 core isn't fast enough to do all the calculations.

XWarrior, its just random advanced visuals.
Mirror's Edge shows it pretty well http://www.firingsquad.com/hardware/mirrors_edge_physx_performance/

Basically what dopefish said, if you are already on Nvidia, throw in a cheapo card with a handful of stream processors and dedicate it to physx (will be cheaper than the aegia cards ever were)

If you are ATi use the work around to throw a nvidia card in

25 bucks for a 8400gs
50 dollar area for a geforce 210/9400gt