Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,602
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
SemiAccurate wrong about Nvidia 480GTX power use
- Thread starter argor
- Start date
Right, much like all teh fanATIcs around here fall over themselves for crap for physics. Its not PR, try reading the bloody interview. He gets asked very point questions and he answers them very pointly, he doesn't dance around them like a PR drown.
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,602
bc2 doesn't have crap physx, Nor Does Crysis.
Neither one of them use Nvidia Physx...
P.S. Forgot about Ghostbusters too
Last edited:
Actually, BC2 does have crap for physics. its the same, none realitics looking crap everytime you do something and never changes, no matter what you use. Its preprogrammed. Crysis is the exception to this rule because they used in house coding and sadly, I'd liek to have seen it used in more games but noone would license it from CryTek. and FYI: Crysis physics is 10000000 times more realistic than BC2s will ever hope to be. Hell, if they adjust a few things to their in house physics code for the terrian aspect, it'd be better than PhysX by a long 1000 times, right now as is, its about 100 times better.
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,602
And Developers dont preprogram Physx for Nvidia games they support?...LOL come on man.
You make more excuses then George Bush did on the war on Iraq.
I am not 1 that is saying Fermi is bad, Real reviews will come out, But man you put down ATI anyway you can.
How did you feel about those Nvidia drivers that killed video cards?....Is that ATI's fault as well?
how can you preprogramm reactionary physics whihc is what everygame that uses GPU based physx has, reactionary. this means that everytime you do something different to an object, ah get this, the object will behave accordingly unlike crap physics in BC2 where no matter what you use or from where you use, ITS THE SAME CRAP everytime you see it. And how am I putting down ATI? Have they not been touting Havok and OCL for GPU based physics for the last 3+ years? Why yes george, they have. And what have they done in those 3 years you might ask george, 1 fucking demo of water. In that time, how many demos and games have been released show casing GPU based PhysX? A far good deal more than 1.
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,602
Well they dont have the physx's programmed in the Game to then have Nvidia come along and say (hey make your physx run better with an nvidia logo!!) No they program in the Nvidia physx into the game.
There is a reason you can run a game without Nvidia Physx, and run with it turned on. ITS Programmed into the game.
They program in to detect nvidia physx so IT can run the game with Nvidia physx and ADD the extra effects they programmed into the game for Nvidia.
Case in point. Batman AA. They programmed into the game to disable HIGH END textures to ATI customers because they cannot run Phsyx. But a neat lil hack works around this and makes it work
Again they PROGRAMMED it in.....DO NOT tell me they dont program Nvidia physx into a game../facepalm
Well they dont have the physx's programmed in the Game to then have Nvidia come along and say (hey make your physx run better with an nvidia logo!!) No they program in the Nvidia physx into the game.
There is a reason you can run a game without Nvidia Physx, and run with it turned on. ITS Programmed into the game.
They program in to detect nvidia physx so IT can run the game with Nvidia physx and ADD the extra effects they programmed into the game for Nvidia.
Case in point. Batman AA. They programmed into the game to disable HIGH END textures to ATI customers because they cannot run Phsyx. But a neat lil hack works around this and makes it work
Again they PROGRAMMED it in.....DO NOT tell me they dont program Nvidia physx into a game../facepalm
I've read about that hack and its reduced physics that doesn't kill performance. And if I recall Kyle's(or was it bruces') review of B:AA performance review, they said outside of PhysX, the game played and looked the same the major difference being the PhysX enabled fog and objects the REACT according to how you INTERACT with them.
Brackle
Old Timer
- Joined
- Jun 19, 2003
- Messages
- 8,602
I wouldn't exactly call REDUCED being the same as if you had a GPU or physics card in the machine while playing the game.
Anotehr thing you keep missing about how the reactionary PhysX works. Run thru the fog in B:AA with papers everywhere a dozen times the same way. a dozen times you'll get different reactions. BC2, blow up a building 12 times with a tank, blow up 12 times with a rocket launcher, then a granade launcher. every single time its the same, it doesn't change, there is no variance in it, its the same.
This. I'm waiting to see CPU benchmarks of Metro vs. GPU PhysX. I was actually thinking of using my good `ole 9800GTX for some fun comparisons.
The PhysX SDK does allow multithreaded CPU code, but it's fairly difficult to use efficiently/correctly compared to implementation on the GPU.
I encourage people to look at the documentation themselves:
http://developer.nvidia.com/object/physx_downloads.html
edit: Woops, looks like the API documentation is only available w/ the SDK itself. But hey, you can take my word for it =)
@XMAN:
You're very quick to call developers lazy, but you can't seem to put together one damn post that doesn't make my eyes bleed. Please don't quote this, I would hate to see my post followed by your innane mouth seizure "argument".
Last edited:
siliconnerd
[H]ard|Gawd
- Joined
- Jan 7, 2002
- Messages
- 1,768
Looking at your post history, no.
Looking at your post history, no.
What does the Havok/OCL and ATI and the fact they have accomplished nothing but a demo of GPU based physics have to do with my post history?
What about a dual GPU card(GTX495)?
DontBeEvil
Weaksauce
- Joined
- Mar 6, 2010
- Messages
- 110
What I can conclude from this thread.
XMAN245 is not a Nvidia employee. He knows less than I do and that's quite astonishing.
XMAN245 is not a Nvidia employee. He knows less than I do and that's quite astonishing.
With very few days remaining I have been with holding judgement.Only until then will we really know how this thing will react.There is no need for these conversations to get out of hand.I know there are those that like certain brands and loyallty is good,but not to this point.Just my opinon.Carry on.
Vermillion
Supreme [H]ardness
- Joined
- Apr 5, 2007
- Messages
- 4,423
The reason Havok or Bullet physics using the GPU isn't gaining ground is simple: GPU based physics aren't a game changer for anything as of yet. Eyefinity is a huge game changer and most likely NV surround will catch up to Eyefinity in the future. GPU physics just isn't that game changer yet.
Not once has PhysX made a game so amazing that it's a must have. PhysX in Batman: AA was a gimmick. Yes, it looks nice but it adds NOTHING to the game itself. I will always firmly believe that the PhysX effects would have been fine on the CPU if it hadn't been locked to a single core on a multi-core CPU in Batman especially after seeing the Infernal engine used in Ghostbusters.
Plus unless you're on a multi-GPU setup enabling GPU based physics is a performance hit. Multi-GPU setups are still a very niche market. I doubt Fermi will change that especially with it's rumored power requirements.
All evidence points to Fermi being under performing, overpowered, and overpriced. PhysX isn't going to save Fermi so why even bring it up? If we're lucky Fermi finally brings competition to AMD causing price drops. If not well...then as consumers we get screwed.
Chicken and egg.
Also, the physics engine in the Ghostbusters game was hugely simplistic. Havok is a lot more impressive if we're talking CPU-only physics engines.
Is that belief based on some technical grasp of the workload or is it just a warm fuzzy feeling?
Less than a dozen FPUs versus hundreds of them on a GPU. Definitely a fuzzy feeling. And fuzzy math too, probably
CrimandEvil
Dick with a heart of gold
- Joined
- Oct 22, 2003
- Messages
- 19,670
Less than a dozen FPUs versus hundreds of them on a GPU. Definitely a fuzzy feeling. And fuzzy math too, probably
I don't know, I remember the Alan Wake PC tech demo thing that was being shown off a while back. The one where the guy was showing the small town, sweeping back and forth across the landscape and then a huge tornado came in and destroyed it all, debris flying everywhere.
All the physics was done on one of the four cores of the Quad they were using. They specifically said that for the advanced physics to work it's best you would need a Quad core since one whole core would go to physics processing.
Yeah, a lot more can be done on a GPU but, seriously, how many games use enough physics effects that require a whole GPU? I'm pretty sure that any physics enabled game out right now, and probably for the next few years, could do all it's physics on an entire CPU core just as well.
necrophile
[H]ard|Gawd
- Joined
- Sep 6, 2008
- Messages
- 1,243
I have never seen anything impressive in game done by a gpu that was not smoke and mirrors. It is like having 10 men carry a plank of wood and telling everyone it is very heavy, yet not showing the one man doing the same thing.
Vermillion
Supreme [H]ardness
- Joined
- Apr 5, 2007
- Messages
- 4,423
None of us here have seen the code so technically none of us can say if it can or cannot run on a CPU.
But yes it is a warm fuzzy feeling based on the hack and what CPU physics have been shown to do. Nobody has ever proven that the Batman PhysX couldn't be done on the CPU correctly. Everybody just assumes it can't.
What I can conclude from this thread.
XMAN245 is not a Nvidia employee. He knows less than I do and that's quite astonishing.
If I had to guess, I'd say he's a 'former' nvidia employee
MrLonghair
[H]ard|Gawd
- Joined
- Jun 7, 2004
- Messages
- 1,764
You should see his posts on beyond3d then, DontBeEvil. You'll crack a rib
Too bad you don't know what the MSRP is, and you're just a fanboy with nVidia blinders on.
BTW, I'm an economist. How about you stay to the domains where you excel at, like being unreasonable.
A
akotlar
Guest
Here's a small benchmark using an 8600GT as the physics co-processor. http://www.revioo.com/articles/a13159_3.html
8600GT has ~114Gflops when utilizing the MUL. I don't have any information on whether or not it is being utilized with PhysX. If it isn't then it has less theoretical single precision performance than does a 3.2ghz Nehalem (2.66Ghz Nehalem can reach 75Gflops/s SP in benchmarks, 2/3 of a 8600GT is 76Gflops theoretically). I also doubt utilization is as high on a G9x gpu as it is on Nehalem (or C2Q). From what I've read physics runs less efficiently on a G92-GT2xx class GPU than it does on a CPU, but I can't really corroborate that at this time. Maybe its not a big enough efficiency difference to talk about. Do you know what kind of collision load a Nehalem can handle vs a similar Gflop spec GPU?
It seems quite reasonable that all of the effects we see in Batman could be run on an reasonably clocked Nehalem at speeds exceeding that of a 8600GT, which performs reasonably well as a physics co-processor. I'd also assume that a CPU-centric programming model would discover efficiencies, or "good enough" approximations, for example maybe less accurate but more efficiently processed collision shapes, that would allow comparable performance at some minor visual tradeoff compared to something like a 9600GT, which gives considerably better results. Personally I ran Batman on a GTX 285 with physics on high (this bench shows that GTX 280 runs about 20% slower than 280 + 8600GT, about 41 vs 52 fps), and that level of performance would be acceptable. Nehalem would probably deliver beyond the 8600GT results. Never mind that there is room for discussion about whether the PhysX effects are done in the most efficient way possible.
Do you know what the Batman physics workload was in Gflops/s? Could a good CPU physics package handle it? You have a superior tone, both here and B3D, where's it coming from? But it's cute so go ahead and keep it
edit: http://74.200.65.90/showthread.php?p=1179086 I'm not sure why this is, but seems MUL utilization on G80 & G92 (so presumably G84 as well, or was that in any way an improvement on either in terms of MUL utilization?) is no more than 50% of theoretical. So that puts a G8600GT right below a 3.2Ghz Nehalem (~ 85.55 vs ~90Gflops, using 50% MUL utilization and 1:1 scaling from 2.66Ghz resp. ) From what I've seen in Nvidia's CUDA documents however, it is possible to utilize MUL with no penalty. Not sure what's up. Regardless performance will certainly be close to a 8600GT.
Last edited by a moderator:
Manufactor
Suggested
Retail
Price
smart ass. You wanna cookie for working to help screw up our economy. excellent job there Mr. Economist.
I'm just curious, but where are you getting your GFlops info?
I ask because of the following for a single nehakem based cpu, scroll down. But from googling it appears your numbers are based on Nodes and not single CPUs.
http://www.maxxpi.net/pages/result-browser/top10---flops.php
A
akotlar
Guest
Sure http://vuduc.org/pubs/arora2009-nbody.pdf I think you're confused about what node means.
here's another source: http://www.hpcwire.com/blogs/Benchmark-Challenge-Nehalem-Versus-Istanbul-48551797.html
~ 170 Gflops achievable peak, meaning 85 per cpu, but it doesn't state frequency. If you browse around you'll find 75 Gflops linpack performance for a single 8 thread Bloomfield @ 2.66ghz. Assuming linear scaling that's about 90Gflops for a 3.2Ghz Bloomfield. Those same sources paint Bloomfield as ~85% efficient so 90Gflops/.85 gets you Bloomfields theoretical SP performance. 106 Gflops. It's most likely more efficient than Geforce 8,9 & 200 in GPGPU tasks. That seems to ring true for other GPGPU tasks, and I'm assuming that physics isn't somehow exempt. Assuming everything up to GT200 doesn't reach more than ~50% utilization of that MUL on average in GPGPU tasks (coming from B3D, previous post, not sure if this is completely accurate), Nehalem is actually quite a bit more powerful than 8600GT, and about half the performance of a 9600GT, but we'll get to that.
As for this dicussion. Well I think it can be best summarized with this:
http://developer.download.nvidia.com/presentations/2008/GDC/GDC08_ParticleFluids.pdf (Nvidia Performance figures)
http://www.youtube.com/watch?v=fet-TrNssjQ (Nehalem physics performance)
The Nehalem "smoke" demo, runs about 60k particles at ~17-19fps. I'm assuming they're simple collision particles, and aren't using any sort of fluid simulation (Navier Stokes, SPH), which seems perfectly reasonable.
This Nvidia doc http://vuduc.org/pubs/arora2009-nbody.pdf points to an 8800GT (which is ~98% Gflops performance of a 8800GTX) being able to handle 65K particles simulating simple collision at ~100fps (they say > 100, which probably means 101 lol)
So going by that figure we'll assume ~ 5.5x speedup from an efficiently loaded (~75% avg) Nehalem (ghz?) (they state 75% average utilization in the video). This would put a Nehalem at ~19% the performance of an 8800GT, whereas a 8600GT posts at about 22% (not sure this is the case. Considering how castrated the G84 is, it wouldn't be very surprising that it does not reach similar Gflops/theoretical Gflops efficiency). So going by these figures, if an 8600GT is the limiting factor and provides ~ 52fps (previous post has these numbers) in Batman (with a GTX 280 as the GPU), Nehalem would give you somewhere on the order of 45 FPS, still above what a single GTX 280 can render on its own. Obviously with more intense GPU loads and similar physics load, this would be more in favor of Nehalem.
Going by the Nvidia document, we also see that the 8800GTX running N-body simulation reaches 260Gflops (probably peak). Assuming that is an ideally optimized example (I'm not a programmer, I don't know that this is the case), and assuming a 3.2GHz nehalem reaches 75% utilization (from the video, and taking into account a 90Gflops peak linpack performance as an indication of peak n-body performance), we see a <4x speedup (~69Gflops vs 260) by going to a 8800GTX over an ideally loaded 3.2 Nehalem.
In this case Nehalem reaches ~25% of the 8800GTX vs 22% for the 8600GT . Therefore Nehalem should give you roughly 12.5% PhysX performance improvement over 8600GT. If Batman is 100% bottlenecked by the 8600GT with PhysX on high, maybe you'll see ~58fps for a Nehalem that is utilized 75% of the time for PhysX. Considering how low CPU usage is in Arkham Asylum running on Bloomfield, 75% utilization for physics seems to leave enough room for the CPU to calculate everything else (basic collision tests, a.i, whatever normal CPU physics is running).
Batman CPU utilization: http://www.hardocp.com/article/2009/10/19/batman_arkham_asylum_physx_gameplay_review/11 (37% average... on a QX9770 at 3.6ghz...will certainly be no more than 50% of that figure on Nehalem, accounting for 2x threads)
Of course this is very theoretical, and I don't have the expertise to evaluate how accurate this supposition is, but trinibwoy, that's the basis of my reasoning.
Last edited by a moderator:
Zellio2010
Limp Gawd
- Joined
- Dec 26, 2009
- Messages
- 460
What I can conclude from this thread.
XMAN245 is not a Nvidia employee. He knows less than I do and that's quite astonishing.
He could be the guy who programmed 196.75....
Zellio2010
Limp Gawd
- Joined
- Dec 26, 2009
- Messages
- 460
I'm not a physics programming expert, but I'm pretty sure that ALL physics is proprogrammed using math routines (after all, physics is math), You can't create mathmatics in a game or anywhere in thin air, it has to be programmed. My god you are dumb.
Zellio2010
Limp Gawd
- Joined
- Dec 26, 2009
- Messages
- 460
I love the broad brush strokes you give people kid. No wonder you act unintelligent.
What he's talking about is pre-scripted instead of generated by routine. A scripted interaction would involve the model collapsing/blowing up the same way every time because it follows a script. When it's generated by a routine the equation is solved with impact velocity to determine where things are gonna be flying, so things rarely end up the same.. But the physics aren't scripted in BF2. He either hasn't played it or is purposely conflating separate issues.
The destructed environment is predetermined - certain walls are non-destructible and will always look the same after the destructible portion has been blown up. But the physics - which is only concerned with the debris from the destructable portions - is real, on-the-fly physics.
Zellio2010
Limp Gawd
- Joined
- Dec 26, 2009
- Messages
- 460
That's not physics, that would be scripted events. Again, physics cannot be made out of thin air. And any physics platform would have the ability for real physics and not scripts if you took the time to program math.
You tell me?
http://www.rchoetzlein.com/eng/graphics/sph_paper.pdf
And the tone? Well that's just part of my charm. The ladies can't resist.
Uhm, ok. By 'script' he and I mean, predetermined movement with no calculations happening involving the interaction of the objects. By 'physics' we mean, no predetermined movement and using calculations to determine the inertia of the objects and their interaction to emulate real world physics.
If you want to be abstract, you can claim all physics is scripted because the math can be predetermined, or is run through a routine that you could call a script, but that's not the way we're using the words.