AMD considers buying Ageia

Interesting.



http://www.custompc.co.uk/news/601680/amd-considers-buying-ageia/page1.html

how does that children's song go "anything you can do, I can do better.."

Thats great. Expect the little developer to be hurt somehow.

AMD Already bought ATI who can see a chip soon that does everything outta the box.

AMD's move was in vain. I switched from AMD and ATI to Intel and nVidia. I like my new choice. "puki" is happy.

AMD - who are they?

ATI - what's that then?

Sausages.

I always liked AMD when they were the underdog producing typicaly better CPU's at lower prices. Intel seem to have had to face the music and put the effort in and are way ahead right now. Even AMD's new quad cores are 13% slower than the Q6600.

Ati are quite good, but they have never had very good GL drivers with a stable pro feature set. I've had a lot more reliability from Nvidia even with Nvidias beta drivers. They seem to have been very prolific releasing many betas and showing a vast improvement of late.

Not sure what to think of Ageia's Physx. I think the API is pretty good but the PPU is too expensive, I'd rather see better use of GPU's or multicore than have to spend $200 of a seperate physics card, and want an open standard for physics.

They're already in talks about this, or they wouldn't make an announcement like that. It'll probably happen.

The PhysX API is terrible. They need to open it up so physics programmers can use the low-level API like CUDA.

I would buy Ageia if the price was right...

...*starts counting coins in wallet*

I can, and Intel are making it. They're so far ahead of AMD at the moment, that Intel CPU's can already be used to do physics, graphics and general processing. AMD's quad core CPU's aren't even matching Intel's dual core CPU's in some respects.

If AMD buy Ageia as well, they'll probably still be behind on balance.

Based on my experience thus far, ATI and AMD are screwed. That is all.

I prefer AMD processors, all the intel processors I have had (above the 233MMX one) have all died a death. All of the AMD processors I ever had havent died once.

That step of AMD/ATI is their only chance ...

Intel bought Havoc and Intel collaborates with NVIDIA
So if AMD/ATI wants to remain in the GPU driven physics, their only real chance is to get it for sure ... ie make sure that their next generation of GPU has Ageia Physics units or at least co GPUs compatible to it so they have a hardware driven physics engine at their hand ...

Especially after the recent move of MS and TGC this might be an interesting option when looking at DarkPhysics and FPSC X10 are both using it ...

AMD needs to do better GPUs and CPUs, not losing time on hardware like this Ageia!

Thats what they plan to do by 2009 (given the regular AMD and ATI timeline misses, more 2010 or 2011 thought) with their Fusion platform ...

Question is if this will be soon enough to be crashed by NVIDIA and Intel ...

Of course that doesn't really mean anything.

I wouldn't rule out buying a spaceshuttle if the price were right, but it still ain't gonna happen. ;-)

If intel and NVIDIA strenghten their work together, it definitely will, as intel could make havocFX more "quantum physics" centric which would be catastrophic for AMDs "physics onboard" ambitions with the HD3000 and HD4000 series ...

and 14% cheaper.

For the moment.
When the 45nm roll out at intel, the current generation will see a price drop again and AMD will be more expensive.
thats just a short flicker of a "light at the end of the tunnel" for AMD ...

Intel are far faster right now. Was a really good deal on Quad core for thanksgiving too. $199 for Q6600 + mobo, $150 off regular price of $356.



The fastest AMD hovers around 10th place in the rendering and most of the game performance charts.

http://www23.tomshardware.com/cpu_2007.html?modelx=33&model1=946&model2=871&chart=424

Theres a whole bunch of comparrisons that can be made from that page. I still only have a 3000+ which is even slower than the one in their chart, mine being 2 sockets older.

The worst about AMD is that their "wonder CPUs" seem to have more problem that they officially stated. There are different sources that point out that the higher Phenom X4 are having some serious issue which is why they won't appear until Q1 08 at least ... so if Intel speeds up the Nehelem stuff now, AMD might even "die down" before they get crushed by Merom which they will definitely with their joke platform they currently intend to use "against intel"

Especially as they tricked too many AM2 costumers with making them believe that they wouldn't play their dirty game again and drop technology soon after ... yet they already are on AM2+ and users are screewed again.

One would think that they learned at K7 where they took market shares from Intel that the "consistency" is what made them having success, not their trashy CPU ... but it seems like they need a heavy wall crash again to find back to their roots instead of spreading that thin that intel can blow them away.

This cpu seems aimed at upgraders. For the cost effectiveness to work, you need to remember these chips can be used in old motherboards.

If you factor in both mobo and cpu changes, perhaps its still a very good deal.

"13%" slower is a little misleading, since that's an averaged value between all the benchmarks. If you look at the benchmarks, you'll find that AMD's new quad cores are around 25 - 50% slower than Intel's Q6600, while in a few other applications it performs a little better.

Regarding Ageia PhysX hardware, I think that physics just isn't specialized enough to justify dedicated hardware. Multi-core processors seem ideal for physics, AI, etc. in games because you're not locked to a single API, and the overall cost is much lower. For example, why would you buy a ~$200 physics card + CPU, when you can get Intel's Q6600 for ~$300 that will do everything the physics card does and more (AI, etc.)

However, I still think the Ageia PhysX API is useful. It's certainly one of the most powerful, feature-rich, and fast physics engines you can get for free, and doesn't need a PhysX card to work (it will even take advantage of multi-core CPUs). The only problems with Ageia's PhysX SDK is that the user needs to install ~30mb of Ageia Drivers before your game will work (you can reduce that to ~1 MB, but the license agreement doesn't permit it), and the code style of the PhysX API / tutorials / examples is a little "weird".

I think general-purpose GPU physics doesn't make much sense. Obviously it moves the physics load from the CPU to the GPU, but is that really a good thing? With today's quad-core CPU's, there's more than enough parallel processing power for physics. In this case, moving physics to the GPU will only slow down the frame rate of graphically-intense games because now the GPU has to share it's power with the physics engine, while your CPU most likely has an entire core idle.

John, GPU physics are pretty much only in use for small things like particles and such where they won't interact with much of the environment. So's been my experience looking around at what people have been doing with these new means of coding for GPUs.

Yeah, that's why I said "general-purpose" GPU physics. Particles and many small objects definitely benefit from GPU "physics" because otherwise this volume of dynamic objects would require a huge amount of bandwidth. You can even do limited AI on the GPU, like flocking, etc.

I haven't done much research on GPU based physics, so maybe I incorrectly assumed that there was a general-purpose physics engine implementation on the GPU.

The DX10 generation cards have physics capability, at least NVIDIA with their Quantum Effects stuff.
I assume AMD has as well as they hyped their HD trash that far with their "sound and physics onboard" (who the heck came up at them with the master piece idiot idea of sound on a graphic card?!)

Problem is just that you need to use OpenGL 3 or DX10 to use the unified shader units used for geometry shader and the like. Will not work with VS / PS approach on a usefull scale.

You don't have to use OpenGL 3 for this. What you need is CUDA, and it is still very young, so I don't expect anything to come from this for a while. I think multicore processors are a better approach, simply because a good CPU is more common than a good GPU.

CUDA is nice but has 3 drawbacks:

1. It needs CUDA capable hardware, not any crap nvidia will cut it. IE DX10/OGL 3 hardware, so I do not see the point of not directly using it.
2. ATI?
3. Its for scientific usage, not realtime. Especially the matlab plugin shows that