Hawken, holy gpu physx!

[parvadomus]

So because it happend a long time ago...I should just forget it? LOL

And I should be gratefull that instead of moving on from X86, AMD's AMD64 made us stuck on x86?

I don't see AMD as a "positive"....quite the opposite.

:'(
You should see it, if AMD moved from X86 they would be dead by now. And Intel would be a monopoly.

 

[Jaydip]

Intel is already a monopoly.

 

[cmdrdredd]

Intel is already a monopoly.

So there is no AMD CPU?

 

[Olikan]

But noot good enough...so back to square one.

the link was comparing a 244 watt gpu against a 95 watt cpu, yet the cpu didn't fall much behind....

but you still think that gpus are overall better for physics?

heh!.... Nvidia is going to ignore AVX-2 for years

 

[Lonbjerg]

the link was comparing a 244 watt gpu against a 95 watt cpu, yet the cpu didn't fall much behind....

but you still think that gpus are overall better for physics?

heh!.... Nvidia is going to ignore AVX-2 for years

I have heard all claims about what will kill PhysX.
Once it was NVIDIA and their GPU's.

Now the "magic" is called AVX2.

Let me ask ask you this, since AVX2 is SO powerfull...we will change to a software render (CPU)when it comes out right?

If not...I would wait on predicting the death of PhysX...since people like you have declared it dead since 2006.

Best working dead software I ever have seen.

 

[Lonbjerg]

SirPauly
Sorry, was forced to take that down since YouTube has messed it up a little.

Here is the right one
http://www.youtube.com/watch?v=PAlC3VSN4Ks

--------------
Sorry, Lonbjerg, I can't send private messages until I'll get 25 posts =)
I haven't experienced any audio sync issues so far. However, I never uploaded RAW fraps footage. I currently rendering videos with WMV V9 6Bps profile (Audio: 192 Kbps, 48,000 Hz, 16 Bit, Stereo, WMA9).

Post more then
Love you videos and I am re-encoding my files to see if that does the trick..odd that you have to record....and...re-encode...in order for Youtube to encode the file correctly :S

 

[Jaydip]

So there is no AMD CPU?

Yes but sadly irrelevant.

 

[Olikan]

I have heard all claims about what will kill PhysX.
Once it was NVIDIA and their GPU's.

Best working dead software I ever have seen.

It is alive because Nvidia keeps begging for developers to use it

read here, it's a very tecnical thread about cpu rendering, first page they speak about cpu physics , see rest of the pages to see how impressive avx-2 is

 

[Lonbjerg]

It is alive because Nvidia keeps begging for developers to use it

read here, it's a very tecnical thread about cpu rendering, first page they speak about cpu physics , see rest of the pages to see how impressive avx-2 is

ZZZZzzzzzZZZZZzzzZZZZ...I have heard it all before.

PhysX is still alive, 7 years after it was declared soon dead...it's will survive AVX2 too...as it will help PhysX run better on the CPU...but still below a GPU.

Logic have a way of going away when on a crusade Ollie...unless you are saying PhysX is uncapable of utilizing AVX2 in the future? ^^

 

[Olikan]

Logic have a way of going away when on a crusade Ollie...unless you are saying PhysX is uncapable of utilizing AVX2 in the future? ^^

PhysX can use AVX2, yes...but will be redundant, AVX2 alone will do a better job than the gpu itself...

unless, of course, nvidia keep adding TONs of particles

 

[BallaTheFeared]

PhysX will be dead by then.

 

[Lonbjerg]

PhysX can use AVX2, yes...but will be redundant, AVX2 alone will do a better job than the gpu itself...

unless, of course, nvidia keep adding TONs of particles

So we are going software redering is what you say?

If not.....your post is hillarious...but useless.

 

[BallaTheFeared]

So we are going software redering is what you say?

If not.....your post is hillarious...but useless.

Not to mention x87 is FPU only, which is what PhysX uses.

Integer is faster, but less accurate. SSE2 does both. AVX2 adds integer.......

 

[NTMBK]

Not to mention x87 is FPU only, which is what PhysX uses.

Integer is faster, but less accurate. SSE2 does both. AVX2 adds integer.......

AVX2 also adds gather, which will be useful for both integer and FP.

 

[Olikan]

So we are going software redering is what you say?

If not.....your post is hillarious...but useless.

at physics, dude

but if you did read the link, cpu rendering is possible... if used some crazy cpu with more than 20 cores

 

[Lonbjerg]

Not to mention x87 is FPU only, which is what PhysX uses.

Integer is faster, but less accurate. SSE2 does both. AVX2 adds integer.......

Google PhysX SDK 3.0

SSE2 has replaced x87..and didn't do anything...

 

[Olikan]

Google PhysX SDK 3.0

SSE2 has replaced x87..and didn't do anything...

Cloth - SSE2: 771.9 FPS
Cloth - x87: 175.3 FPS

Mesh - SSE2: 881.1 FPS
Mesh - x87: 360.4 FPS

ForceField - SSE2: 463.8 FPS
ForceField - x87: 351.5 FPS

http://forum.gamevicio.com/i/topicos/97/97219-nvidia-physx-primeiros-testes-usando-sse2/index.html


it was the first that i found, it's in portuguese... i am Brazilian

 

[Lonyo]

Cloth - SSE2: 771.9 FPS
Cloth - x87: 175.3 FPS

Mesh - SSE2: 881.1 FPS
Mesh - x87: 360.4 FPS

ForceField - SSE2: 463.8 FPS
ForceField - x87: 351.5 FPS

http://forum.gamevicio.com/i/topicos/97/97219-nvidia-physx-primeiros-testes-usando-sse2/index.html


it was the first that i found, it's in portuguese... i am Brazilian

Yeah but that's only a 32% to 340% increase in performance, which is nothing.
And it's Portugese, so it's clearly biased and/or wrong. NV couldn't purposefully hobble PhysX CPU performance and changing from x87 to SSE2 wouldn't help because that was debunked ages ago.

 

[SirPauly]

Developers wouldn't use CPU PhysX if nVidia purposely sabotaged the CPU -- there are many more CPU PhysX applications than GPU, imho!

 

[Reticula]

Lonyo
Yeah but that's only a 32% to 340% increase in performance, which is nothing
Actually, that performance gain was not from SSE.
It was because CPU cloth solver was rewritten in PhysX SDK 2.8.4, giving it 2x-3x performance boost compared to SDK 2.8.3. And all of those test scenes are using either cloth or softbody.

Originally, it was posted by Geeks3D
http://www.geeks3d.com/20100922/physx-sdk-2-8-4-optimized-cpu-cloth-simulation/

We have tested a gain from SSE in FluidMark (or, at least, I think it was related to SSE) - around 15%
http://physxinfo.com/news/4267/fluidmark-1-2-2-compiled-with-newest-physx-sdk-2-8-4/

 

[railven]

Yeah but that's only a 32% to 340% increase in performance, which is nothing.
And it's Portugese, so it's clearly biased and/or wrong. NV couldn't purposefully hobble PhysX CPU performance and changing from x87 to SSE2 wouldn't help because that was debunked ages ago.

That sarcasm taste good!

Who ever wants to do it, I frankly don't care at this point, just someone do it, and hopefully it's Intel so they can control the market, whip us as their slaves and hopefully shut Lonberg up.

 

[96Firebird]

Lonyo
Yeah but that's only a 32% to 340% increase in performance, which is nothing
Actually, that performance gain was not from SSE.
It was because CPU cloth solver was rewritten in PhysX SDK 2.8.4, giving it 2x-3x performance boost compared to SDK 2.8.3. And all of those test scenes are using either cloth or softbody.

Originally, it was posted by Geeks3D
http://www.geeks3d.com/20100922/physx-sdk-2-8-4-optimized-cpu-cloth-simulation/

We have tested a gain from SSE in FluidMark (or, at least, I think it was related to SSE) - around 15%
http://physxinfo.com/news/4267/fluidmark-1-2-2-compiled-with-newest-physx-sdk-2-8-4/

I love when people who actually know what they are talking about post, it takes away from the bull that most other people post.

:thumbsup:

 

[Lonbjerg]

Lonyo
Yeah but that's only a 32% to 340% increase in performance, which is nothing
Actually, that performance gain was not from SSE.
It was because CPU cloth solver was rewritten in PhysX SDK 2.8.4, giving it 2x-3x performance boost compared to SDK 2.8.3. And all of those test scenes are using either cloth or softbody.

Originally, it was posted by Geeks3D
http://www.geeks3d.com/20100922/physx-sdk-2-8-4-optimized-cpu-cloth-simulation/

We have tested a gain from SSE in FluidMark (or, at least, I think it was related to SSE) - around 15%
http://physxinfo.com/news/4267/fluidmark-1-2-2-compiled-with-newest-physx-sdk-2-8-4/

Thanks for saving me the time.
I participated in a thread on this forum in 2010 back when David Kanter took a shot at PhysX for X87 vs SSE.
I thought we got it settled back the (the anti-PhysX crowd left the thread after benches proved NVIDIA hadn't gimped PhysX on the CPU).

Sadly I can see that 3 years later, the debunked argument is still being used

 

[BallaTheFeared]

I love when people who actually know what they are talking about post, it takes away from the bull that most other people post.

:thumbsup:

Refreshing isn't it? :thumbsup:

 

[Lonyo]

Thanks for saving me the time.
I participated in a thread on this forum in 2010 back when David Kanter took a shot at PhysX for X87 vs SSE.
I thought we got it settled back the (the anti-PhysX crowd left the thread after benches proved NVIDIA hadn't gimped PhysX on the CPU).

Sadly I can see that 3 years later, the debunked argument is still being used

So Kanter took a shot about x87 vs SSE, and that only gains some performance.
Rewriting the code to make it work nets a 2x to 3x performance increase, but CPU PhysX isn't gimped at all.
Because 2x to 3x is a normal improvement to get when you optimise code.