Phuture of PhysX?

[kumquatsrus]

just ran across the review of the gtx 280 from guru3d and it was talking about physx implementation via cuda on 8 series gpu's and beyond.

http://www.guru3d.com/article/geforce-gtx-280-review-test/9

it seems there are quite a few games being developed with physx in mind. and then i ran across these lists of physx titles out or in production.

http://www.nzone.com/object/nzone_physxgames_home.html

http://personal.inet.fi/atk/kjh2348fs/ageia_physx.html

so my question is how is the future of physx looking? it appears there are more games with physx support now with nvidia in the picture. does this mean the ppu still has a use for the near foreseeable future? will physx via nvidia gpus work as well as the standalone ppu? is it possible for the ppu to be used in tandem along with an nvidia 8800gt, for example, to provide boosted physics performance?

what about amd/ intel and havok? will havok supported games work with physx supported games and vice versa, or exclusively?

what role will a future directx (11?) play in the future regarding physics in games?

will there be a faster progression towards gameplay physics instead of just the run-of-the-mill effects physics that we see in most games nowadays?

just a couple questions i'd thought i'd throw out there to see if anyone had a clue as to what the future holds for next generation gaming.

crystal ball gazers, come out from your caves, and let me know

 

[pxc]

My crystal ball says we'll know more about the state of legacy hardware PhysX support on GPUs next month and won't know about the future until game manufacturers announce support.

 

[SuperGee]

A old PC component out of 5-2006. the PPU is old. So it stands in retirement status.
PPU had a produckt cycle of 2 years. Thats over now.
Around this time there would be a need of a nextgen PPU.
Like 300 to 500 mil tranistors on 65nm. P2.
PPU P1 more comparable with a T-bred or nV40.

But as specialised produckt it still beats CPU.

Now after 2 years my PPU could be replaced. But as currently PhysX games optional support PPU P1 there is no need.
If there comes a new nV PhysX by TWIMTPB pushed and harest Dev's game.
And uses up to a full G92B or GT200 as dedicated PhysX accelerator. And models below. then it's time to replace. By something that at least is decent bit vaster then PPU P1.

So wenn a Good nV Physx game uses a 200M tot 1 Teraflop of PhysX in a right way. In a game I like.
It's time to replace my old PPU. They had there duty cycle.

The budged SM4.0 series arent much of a step up. But the high-end are.

So lets wait agian for the PhysX gen 2 killer game(s).
I see GPGPU nV PhysX replacing PPU gen 2.

 

[GotNoRice]

A old PC component out of 5-2006. the PPU is old. So it stands in retirement status.
PPU had a produckt cycle of 2 years. Thats over now.

I really find that to be a poor argument, regardless of how many times you post it here. Nvidia 8 series is 2 years old now, quad core is 2 years old. Tech such as the X-Fi have been around for even longer, yet all of this is still current tech. It is absolutely absurd to call the PPU old in that context. For what it does there is no president for how long it could remain a viable piece of technology. "Retirement status" after 2 years

 

[SuperGee]

I really find that to be a poor argument, regardless of how many times you post it here. Nvidia 8 series is 2 years old now, quad core is 2 years old. Tech such as the X-Fi have been around for even longer, yet all of this is still current tech. It is absolutely absurd to call the PPU old in that context. For what it does there is no president for how long it could remain a viable piece of technology. "Retirement status" after 2 years

Poor? totaly not.
After first G80 the g80U came along. Then G92 and all that can in SLI and G92X2 and on top of it GT200.
most G80 are long EOL replaced by G92. PPU not, just recently.
CPU are updated with new models each quarter end older once go also EOL
PPU is sold for real for 2 years long.

You mention some computation needing chips made on a more modern Dieshrink procede out of 2006 and up. Wich may draw a lot more power to to put yet agian a lot Flops in it. CPU GPU that is. CPU that was the competition and GPU was promised to become its competition.
130nm is much older. It was already dated in 2006. But as a small company there wasn't much choice. The where a small customer to TMSC or the Chipfoundry they went to.
130nm is Tbred-Northwood tulatin nV40 etc. But then a 125mil flavor.

And Physics use is just as computingpower needing upschaling feature as 3D rendering and CPU computing needs.
Xi-fi works on sound and that doesn't scale much further then HD sound flafors. So for that there is a ceiling. just like IO2 controlers and Nic chips. Wich often get intergrated.
So a PPU you can put in line with CPU and GPU. Where the computing innovation is push to the limits. Wich exclude sound chips. And PPU is a first smal step as being the first of it's kind.

For competing agianst CPU a 2 year dutycycle suffice. And a old Dieshrink would do. Because it's specialist archteckture and dedicated use it made up for it largly. The 125mil 130nm is light chip design powerfull enough to beat cPU from 2006 till now. Quadcore that is.
But the promise from GPU vendors to storm the Physics market to, promised around it's release. It turned out Ageia would compete also agianst GpGPU somewhere in the future. Wich need a more powerfull PPU Design made on a more modern procede. But because GPGPU kept Away till now it would be the time. And ageia didn't anticipate this move in the GPGPU market.

But PPU is killed by nV. That no flop, that hostile take over. nV true reason is the SDK to replace the also true killed HavokFX in the 1st degree by iNtel.
Might be picked up agian by AMD.

Well I realy wished for 80nm PPU a year ago. But a 65nm PPU2 now with a new design would be nice. Instead it's replaced by a full SM4.0 GPU lines. So I have to pick something wich is a lot faster then the old light weight PPU.

For repalcement of the old PPU 125mil on 130nm it's counter part in transistor count would be the shader array of a GPU. DIE pictures of GT200/RV770 looks like a large part is filled in with shaders so a good step forward would be 300 to 500mil of shaders a G92B offer that and then you need that amound for 2D rendering. Possible a lot more PhysX computing is reachable so more PhysX in PhysX games in the future.
A GTX260/280 wil do. Enough for both task on a decent level. To push PhysX games further.

But to know, what is realy a lot better like a decent step forward. They all must be benched in a big review. with real games.
CPU PPU and the whole line of SM4.0 GPU's

G86
Chip - Transistors 210 mln // shader part +/- 100mil
Chip - Productieprocédé 80 nm // 80nm vs 130nm so higher clock
Comes close to the old PPU.


A GTX260 would be a good replacement for my 8600GT and PPU. For bencin I could leave them in. On a ATI CF chipset.

 

[GotNoRice]

Poor? totaly not.
After first G80 the g80U came along. Then G92 and all that can in SLI and G92X2 and on top of it GT200.
most G80 are long EOL replaced by G92. PPU not, just recently.
CPU are updated with new models each quarter end older once go also EOL
PPU is sold for real for 2 years long.

My point wasn’t that they haven’t come out with anything newer than an 8800 GTX in the last 2 years, but rather that I know plenty of people who bought one then and still use one and consider it far from “retirement status” worthy yet you are ready to apply that label to the PPU using no directly comparable criteria other than the passage of time.

130nm is much older. It was already dated in 2006. But as a small company there wasn't much choice. The where a small customer to TMSC or the Chipfoundry they went to.
130nm is Tbred-Northwood tulatin nV40 etc. But then a 125mil flavor.

You’re comparing completely different architectures and assuming you can draw relevant conclusions using these details. The PPU showed just how powerful it was vs the CPU and the truth is that we won’t know just how well it performs vs. CUDA emulation until we see some real benchmarks. I’m not saying it’s going to be great or even as good or better than Cuda emulation of PhysX but until we see some REAL Benchmarks or some REAL numbers all this speculation being thrown around as if you actually have a clue is pretty pointless…

For competing agianst CPU a 2 year dutycycle suffice. And a old Dieshrink would do. Because it's specialist archteckture and dedicated use it made up for it largly. The 125mil 130nm is light chip design powerfull enough to beat cPU from 2006 till now. Quadcore that is.

Uh, because you said so? If you have some benchmarks or any sort of real data that you are drawing these conclusions from please feel free to share it with us. Otherwise there is nothing special about the here and now which makes the PPU any more or less obsolete than it was a few months ago or a few months from now.

But to know, what is realy a lot better like a decent step forward. They all must be benched in a big review. with real games.
CPU PPU and the whole line of SM4.0 GPU's

There’s something we agree on, though that doesn’t seem to have stopped you from drawing many conclusions already. Clearly GPU physics are the way of the future but personally I have a lot of the same questions as the OP has and I’d like to see them answered at some point. Who knows what sort of interesting niche configurations may be possible. I continued to run a pair of Voodoo2’s in SLI for years even when I had upgraded to later cards like the Geforce3 because I preferred to run the original UT using Glide. PPU’s may be an interesting alternative for ATI/AMD card users who want PhysX if nvidia decides not to open it up.

 

[SuperGee]

You find it a poor argument and comes with G80's unexpected long duty cycle?
My point is about PPU retirement status.
That status has nothing to do with atom, itanium or my CPU in my ECU of my car. Or sound chips. Or intel stuf. etc etc etc. It's about PPU duty cycle.
Dutycycles are uniek and adoptive to there market and situation it is in, like competition.
G80 is more of a exception for GPU's. But that has nothing to do with PPU duty cycle.

Ageia planned for a duty cycle of 2 years.
That was Ageia choice. I think its was to long. After the soon announced competition of GPU.
Good to keep up with CPU. But for GPU to short for there first PPU.

There are 80nm PPU for mobile use. So Ageia had there G92(B) flavor to. But use for mobile.
Like I said it's a light design on to old Proces for that time. 125Mil on 130nm could have been 250mil on 90nm. But of course a new vendor to chipfoundery's don't get the luxory of latest Procedes.
Now PPU2 had not only to extend on CPU but also compete with GPU.
So the duty cycle of the first PPU was 2 years. and it was it's last.
As it's now nv GPU's replacing PPU2 it takes there duty cycles.
And doesn't take the add-on rout anymore.

 

[GotNoRice]

You find it a poor argument and comes with G80's unexpected long duty cycle?
My point is about PPU retirement status.
That status has nothing to do with atom, itanium or my CPU in my ECU of my car. Or sound chips. Or intel stuf. etc etc etc. It's about PPU duty cycle.
Dutycycles are uniek and adoptive to there market and situation it is in, like competition.
G80 is more of a exception for GPU's. But that has nothing to do with PPU duty cycle.

Ageia planned for a duty cycle of 2 years.
That was Ageia choice. I think its was to long. After the soon announced competition of GPU.

That’s obviously not how it played out, so some reference to a 2 year duty cycle back in 2006 doesn’t really mean anything in the current context.

Good to keep up with CPU. But for GPU to short for there first PPU.

This is based on what data?

Like I said it's a light design on to old Proces for that time. 125Mil on 130nm could have been 250mil on 90nm. But of course a new vendor to chipfoundery's don't get the luxory of latest Procedes.
Now PPU2 had not only to extend on CPU but also compete with GPU.
So the duty cycle of the first PPU was 2 years. and it was it's last.
As it's now nv GPU's replacing PPU2 it takes there duty cycles.
And doesn't take the add-on rout anymore.

Yes, obviously GPU physics are the way of the future, but that says NOTHING as to whether our current cards will continue to be functional or worthwhile in at least some configurations or how the performance will compare against cuda emulation on the GPU. If you hadn’t noticed, that was one of the main topics of this thread…

There are so many questions still left to be answered.

If the PPU is an old product ready for “retirement status”, why is it that we have just now within the last month or so FINALLY began to see PCI-E PPU’s hit the market in force?
http://www.hardforum.com/showthread.php?t=1311432

 

[Uberbob102000]

I don't think that a new PPU with more power is necessary considering that the current PPU is barely getting used.

 

[kumquatsrus]

I don't think that a new PPU with more power is necessary considering that the current PPU is barely getting used.

true, which comes back to my question of how the cuda enabled gpus will perform with physx against a ppu; and if they could be used in conjunction with a ppu to boost performance even more in physx enabled games. that's what i'm most curious about.

 

[SuperGee]

CPU had 2 years freedom to catch up with PPU. Now it has to catch up with GPU's.
Will do with larrabe.