No PhysX on OpenCL soon

i heard a rumour that Microsoft was working on a direct X compute implementation which included yet another implementation of physics. Is this likely to be a pipe dream or could it be a reality?
 
i heard a rumour that Microsoft was working on a direct X compute implementation which included yet another implementation of physics. Is this likely to be a pipe dream or could it be a reality?

Microsoft doesn't provide implementations of anything via DirectX. They provide programming interfaces to access hardware functionality.
 
Well that's no different to Java's convenience/helper classes that wrap the lower level APIs. It's still way lower level than a full bown application / middleware package / engine etc, etc.
 
So out of 200 games that use PhysX, 81 of them are crap. The others simply use it because it comes with UE3. I see some decent titles there but I still think Nvidia's strategy of leaving other GPUs out will backfire in the future. It's Glide all over again.

Only commenting on the underlined part: you can't make that statement. In theory, they could all have opted for PhysX even if it hadn't come with the engine. We can't quantify how many would not so it's pointless to argue this.
 
So out of 200 games that use PhysX, 81 of them are crap. The others simply use it because it comes with UE3. I see some decent titles there but I still think Nvidia's strategy of leaving other GPUs out will backfire in the future. It's Glide all over again.
It's not: PhysX works on all brands of PC's and consoles. Glide did not. But PhysX has the additional bonus that it also sometimes works even faster on the GPU of a significant percentage of the users.

AFAIK Havok still doesn't have that bonus.

If your requirements are such that PhysX CPU is sufficient for your needs, why would you pay more and live with the no-GPU limitation of Havok?
 
...PhysX has the additional bonus that it also sometimes works even faster on the GPU...

Has this really been investigated? From the many things I've read about Physx I have not seen someone take the time to create a synthetic test to benchmark the CPU versus the GPU implementations of Physx. You'll excuse me if I exclude current game implementations of the tech because there is no way to know if they are purposely crippled when executed on CPU's. Even if the end result shows that it will in fact run faster on a GPU, I'd still like to know how much faster to determine if it is worth it to invest into the technology as a consumer.
 
Only commenting on the underlined part: you can't make that statement. In theory, they could all have opted for PhysX even if it hadn't come with the engine. We can't quantify how many would not so it's pointless to argue this.

You are right. Let me reword it: I have the gut feeling that many titles that use UE3 will also use PhysX because it comes bundled with the engine.

It would be very interesting to see games using UE3 + Havok and the reasons why the devs chose to go that way.
 
It's not: PhysX works on all brands of PC's and consoles. Glide did not. But PhysX has the additional bonus that it also sometimes works even faster on the GPU of a significant percentage of the users.

AFAIK Havok still doesn't have that bonus.

If your requirements are such that PhysX CPU is sufficient for your needs, why would you pay more and live with the no-GPU limitation of Havok?

The way I see it physics middleware is something that will end up running on the GPUs either using OpenCL or CS (or console equivalent for next gen). On PCs PhysX only supports CUDA and, what's worse, only does it if your primary graphics card is also an Nvidia part.

I got the impression from the Batman:AA thread that PhysX's performance of the CPU implementation left a lot be desired.

If you read the link that was posted before it makes it clear that Havok is currently the most complete solution with better support for devs and the choise for multiplatform.

So what exactly does NV expect from PhysX. Are they using it to try to move more GPUs? Are they going to try to get more AAA titles using it?

I'd love to see a benchmark showing whether PhysX is faster/slower than Havok when running on the CPU. If it's slower using PhysX on your game will be bad for a significant portion of the customers who don't have 2 NV GPUs.
 
It's not: PhysX works on all brands of PC's and consoles. Glide did not. But PhysX has the additional bonus that it also sometimes works even faster on the GPU of a significant percentage of the users.

This is debatable. PhysX may work without a GPU but if it drops the fps to single digits, is it relevant? You could run GLQuake in software afterall, if you could stomach framerates measured in seconds per frame.

If you mean CPU-based PhysX can scale back complexity to bring back fps... DOOM 3 works on a VooDoo2 but with those kinds of graphics, is it still DOOM 3?

ANYWAY, the point of the thread was PhysX and no OpenCL support (for now). What do you think this means? Downfall of OpenCL? Downfall of PhysX? Does this influence ATI in anyway (speed up/slow down their OpenCL implementation)?
 
This is debatable. PhysX may work without a GPU but if it drops the fps to single digits, is it relevant? You could run GLQuake in software afterall, if you could stomach framerates measured in seconds per frame.

PhysX does what every other package does AND it can scale up to run on GPUs too. It baffles me how this additional functionality could be spun as a negative thing.

ANYWAY, the point of the thread was PhysX and no OpenCL support (for now). What do you think this means? Downfall of OpenCL? Downfall of PhysX? Does this influence ATI in anyway (speed up/slow down their OpenCL implementation)?

It doesn't mean anything really considering there is no competition. PhysX's lack of OpenCL support isn't a disadvantage in a world where no OpenCL implementation exists.
 
PhysX does what every other package does AND it can scale up to run on GPUs too.

In my experience PhysX brings a colossal performance drop when running on the CPU while other physics middle-ware do not for similarly looking games. We can argue features versus performance but like my previous examples allude to: extra features at the expense of playable framerate aren't features worth having.

I.e. I'm ONLY discussing performance which was what silent_guy was overlooking when he says PhysX runs on all machines (CPUs). By the same token GLQuake runs on a Pentium MMX too: technically correct, but irrelevant in the real world.

It baffles me how this additional functionality could be spun as a negative thing.

If you can find in my post where that was spun that way.

It doesn't mean anything really considering there is no competition. PhysX's lack of OpenCL support isn't a disadvantage in a world where no OpenCL implementation exists.

Right, so it means something. ;) Go on...
 
In my experience PhysX brings a colossal performance drop when running on the CPU while other physics middle-ware do not for similarly looking games. We can argue features versus performance but like my previous examples allude to: extra features at the expense of playable framerate aren't features worth having.

Hmmm I can't think of an example of a game doing stuff on the CPU where PhysX would require a GPU to do the same thing. When you say "PhysX" you're referring to the whole platform and not just specifically the GPU accelerated bits right?

Right, so it means something. ;) Go on...

Nah, maybe it WILL mean something if the dynamic changes in the future and there's real competition. Do you have word of this happening sooner rather than later? :)
 
I'm not claiming that PhysX is at par or better than Havok feature-wise or performance-wise. It may or may not be: I simply don't know.

My argument is all about this: "If your requirements are such that PhysX CPU is sufficient for your needs..."

I don't think it's unreasonable to expect that there are many cases where you don't need the state-of-the-art and decent is good enough.

As for the question of whether the GPU is only faster because the CPU version is crippled, I don't buy it: Bullet physics is faster on a GPU than a CPU. Why would PhysX not be?
 
no ones saying its only faster because the cpu version is crippled, but nv are artificially enhancing the difference by doing no work on the cpu version

ps: bullet on the gpu ? im sure i read a whole lot of nvidia speil tha physx is the only gpu physics
 
Back
Top