Ageia PPU...maybe not so great after all.

[bunnyfubbles]

quit complaining, its going to happen.

 

[the Chase]

One of the more interesting things to do is watch my girlfriends 6 year old play some of these games. (The less graphic and vulgar ones). He really got a kick out of the GRAW demo in that you could see your leg(s) and feet and could actually "dribble" an empty can that was lying around with your feet soccer style. He never could understand why you couldn't eat one of the bananas in the fruit stands in the desert combat mod of BF1942 lol. Looking forward to more, however it happens...

 

[Drayvn]

As a few others have said here. Ok so the guy had to lower his res to play the game with 1000's of objects going round because the GPU couldnt handle rendering all those things.

What do you guys think the performance would have been like just using the CPU without the help of the PPU. Probably about 100x worse!

So in fact what this post is saying is its a graphics decelerator, well it isnt because for this type of game its actually helping to keep the FPS up because if the CPU did it all, you probably wouldnt have even a frame per second.

 

[akugami]

For various reasons, I've always been of the opinion that Aegia will fail. Not that I don't believe their product will truly help bring in the next level of gaming. Just that by the time games start being coded with advanced physics in mind, both nVidia and ATI will have some form of physics acceleration built into their GPU's. The initial batch of games with physics acceleration seems about par for the course. More pretty effects, maybe a few extra damage in the terrain but nothing ground breaking and nothing that will make or break a game.

Advanced physics as it stands today just seems like a more natural extension of a modern GPU. It took Aegia to spur the advancement of the next level of physics processing but unfortunately it looks like they will go the way of 3dfx.

 

[otispunkmeyer]

having seen cell force (video) i would play at 640x480 just to go mental with the physics.....looks like overwhelming fun that game

 

[Dethfrumbelo]

Originally posted by: akugami
For various reasons, I've always been of the opinion that Aegia will fail. Not that I don't believe their product will truly help bring in the next level of gaming. Just that by the time games start being coded with advanced physics in mind, both nVidia and ATI will have some form of physics acceleration built into their GPU's. The initial batch of games with physics acceleration seems about par for the course. More pretty effects, maybe a few extra damage in the terrain but nothing ground breaking and nothing that will make or break a game.

Advanced physics as it stands today just seems like a more natural extension of a modern GPU. It took Aegia to spur the advancement of the next level of physics processing but unfortunately it looks like they will go the way of 3dfx.

Exactly. Since the GPU is the bottleneck, Nvidia and ATI will have the final say, for better or worse. By the time GPU rendering power is able to cope with this level of objects (I doubt G80/R600 will even be able to do it, probably the generation after that), they will almost surely have a PPU integrated on their cards.

 

[otispunkmeyer]

Originally posted by: The Dome
Here you go, take the time to read the whole thing. LINK

that dudes got a yonah @ 3100mhz and a x1900xtx.....that thing is gonna fly PPU or not lol

 

[otispunkmeyer]

Originally posted by: DeathReborn

Originally posted by: ElFenix

Originally posted by: the Chase
I like munky's thoughts on this. How much phyx do you need in a game to make it fun/immersive? 10,000 flying barrels is neat and maybe fun to mess around with for 10 mins. but hardly realistic and will choke any graphics card out there. There are a lot of phyx effects that are more subtle (water movement, the way cars, boats, etc. move in relation to their enviroment, etc.) that won't take so much graphics power to make work. But is this using the PPU to it's full extent and/or could a dual core proc. handle it alone? Dunno.

I do feel though that this puts a serious crimp in the notion of having GPU's calculate phyx. If they can barely handle Oblivion now, and are totally choked with 10,000 barrels that a PPU creates, how are they going to calculate the phyx of the 10,000 barrels (or even 500 barrels) and render them at the same time?

But yeah buying a $300 card so I have to go buy $1000 worth of GPU power to play at decent framerates doesn't sound to appealing....

Edited for spelling.....How many l's are in barrels?? Barrels or barrells?

i mean, no one would ever want MORE realism in their games? after all, no one has ever switched a game from 'arcade mode' to 'sim mode'

I could see a PPU being used in games like Armed Assault, America's Army and other "aiming for reality" games like that. Not to the extent of CellFactor but at least creating a realistic environment for some realistic DM's, lol.

i know, cell factor looks great but its obviously a show case game, its well over the top. id imagine proper integration of the physics now capable will be more subtle in coming games

just things like proper fluid modelling, chassis dynamics, explosions, object interactions etc...all can probably be done on the cpu, but the ppu will speed it up further, and be more in depth at the same time, without the need for a million objects floating round

cell factor may present a bit of a hurdle for GPU's now, but id imagine more serious games like GRAW and race sims and stuff wont be a problem

 

[Gstanfor]

I always felt that physics would result in a much larger load being dumped on the graphics card. If it's this bad with a dedicated PPU doing the physics processing imagine how much worse the Havok systems running on the GPu are going to end up performing.

Anyhow, to me, this just further exposes the current archilles heel of 3D graphics at the moment in my opinion - lack of vertex/geometrt processing power. Physics is all about manipulating objects in space, vertex shaders set those objects up in 3d space, pixel shaders/texture units merely render the surfaces. For far too long now the 3D harware vendors have favored pixel processing power over vertex processing power (because it is easy for people to see the benefit of pixel processing power and much harder to see the benefits of vertex processing power). If we want to truly move forwards then vertex processing power is going to have to significantly increase. This is a major reason why IMO microsoft unified pixel/vertex shaders in DX10 and introduced the geometry shader.

 

[Jules]

Originally posted by: Gstanfor
I always felt that physics would result in a much larger load being dumped on the graphics card. If it's this bad with a dedicated PPU doing the physics processing imagine how much worse the Havok systems running on the GPu are going to end up performing.

When you dont thave the PPU card it doesnt use those settings there for it wont perform slower.

 

[Gstanfor]

Originally posted by: MyStupidMouth

Originally posted by: Gstanfor
I always felt that physics would result in a much larger load being dumped on the graphics card. If it's this bad with a dedicated PPU doing the physics processing imagine how much worse the Havok systems running on the GPu are going to end up performing.

When you dont thave the PPU card it doesnt use those settings there for it wont perform slower.

You obviously missed the discussion on NVIDIA's collaboration with HAVOK. ATi also has PPU on a GPU in the works.

 

[TanisHalfElven]

well duh. really what did you expect. this thread is really badly titled. i thought i physx card had some bug with cross fire.

i mean come on ppl ofcourse there'd be slow down. atleast now its due to sometihng fun rather than 10 shawdows vs 20 shadows or low-res textures vs high res. onceit comes out i personally will reduce everysingle shadow and other crap just to enjoy the awesome physics. most others will do the same.

 

[Ackmed]

Originally posted by: Dethfrumbelo
Even with top of the line hardware, games like Oblivion aren't running anywhere near 60 fps. More likely you'll be going from 30 fps to 2 fps. I can understand why Bethesda dropped most of the physics from the game. GPUs have a long, long way to go before they can match these kinds of demands.

2fps? Not likely.

I dont think they should have dropped some things. Even if systems today cant handle them very good, next years systems will. Farcry slapped around any system when it came out. A year later, or even today, you can get great frames with everything maxed. Oblivion will be played for years, likely have some add ons, and it would be nice to be able to crank up the details then.

 

[Extelleron]

Forgive me if there's more info elsewhere, but how do you know his "twin graphics setup", isnt something like 6800GT SLI, or 6800U SLI? Just because it's "twin graphics" doesnt mean its 7900GTX SLI or X1900XTX CF.

 

[the Chase]

Originally posted by: Extelleron
Forgive me if there's more info elsewhere, but how do you know his "twin graphics setup", isnt something like 6800GT SLI, or 6800U SLI? Just because it's "twin graphics" doesnt mean its 7900GTX SLI or X1900XTX CF.

Also on that note - what did he have to reduce the settings to? If it was just 1880x1600(or whatever) to 1600x1200 that's not really a huge deal. Now if it was 1600x1200 and he had to go to 640x480 that would suck....

Another thing that hopefully peeps could fill me in on. When we talk about integrating the PPU onto the graphics card(s), do we mean putting an additional chip on the PCB of the card that shares the memory with the GPU? Or do we mean using the "extra" vertex processing power of the card or extra created pipelines in the GPU core created specifically for phyx?

 

[RyanVM]

Originally posted by: akugami
For various reasons, I've always been of the opinion that Aegia will fail. Not that I don't believe their product will truly help bring in the next level of gaming. Just that by the time games start being coded with advanced physics in mind, both nVidia and ATI will have some form of physics acceleration built into their GPU's. The initial batch of games with physics acceleration seems about par for the course. More pretty effects, maybe a few extra damage in the terrain but nothing ground breaking and nothing that will make or break a game.

Advanced physics as it stands today just seems like a more natural extension of a modern GPU. It took Aegia to spur the advancement of the next level of physics processing but unfortunately it looks like they will go the way of 3dfx.

I don't see AGEIA failing as much as I see them being bought out by nVidia or ATi. They do make a solid product and one of them could certainly make use of the technology.

 

[Gstanfor]

Originally posted by: the Chase

Originally posted by: Extelleron
Forgive me if there's more info elsewhere, but how do you know his "twin graphics setup", isnt something like 6800GT SLI, or 6800U SLI? Just because it's "twin graphics" doesnt mean its 7900GTX SLI or X1900XTX CF.

Also on that note - what did he have to reduce the settings to? If it was just 1880x1600(or whatever) to 1600x1200 that's not really a huge deal. Now if it was 1600x1200 and he had to go to 640x480 that would suck....

Another thing that hopefully peeps could fill me in on. When we talk about integrating the PPU onto the graphics card(s), do we mean putting an additional chip on the PCB of the card that shares the memory with the GPU? Or do we mean using the "extra" vertex processing power of the card or extra created pipelines in the GPU core created specifically for phyx?

At the moment when we talk about on GPU physics, we mean utilization of the pixel shaders (with some usage of the vertex shaders also) to do the physics calculations. In current GPU's there are no "extra" quads or pipes for PPU operation, though that could possibly change in the future (highly unlikely IMO). The physics calculations take place in GPU idle time (easy enough for older games that don't make much use of pixel shaders, but with newer games there is less and less "idle time" in the pixel shaders to begin with).

Also unless there isa custom API at work then setup time to send the data to and from the GPU will end up being a throttle on performance (performing scene setup already holds GPU's back enormously - something microsoft is trying to address with DX10), not to mention restoring the GPU state to a usable condition for D3D/OGL.

 

[DeathReborn]

Originally posted by: otispunkmeyer

Originally posted by: DeathReborn

Originally posted by: ElFenix

Originally posted by: the Chase
I like munky's thoughts on this. How much phyx do you need in a game to make it fun/immersive? 10,000 flying barrels is neat and maybe fun to mess around with for 10 mins. but hardly realistic and will choke any graphics card out there. There are a lot of phyx effects that are more subtle (water movement, the way cars, boats, etc. move in relation to their enviroment, etc.) that won't take so much graphics power to make work. But is this using the PPU to it's full extent and/or could a dual core proc. handle it alone? Dunno.

I do feel though that this puts a serious crimp in the notion of having GPU's calculate phyx. If they can barely handle Oblivion now, and are totally choked with 10,000 barrels that a PPU creates, how are they going to calculate the phyx of the 10,000 barrels (or even 500 barrels) and render them at the same time?

But yeah buying a $300 card so I have to go buy $1000 worth of GPU power to play at decent framerates doesn't sound to appealing....

Edited for spelling.....How many l's are in barrels?? Barrels or barrells?

i mean, no one would ever want MORE realism in their games? after all, no one has ever switched a game from 'arcade mode' to 'sim mode'

I could see a PPU being used in games like Armed Assault, America's Army and other "aiming for reality" games like that. Not to the extent of CellFactor but at least creating a realistic environment for some realistic DM's, lol.

i know, cell factor looks great but its obviously a show case game, its well over the top. id imagine proper integration of the physics now capable will be more subtle in coming games

just things like proper fluid modelling, chassis dynamics, explosions, object interactions etc...all can probably be done on the cpu, but the ppu will speed it up further, and be more in depth at the same time, without the need for a million objects floating round

cell factor may present a bit of a hurdle for GPU's now, but id imagine more serious games like GRAW and race sims and stuff wont be a problem

CellFactor would make a great tech demo and quite possibly a great game too (albeit a slow one). I do like how in GRAW bullets do cosmetic damage better than in almost every other game but Dual Core (Opteron 170 @ 2.8Ghz) doesn't stop it being sluggish when large explosions happen (like the petrol Station, US Embassy etc) so the CPU does seem to be a bottleneck in GRAW. I know X1800XT CF isn't in the league of X1900 CF but even when I stick both in and oc the Opty it's choking at points.

I'm going to try and buy a PhysX PPU in May and see how it affects GRAW. I want to see the realistic games creating scenery with the properties it has in real life so in them you get a realistic experience (as real as a PC gets anyway). We got a long way to go and I supsect it'll be a good 5 or 6 years before we see something like that happening.

 

[Avalon]

I'd rather drop a notch in resolution than not get to have 500 things flying about me at once.

 

[PingSpike]

I'd always assumed this was how it would work. More things = more things that need to be rendered. And obviously the PPU isn't involved in the rendering process.

 

[dunno99]

Wow, obviously the person who wrote the original article or the presenter weren't paying attention back in school. In any comparison, the experiement should only vary one variable at a time (of course, there are lots of situations that this can't apply, because the system is too complicated...but this case isn't one of them). Which means the demo should've either compared a system with no PhysX card vs one with the card...and I'm sure the one without the PhysX card would run at about 1 frame per minute. Remember kids, the number of flying objects is also the variable here. The PhysX card isn't a graphics decelerator (the statement in the original Yahoo! article is akin to saying "OMG, my new 3.2L engine can't accelerate from 1 to 1000 in 5 seconds, whereas my 2.6L can accelerate from 1 to 100 in 4! My new engine is a decelerator! Wahh!"...and no, the 1000 or the 100 aren't typos), it's in fact an accelerator. The reason that they have to lower the resolution is because they have INCREASED the amount of objects on screen to demonstrate what the PhysX PPU can do, but because of the increased geometry, this slows down the graphics output.

I find it retarded that these types of Yahoo! articles even exist despite their total lack of intellectual insight (however much that goes into reviewing tech demos). Good thing at least half the Anand forums crowd knew bullsh*t when they smelled one.

 

[Extelleron]

Originally posted by: dunno99
Wow, obviously the person who wrote the original article or the presenter weren't paying attention back in school. In any comparison, the experiement should only vary one variable at a time (of course, there are lots of situations that this can't apply, because the system is too complicated...but this case isn't one of them). Which means the demo should've either compared a system with no PhysX card vs one with the card...and I'm sure the one without the PhysX card would run at about 1 frame per minute. Remember kids, the number of flying objects is also the variable here. The PhysX card isn't a graphics decelerator (the statement in the original Yahoo! article is akin to saying "OMG, my new 3.2L engine can't accelerate from 1 to 1000 in 5 seconds, whereas my 2.6L can accelerate from 1 to 100 in 4! My new engine is a decelerator! Wahh!"...and no, the 1000 or the 100 aren't typos), it's in fact an accelerator. The reason that they have to lower the resolution is because they have INCREASED the amount of objects on screen to demonstrate what the PhysX PPU can do, but because of the increased geometry, this slows down the graphics output.

I find it retarded that these types of Yahoo! articles even exist despite their total lack of intellectual insight (however much that goes into reviewing tech demos). Good thing at least half the Anand forums crowd knew bullsh*t when they smelled one.

:thumbsup:

 

[Munky]

Originally posted by: dunno99
Wow, obviously the person who wrote the original article or the presenter weren't paying attention back in school. In any comparison, the experiement should only vary one variable at a time (of course, there are lots of situations that this can't apply, because the system is too complicated...but this case isn't one of them). Which means the demo should've either compared a system with no PhysX card vs one with the card...and I'm sure the one without the PhysX card would run at about 1 frame per minute. Remember kids, the number of flying objects is also the variable here. The PhysX card isn't a graphics decelerator (the statement in the original Yahoo! article is akin to saying "OMG, my new 3.2L engine can't accelerate from 1 to 1000 in 5 seconds, whereas my 2.6L can accelerate from 1 to 100 in 4! My new engine is a decelerator! Wahh!"...and no, the 1000 or the 100 aren't typos), it's in fact an accelerator. The reason that they have to lower the resolution is because they have INCREASED the amount of objects on screen to demonstrate what the PhysX PPU can do, but because of the increased geometry, this slows down the graphics output.

I find it retarded that these types of Yahoo! articles even exist despite their total lack of intellectual insight (however much that goes into reviewing tech demos). Good thing at least half the Anand forums crowd knew bullsh*t when they smelled one.

Which is of course the whole point of the physx demo - to show off it's capabilities. But how those capabilities will be used in actual games is a different matter, especially when a) the point of the game is not to just show off the physics, and b) the rest of the system can't handle the additional physics load without slow downs.

 

[Drayvn]

Originally posted by: dunno99
Wow, obviously the person who wrote the original article or the presenter weren't paying attention back in school. In any comparison, the experiement should only vary one variable at a time (of course, there are lots of situations that this can't apply, because the system is too complicated...but this case isn't one of them). Which means the demo should've either compared a system with no PhysX card vs one with the card...and I'm sure the one without the PhysX card would run at about 1 frame per minute. Remember kids, the number of flying objects is also the variable here. The PhysX card isn't a graphics decelerator (the statement in the original Yahoo! article is akin to saying "OMG, my new 3.2L engine can't accelerate from 1 to 1000 in 5 seconds, whereas my 2.6L can accelerate from 1 to 100 in 4! My new engine is a decelerator! Wahh!"...and no, the 1000 or the 100 aren't typos), it's in fact an accelerator. The reason that they have to lower the resolution is because they have INCREASED the amount of objects on screen to demonstrate what the PhysX PPU can do, but because of the increased geometry, this slows down the graphics output.

I find it retarded that these types of Yahoo! articles even exist despite their total lack of intellectual insight (however much that goes into reviewing tech demos). Good thing at least half the Anand forums crowd knew bullsh*t when they smelled one.

Exactly.

 

[imported_JamesDax]

Originally posted by: dunno99
Wow, obviously the person who wrote the original article or the presenter weren't paying attention back in school. In any comparison, the experiement should only vary one variable at a time (of course, there are lots of situations that this can't apply, because the system is too complicated...but this case isn't one of them). Which means the demo should've either compared a system with no PhysX card vs one with the card...and I'm sure the one without the PhysX card would run at about 1 frame per minute. Remember kids, the number of flying objects is also the variable here. The PhysX card isn't a graphics decelerator (the statement in the original Yahoo! article is akin to saying "OMG, my new 3.2L engine can't accelerate from 1 to 1000 in 5 seconds, whereas my 2.6L can accelerate from 1 to 100 in 4! My new engine is a decelerator! Wahh!"...and no, the 1000 or the 100 aren't typos), it's in fact an accelerator. The reason that they have to lower the resolution is because they have INCREASED the amount of objects on screen to demonstrate what the PhysX PPU can do, but because of the increased geometry, this slows down the graphics output.

I find it retarded that these types of Yahoo! articles even exist despite their total lack of intellectual insight (however much that goes into reviewing tech demos). Good thing at least half the Anand forums crowd knew bullsh*t when they smelled one.

QFT