Maybe someone else has the energy to question you about these claims themselves, me I'm mainly just curious about the real reasoning behind why you and some others here dislike Physx so strongly.
In general, what's wrong with a little competition?
Amd Insight
- Thread starter Davros
- Start date
Maybe someone else has the energy to question you about these claims themselves, me I'm mainly just curious about the real reasoning behind why you and some others here dislike Physx so strongly.
In general, what's wrong with a little competition?
What about the CPU implementation of PhysX? From the cpu-utilization graphs I've seen (where it basicly looks single threaded) I find it hard to take it serious as an API.
They can obviously do it massively parallel for the gpu implementation, but haven't done it for the cpu implementation where especially quad core users have LOADS of unused cpu ressources (unlike the gpu ressources that are often close to 100% in use already.
They can obviously do it massively parallel for the gpu implementation, but haven't done it for the cpu implementation where especially quad core users have LOADS of unused cpu ressources (unlike the gpu ressources that are often close to 100% in use already.
Have they?
Historically Havok's revenue model was solely based on the quality of their software and the payments they got from the licensing of the middleware. Ageia found it difficult to gain traction in relation to Havok up until the point they started offering their entire software stack (not just the PPU accelerated elements) for free; even then most opted to continue paying for Havok.
With the purchase by Intel, Havok's survival is no longer as dependant on the need for that PC licensing revenue (along with the revenues brought in from console licensing) hence the relatively recent move freely license the core middleware on the PC.
Killer-Kris
Regular
...and the numbers make it look like Havok is losing support even faster. So what is replacing them?
A little Google-fu with Mr. Spink's name should reveal his employer, and possibly give you some insight into his views.
aaronspink
Veteran
Physx from its inception was designed as a proprietary library in order to sell the hardware. The amount of effort and money going in to optimizing it for its propriety hardware is much much more than for other hardware in general. The whole point is proprietary lock in. This goes down to their benchmarks which don't so much show real physics but just eye candy with little to no effect on actual game play.
Nothing has changed since the acquisition by Nvidia, its just shifted which proprietary hardware its meant to sell. The problem is that unless it is believed that the vendor is relatively neutral there is little chance that the game designers will actually add in meaningful physics into their games.
aaronspink
Veteran
I speak for myself thank you. I've always been down on physx as a middleware solution because its whole purpose is proprietary hardware lock in with little to no work making it fast on any other hardware.
in addition, most of what the physx libraries accelerate is just eye candy. I want real, interactive, part of the actual game physics.
The inception of PhysX was called NovodeX, a cross-platform multithreaded physics library (PC and various consoles), with no special hardware. It was acquired by Ageia and renamed to PhysX.
I could say the same about Havok. Currently it's CPU-only, no GPU or PPU-support available at all. And even if (a very big if) Radeon-support ever surfaces, that still leaves out nVidia GPUs and Ageia PPUs.
That depends on what the developer does with it. Nothing limits you from using a GPU or PPU to accelerate actual game physics (and in UT3, *all* physics are accelerated with GPU/PPU for example).
Havok on the other hand clearly states that they are not going to use the GPU for everything (read: eye-candy only).
So pot calling the kettle black, it seems.
The PPU is dead, we should start relegating it to a footnote in history at some point soon. And if they ever add GPU support to the core libraries, I'd imagine they won't tie it to one vendor. And if they did, why not the parent's GPUs?
You're probably right there. That would kinda seem to make business sense considering who owned it and owns it. Conversely, Intel has nothing to gain from optimising for anything but CPUs, and Havok becomes just another vehicle for selling quad core hardware.
So how is that really different? How is the x86 architecture somehow non-'proprietary', but a PCI or PCIE card - which will work on pretty much any platform - is?
Seeing as you keep mentioning this, what exact limitation in Physx restricts it to eye candy use rather than in game play?
There some interesting view of the history of both physics engines here...
Lets recap. This is a pretty bad history (it misses lots of things) and is also from my memory, but hopefully gives and idea of the complex state of both companies physics engines.
Havok Timeline
------------------
Originally a single threaded PC physics engine
First multi-threaded implementation (PS2 vector units) showed how much the original architecture didn't fit NUMA vector units well
2nd Generation PC physics engine. Multithreaded, scales well across SMP system (360 + 4 core PC)
2nd Generation multi core vector units (PS3 v1), still had issues fitting into small job based NUMA architecture. so redid pretty much from scratch.
3rd Generation multi core vector units (current PS3). Its well broken into small jobs that stream data in a massively parellel fashion.
Intel buy Havok
I guess (presume from press releases), that the PS3 core is being used as the basis for Larrabee Havok and AMD GPU Havok.
Ageia Timeline
-----------------
PPU implementation - 1st (largely closed) API
For commercial reasons needed a x86 fallback, so Ageia brought a small up and coming software physics engine (Novodex)
2nd CPU + PPU implementation. Took the novodox front end and translated to PPU where possible, some parts of Novodex were removed as they didn't fit the PPU model
Ageia CPU is made SMP friendly, reasonable ports to 4 core PC and 360
Ageia PS3 1st pass is a disaster
Ageia PS3 2nd pass is done by SCEA itself, based on the CPU backend not the PPU backend.
NVIDIA buy Ageia
NVidia port Ageia to CUDA, some parts done on the CPU. Whether they used to PPU, CPU or SPU backend as a base is AFAIK unknown...
Another interesting thing to note, is that Sony provide free licenses to both Havok and Ageia on PS3. Havok used to cost on all other platforms, Ageia have (mostly) been free on PC if you also supported the PPU. Havok is now free on PC as well.
Lets recap. This is a pretty bad history (it misses lots of things) and is also from my memory, but hopefully gives and idea of the complex state of both companies physics engines.
Havok Timeline
------------------
Originally a single threaded PC physics engine
First multi-threaded implementation (PS2 vector units) showed how much the original architecture didn't fit NUMA vector units well
2nd Generation PC physics engine. Multithreaded, scales well across SMP system (360 + 4 core PC)
2nd Generation multi core vector units (PS3 v1), still had issues fitting into small job based NUMA architecture. so redid pretty much from scratch.
3rd Generation multi core vector units (current PS3). Its well broken into small jobs that stream data in a massively parellel fashion.
Intel buy Havok
I guess (presume from press releases), that the PS3 core is being used as the basis for Larrabee Havok and AMD GPU Havok.
Ageia Timeline
-----------------
PPU implementation - 1st (largely closed) API
For commercial reasons needed a x86 fallback, so Ageia brought a small up and coming software physics engine (Novodex)
2nd CPU + PPU implementation. Took the novodox front end and translated to PPU where possible, some parts of Novodex were removed as they didn't fit the PPU model
Ageia CPU is made SMP friendly, reasonable ports to 4 core PC and 360
Ageia PS3 1st pass is a disaster
Ageia PS3 2nd pass is done by SCEA itself, based on the CPU backend not the PPU backend.
NVIDIA buy Ageia
NVidia port Ageia to CUDA, some parts done on the CPU. Whether they used to PPU, CPU or SPU backend as a base is AFAIK unknown...
Another interesting thing to note, is that Sony provide free licenses to both Havok and Ageia on PS3. Havok used to cost on all other platforms, Ageia have (mostly) been free on PC if you also supported the PPU. Havok is now free on PC as well.
Last edited by a moderator:
From what I recall, it was already multithreaded (unless 'SMP friendly' means something different?) when it was still called NovodeX.
Psycho said:
I can agree that Havok Physics is likely more efficient and better suited for multi threading & multi-core utilization on the CPU. PhysX actually supports multi-threading too, but the docs doesn't advise that. Safe for some aspects that are deemed thread safe. But The fact that there are more games under development that choose CPU/software only PhysX over Havok Physics anyway - shows that developers doesn't judge an engine only based on how many core of the CPUs it can utilize. There are other, no less important, factors to consider; features, price, documentations, tools, support, stability, speed, accuracy, tweakability, ease of use and integration to the rest of the system, etc.
Speccy said:
That may be true, I'm not sure. However, IMO, the argument for AMD/ATi's decision to support certain physics engine/technology in this discussion is not about which engines make more money/revenue for their maker. But rather; which engines being used more in games.
aaronspink said:
Not for those who understand that the numbers for past/released titles are based on titles that were released during the last 5 years or something like that. At least half of the titles under development should be ready for release around this christmas. The rest will follow. So the support for PhysX is actually growing. And one can argue based on those numbers that 80% of upcoming PC games that are using either engines are adopting PhysX.
DeanoC said:
Hmm...
Source.Dave Baumann said:
The point being is that Havok is more of a complete middleware toolset. The primary reason for Ageia's current share is certainly not for its PPU acceleration, its because the CPU core toolset was being offered free. That has since been nullified by Havok.
And the current "interactive world" showcase, Force Unleashed, which went off and implemented Pixelux's engine.
I think that is your own assumption. Which may apply for indies and small/limited budget developers. Serious developers who build games that matter, however, tend to opt for paid support and source code. And PhysX in this regard is still cost much less than Havok Physics, as ever.
And in cases like that nvidia is actually offering to port those engines to physx
Similar threads
- Locked
