Unsymmetrical Physics Acceleration Mode of Radeon HD 2000

Tom|Nbk · 12 May 2007 at 00:37

Very interesting indeed.....

Though the AMD ATI enounce Radeon HD 2900 series VPU has already integrated strong physics acceleration ability, but it seems AMD ATI want to endow Radeon HD 2000 series VPUs with stronger physics acceleration ability, so they put forward unsymmetrical physics acceleration mode, composed of a Radeon HD 2000 graphic card and a Radeon X1900 graphic card, among them the Radeon HD 2000 graphic card is responsible for 3D rendering, the Radeon X1900 graphic card is responsible for physical acceleration.

The unsymmetrical physics acceleration mode can make up the relative weak physics acceleration ability of Radeon HD 2600/2400 series, it also can let Radeon HD 2900 graphic card be free from physics acceleration, and concentrate on 3D rendering tasks. The unsymmetrical physics acceleration mode has already been supported by catalyst driver.

However, if there will be a DX9 game supports ATI physics acceleration in the future, another mode in which X1900 is responsible for 3D rendering, HD 2600/2400 is responsible for physical acceleration is also a quite good choice. But AMD ATI possibly don’t support this kind of mode right now and in the near future.

http://r800.blogspot.com/2007/05/unsymmetrical-physics-acceleration-mode.html

Tooks · 12 May 2007 at 00:39

If I understand that correctly, there might be fewer X1900 cards for sale on MM then as folks upgrading will have a reason to hang onto them?!

You're getting very excited about this launch aren't you Tom?!

Tom|Nbk · 12 May 2007 at 00:42

Tooks said:
If I understand that correctly, there might be fewer X1900 cards for sale on MM then as folks upgrading will have a reason to hang onto them?!

You're getting very excited about this launch aren't you Tom?!

True, and im excited about getting any card right now, this X550 is driving me to insanity, glad ive got work to be getting on with, Give me that GTS back Tooks, I see you got a nice OC on it

.

titaniumx3 · 12 May 2007 at 00:43

However, if there will be a DX9 game supports ATI physics acceleration in the future, another mode in which X1900 is responsible for 3D rendering, HD 2600/2400 is responsible for physical acceleration is also a quite good choice. But AMD ATI possibly don’t support this kind of mode right now and in the near future.

Does this imply that the HD 2600 will be slower than the X1900? Like the 8600GTS?

Tom|Nbk · 12 May 2007 at 00:45

titaniumx3 said:
Does this imply that the HD 2600 will be slower than the X1900? Like the 8600GTS?

Possibly, or the fact the HD series have better physical acceleration than the X1900XT?

Tooks · 12 May 2007 at 00:56

Tom|Nbk said:
Give me that GTS back Tooks, I see you got a nice OC on it .

Not a chance!

Yeah, it clocks pretty well, I guess you must have run her in properly for me!!

You'll have a 2900 soon, and then your pain will be over!

Cyber-Mav · 12 May 2007 at 00:59

looks like typical ati tactics, can't compete with nvidia on pure power level so they just slap on some extra features to sweeten the deal.

either way the truth shall be revealed come monday. im sure there are loads of people waiting for monday to come now.

FreeStream · 12 May 2007 at 09:49

titaniumx3 said:
Does this imply that the HD 2600 will be slower than the X1900? Like the 8600GTS?

HD2800/8800GTS = 15/16k 2005? IIRC
my X1900XTX = 13-14k (Hit 14.5k once)
A Midrange card is gonna sit in the 10-12k gap TBH, which is slightly lower than a X1900XT.

Seems fair to me :/

Pottsey · 12 May 2007 at 10:00

What about game support? Which games use it? If none do whats the point?
Is it still using Havok FX so limited to none gameplay physics?

Hxc · 12 May 2007 at 10:47

Cyber-Mav said:
looks like typical ati tactics, can't compete with nvidia on pure power level so they just slap on some extra features to sweeten the deal.

Because that has always been the case :rolleyes:

NickK · 12 May 2007 at 10:56

The issue is the PCI-E bus bandwidth!

An X2950XT + X1950XTX are going to feel the bottleneck of PCI-E datatransfers.

Nymins · 12 May 2007 at 11:12

I think the PCI-E bandwidth is more than enough for it. Physx cards use the old PCI slots and soon to be PCI-E 1x, so i think PCI-E 16x will be great for it.

Cyber-Mav · 12 May 2007 at 11:40

even agp may be enough for it.

NickK · 12 May 2007 at 11:46

GPGPU use has always felt the effect of PCI-E bandwidth, as long as games are all ok with 256MB that most of the older cards have.

To give you an idea of CTM, which this would use:
X1950XTX doing MADD heavily optimised = 120GB/sec read CPU mem transfer rate. More often than not this would be in the region of 30-60GB/sec with r/w for algorithms.
PCI-E data transfers 1-2GB/sec max.
System memory = ~10GB/sec.

In this form of computing the big issue is the slow data connects, given the fact that graphic card <-> main memory has to use CPU0's hyper transport to the PCI-E cards and the CPU0 memory controller for AMD based solutions..

What is good - the more maths you can load into each point, the better.. so expect more complicated physics (less data transferred) in preference to larger volumes of points (more data transferred).

Sounds that ATI want to stem the tide moving to nVidia by allowing them to use their old hardware.. Unfortunately promise only becomes useful with implementations - show us the games roadmap that use it!

Tom|Nbk · 16 May 2007 at 21:27

Some more information in the new Catalyst suite that came with the 2900.

GPU accelerated physics is available on ATI Radeon™ X2000 series graphics cards. GPU accelerated physics can take advantage of the processing capabilities of more than one ATI Radeon™ graphics card.

Enabling GPU (Graphics Processing Unit) accelerated physics can improve the visual quality of some Direct 3D applications that can take advantage of physics engines. Typically, applications that use extensive geometry can benefit from enabling GPU accelerated physics. Contact your application developer to determine if your application supports GPU accelerated physics. If you have an ATI CrossFire™ configuration, you must disable ATI CrossFire™ on the graphics card that you want to use to perform the physics calculations before you can enable GPU Accelerated Physics on the same card. For more information, see Disable ATI CrossFire™.

From the Graphics Settings tree, expand 3D.
Click More Settings.
From the Graphics Adapter drop-down list, select the graphics card that will perform the physics calculations.
Note: If ATI CrossFire™ is enabled on the graphics card you chose above, the Allow GPU Accelerated Physics option is unavailable.

Select Allow GPU Accelerated Physics to place a check mark in the check box.
Repeat steps 3 and 4 for other graphics cards for which you want to enable GPU accelerated physics.
Note: Enabling ATI CrossFire™ for a selected graphics card automatically disables GPU accelerated physics for that same card.

Personally id love to try it but im not sure if my psu could be up to it.

Cyber-Mav · 16 May 2007 at 21:31

Tom|Nbk said:
Some more information in the new Catalyst suite that came with the 2900.

Personally id love to try it but im not sure if my psu could be up to it.

wondering how effective it will be against a agiea ppu. wonder if my x1900xt can be used as a ppu, ro is it only x2*** series card that can work in this way?

Tom|Nbk · 16 May 2007 at 21:32

Cyber-Mav said:
wondering how effective it will be against a agiea ppu. wonder if my x1900xt can be used as a ppu, ro is it only x2*** series card that can work in this way?

You can only use this with the new 2000 series, im contemplating sticking an X1600 Pro in and see how it does, we need a list of what works and what dosnt first though.

Pottsey · 16 May 2007 at 21:36

“I think the PCI-E bandwidth is more than enough for it. Physx cards use the old PCI slots and soon to be PCI-E 1x, so i think PCI-E 16x will be great for it.“
Physx cards use the PCI slot as its better for physics then PCI-E. PCI-E 16x will be worse then PCI but not by much. It’s not the bandwidth that matters but latency which is worse on PCI-E.

Certain cards like soundcards and network on better on PCI over PCI-E. People who swap sound and netword cards to PCI-E just because PCI-E is newer are often really downgrading.

IceShock · 16 May 2007 at 21:40

i read somewhere that the physX cards arent even 1/8th as powerfull as using a whole graphics card (depending on how good it is) as a physics card.

Pottsey · 16 May 2007 at 21:50

“i read somewhere that the physX cards arent even 1/8th as powerfull as using a whole graphics card (depending on how good it is) as a physics card.“
That was the ATI marketing team spreading BS. They measured there card in gigaflops and compared it to physX card and of course gigaflops make physX look 1/8th as powerful. The catch is gigaflops are meaning less for measuring how good something is at physics. The card with the higher gigaflops is not always faster. When you look at the specs that do matter at physics it’s the other way around ATI have a small faction of the power of a physX card at least in theory.

Get a CPU and GPU with the same gigaflops and they don’t do the same jobs at the same speed.