PowerVR® Shows Off its Curves at GDC 2004

Yes baby !

Now when will I be able to say "PowerVR rocks my world" ?

 

Go, Simon.......

 

I'm a teapot...!

 

Looks like trueform to me or Npatches.

 

Very nice but I'm too poor to buy anything that isn't related to my PC.

PVR5?

 

Definitely not. N-Patches are a relatively simplistic tesselation model meant for increasing LOD from a certain level. This appears to be closer to hardware accelerated progressive mesh support.

 

Why would you need hardware support for progressive meshes? In fact, what would hardware support of progressive emshes consist of exactly?

The standard teapot data set is a set of bezier patches so it probably just tesselates those. But what do I know.

 

On the fly tesselation would be very nice, if done right, which looks like.
That way you don't even need tesselated geometry, gone mesh optimisation, simplification, pre processing for LOD, MRM...

AFAIK what's been available was either not fast enough, or not capable enough.

Of course, screenshots are not enough to KNOW if it's really that good, but it hints that it's quite good.

 

I think the majority could not give 2 hoots about any pvr announcement.

No PC hardware, means Imgtech who?

Im bored waiting already.. Why don't they release something special. We all know they could.

edit: clarification

 

Well, I didn't say that was what it was -- merely that it resembled that much more than it resembled N-Patches. In fact it may not be progressive mesh support explicitely, since that would mean a fixed function feature. More desirable, of course, would be a primative processor that could be programmed to produce results like progressive meshes or N-Patches.

The only problem is that would imply a fixed function process -- not too much different from TruForm.

 

MBX is from what I can tell very much real hardware. Is the specific announcement about something else and I haven't noticed?

 

Me thinks this could work really cool with terrains.

 

Demonic.

Whilst I also am agrieved that there is NO PowerVR PC part at the moment it should not detract from the significance of the work IMGTECH have put into MBX. MBX is head and shoulders above its peers in terms of power consumption, featureset and performance IMO.

What it says to me is that IMGTECH are forward thinkers in their development efforts on all fronts, PC included. They have to be in order to win licencees. The fact that they can't get a PC part out if their lives depended on it is irrelevent to their profit margin at the moment. Damn frustrating for anyone who can see merit in their technical approach mind you.

Personally I'll be amazed if I see another PowerVR PC part on the shelves EVER. There don't seem to be anyone out their prepared to take a gamble in that market and license their products....

 

I figure Wavey asked quite some time ago, if a release is at any price dependant on a license, for a reason. Anything one could figure out in that case is probably not ideal, yet at least far better than having such an ambitious design rotting on shelves.

True. Yet IMHO the purpose of taking the risk of developing a high end design (all by themselves ?), should be very much relevant to their future plans and profit margins. I heard of ambitions to have in a couple of years more than a couple of dozen of chips under development at the same time, those can't be just based on the current palette.

Series5 has covered with the SEGA deal 1/3rd of the original triple target; meeting the console target does seem impossible to me at this point. Yet I don't see reason enough for the time being to exclude the possibility of a second target being met (under any circumstance).

Again all IMHO of course.

***edit:

*wears a large garlic chain, grabs a crucifix and crawls under the desktop on the lookout for Joe....*

 

Bleagh! Definitely not though I suppose I could put in an interface for them some day!

 

http://www.powervr.com/Downloads/Factsheets/PowerVRMBX.pdf

Effective fill rate of MBX 2.5m tri/sec

http://www.theregister.co.uk/content/68/36451.html

ATI Imageon as part of MSM6xxx or more likely MSM7xxx chipsets, gives "3-4 million triangle capability".

I too believe(d) MBX to be the highest performing handset solution, until I read this. Comments anyone?

 

My point was just that there's no point in having hardware support for progressive meshes since there's nothing to support. What would the hardware do exactly? Building the progressive mesh is an offline , one time process and can't (and shouldn't!) be done by the GPU. Evaluating the error metric used to determine LOD at runtime is best done on the CPU. The only work left doing that could be hardware accelerated is any geomorphing between LODs, but that can be done by a simple vertex shader. If PoverVR are claiming that geomorphing on a progressive mesh is supporting curved surfaces in hardware, I'm afraid we will have to beat their marketing department with a very big clue stick. And to be honest, I don't think they ARE claiming that.

Why would tesselating Bezier patches necessarily be a fixed function process? The PS2 VUs can du bezier patch tesselation and they are hardly "fixed-function". A full programmable tesselation processor might be asking a bit much, but if it can support N-Patches as well (I'm no D3D guy but nvidias RT-Patch support was essentially bezier patches as far as I can remember) it certainly seems pretty flexible. MBX is looking mighty impressive.

edit: stupid quotes...

 

Spot on.

 

I say that it would be a fixed function process because they're merely a set of data processing pipelines whose functionality cannot be changed. Instead, only the data fed into them can be changed, of course. There may be many functions, but none of them are modifiable except in the data that's fed into them. Contrast this with a programmable pipeline whose functionality is not defined by a set of static functions, but by a set of individual instructions with which one creates functions.

Of course, now that I think about it, what PowerVR might be doing probably could just be Beziers. Until a specification for a programmable primative processor is unveiled to IHVs, this would have to be an experimental implementation exposed only through an OpenGL extension (of course, it could be this, too).