Talk:Physics processing unit

VU0

Really? Is the part about the VU0 necessary? The VU0 was just a primitive form of SIMD instruction set, the low end MIPS CPU at the core of the Emotion engine wasn't even really designed with it in mind, Sony basically hacked something together by shoving two of the MIPS processors 64-bit FPU's together and creating some new instructions for processing packed SIMD instructions. If we're going to mention that though, why would we leave out the Xbox - as the Pentium 3 it came with had SSE3, which was purpose built and actually had superior SIMD capabilities to Sony's Frankenstein solution. I know that the concept of SIMD in processors may *sound* similar to what the PhysX processor is doing. But that's because the PhysX processor was pretty close to the concept of a modern GPU, which consists of like thousands of tiny cores that perform the same operation many many times. Usually floating point (GPU's have since added integer capabilities, but at the time they were all floating point).

The two concepts, that of a GPU/Physics processor and an SIMD instruction set, sound similar because they are both basically the same concept, SIMD. The GPU just takes the concept to the extreme, whereas even on modern AVX you can at best hope to operate on 8 different packed values at once, with the GPU your operating on thousands. The GPU cores are also usually a lot weaker individually. And AVX sits right next to the processor so of course it's ideal for mixing with general programming. Whereas it can be difficult for the GPU and CPU to communicate effectively over their lengthy bus.2601:140:8980:106F:8576:80E0:C309:7606 (talk) 08:34, 5 May 2019 (UTC)

Untitled

Does anyone else think this comes off as an advertisment? Someone with more knowledge on the subject should probably provide a counter point...

I removed the advertisement and cleaned it up a bit (it was a PPU and a CPU in the previous version :P).

NovodeX/PhysX

There's a lot of confusion between NovodeX (a physics programming API, much like Direct3D is to graphics) and the actual hardware board that Ageia is developing. Despite the common origin of the hardware/API, I don't think it's really relevant that the PS3/Xbox is going to support the software NovodeX simulation pipe. As such, I don't think those facts belong in the article. The same goes for statements like: The NovodeX API can also be used to program physics for general purpose hardware, but it is specialized to run optimally on the PhysX hardware.

Thoughts, anyone? --Kyle Davis 08:11, 5 December 2005 (UTC)

   It was called Novodex when they bought it. They have since renamed the API to PhysX as well. http://www.ageia.com/developers/api.html

I've been bold and removed the software information; it was all old information or duplicated from the PhysX article page --PdDemeter 13:18, 23 September 2006 (UTC)

Useless Anandtech

Calling the link a closer look takes the expression to new depths. Quotiong:

We really don’t know a great deal about the intimate details of the architecture,
but a light weight parallel floating point with lots of communications is a good
start. We’ve had several guesses at how the hardware works that have been confirmed
wrong. But to paraphrase Edison, eliminating all incorrect paths leads to the goal.

The only thing you get is a cookie storm and tons of advertising. I suggest we drop that link.

No objections received, link deleted.

What is a PPU really?

What the heck makes a core a "physics processing unit" different from an ordinary SIMD vector unit anyway? Is there some kind of special opcode for "calculate Laplacian" or "solve critically damped spring" or something? Collabi 10:17, 11 August 2005 (UTC)

I have tried to find out but cannot find details on the internals of the PPU. The advantage PhysX has over the Cell processor, is a SDK ready for use, complete with a tutorial. It is possible a patent search can unearth something more substantial than the glossy whitepapers that are available.

This article states that Sony made an agreement with Ageia to port its SDK to the Cell processor.

An SDK I can certainly understand; physics simulation is software, you can provide a middleware package to encapsulate that software for developers. You write an API for a software package that runs on a processor. But I can't imagine what special instruction set a chip might have for physics, other than the usual linear algebra. I can find only three patent applications by AGEIA and they seem to be mostly concerned with their algorithm for solving the Linear Complementarity Problem. The chip itself is only sketchily referred to, and appears to be a vector processor essentially like the Cell: a group of parallel math units each with its own memory island. Collabi 23:22, 11 August 2005 (UTC)

good question!; when I met the Agiea devrel chap at an X-Fest event, they described that their chip was VERY similar to the Cell -one worry they had was a rival releasing a Cell board for PC's. Hence the Cell is very worthy of being described on this page. Parallel evolution arriving at a similar solution to the same problem - realtime simulation...

There was one different component: 'a processor to manage inter-processor transfers' - I can't see how this could be far off what you can do with SPU/PPU initiated DMA's on the Cell. The whole thing reminds me of the days of DSP accelerator chips. This appears to be an accelerator card marketted around one killer app. But the Devrel guy was quick to describe the cores as custom CPUs and NOT dsp's, when I mentioned that term. I mentioned how I'd always perceived game Physics more as an excercise in collision - parsing complex spatial datastructures - rather than FP, and he answered by describing how the specialized memory architecture helped this.

I suppose they may have analyzed physics code and got their VLIW ISA & execution unit mix tuned exactly to that, wheras desktop CPU's will be based on running a range of common benchmarks & legacy apps..

I don't see anything specific in the processor design of a so called "Physics processing unit". IMO, the term either refers to the special use case of a Graphics processing unit (GPU)... and then this article should make this very clear, or it is simply more marketing bullshit (see buzzword bingo). User:ScotXW^t@lk 10:40, 10 June 2014 (UTC)

Distributed Computing?

How viable would a PPU be for various scientific and mathematical computation performed through distributed computing? Gaming and various 3-D modeling are likely to take only a few hours out of the day, but DC is 24/7 and can use all the speed it can get. Is this something BOINC and BOINC apps can and should support?

Well, for distributed computing apps to support it, they'd need to have some idea of what it actually is. At the moment there's almost no information on this, which leads me to believe that when it's released, all information about how it works, and thus how to make it do anything but game physics, will be top secret. There are some attempts to use GPUs to help with distributed computing, but it's an uphill battle, in part for similar reasons.

Vandalism?

"It is also rumored that the Nintendo Revolution will run with a PPU chip and clock in at 4 quadrillion petaflops, but still pale in comparison to the original Xbox. (Note: PPU in Revolution Rumor was formed by Drinky Crow, a man with tons of credibility.)"

Yeah, someone want to change that?

PPU in Rev

"The Nintendo Revolution is rumored to use a PPU. [3]"

Is there actually going to be a physical piece of hardware for "physics" or is it just software for it? Because it's linked to howstuffworks.com, and while it is informative, it isn't very up to date, and it DOES say "rumored"

Someone put that in there and it caused a ton of vandalism so I removed it, since it is just a rumor anyways. But someone later put it back in, so I added a "source" for the rumor. If you want to modify or remove it go ahead. Qutezuce 19:59, 12 December 2005 (UTC)

The link is now broken, so I'll just remove it. Chris M. 17:29, 2 May 2006 (UTC)

This is about PPUs, not PhysX

I took out the two sections that were entirely PhysX-related from it after copying some info into that article. While it's certainly appropriate to talk about the first major commercial PPU for a bit, the way the article was structured, it made it sound like the two were synonymous. Does anyone have any info about other PPUs or PPUs in general that can be used to buff up the article a bit? (I don't.) --Ted 19:08, 15 June 2006 (UTC)

Right now, it's the ONLY one on the market. ATI is teaming up with Havok, and their solution will be part of their crossfire system. More info is here. Corky842 22:56, 15 June 2006 (UTC)

Well, yes. That doesn't mean this should turn into a copy of the PhysX article, does it? --Ted 03:14, 16 June 2006 (UTC)

Alas I fear this article will never expand beyond a link to the PhysX: Havok FX is still software, albeit partially running on a GPU instead of a CPU. Is the article in need of renaming/removal? (perhaps when we have 2 PPUs in the world we can recreate it!) --PdDemeter 04:03, 23 September 2006 (UTC)

Intel aquired Havok !

Intel To Acquire Havok
This is notable - Intel may suppress Havok FX, to favour sales of multi-processor-core PCs or defer it until its own GPU Larrabee is released ! --195.137.93.171 (talk) 11:12, 7 December 2007 (UTC)