KYRO Power VR3, gamers favourite.

[RSP]

Hello,
Read about KYRo PowerVR3 review...
My Opinion:
No press release announced, suddenly comes as a reference board ready to test.
Tile Rendering techniques seems to be such interesting method to develop.
What Yous See Is What It Renders. Requires less bandwidth.
No T&L stuff, no DDR, it will be very cheap. (but still fast) Yummi..

I guess KYRO pressed up Matrox to soon release ANY of their pre-pre-announced chip or else they will lose their gamer fans.

[MadCat]

You know what, I find it seriously funny that so many people are responding favorably to the KYRO. Many of you were not around for the early PC 3d wars of 1996-97, but at that time NEC was touting their revolutionary Tiled display list architecture with Infinite Planes Deferred Texturing, plus no need for Z buffer. Of course, Infiniite Planes is the technology that eliminates overdraw in all PVR chipsets, and the Tiled Display List allows such things as the 32-bit internal accuracy, because unlike traditional cards each pixel has every texture combined and then is written to the framebuffer.

What Im trying to say here is, this technology is NOT new, its been around for quite some time. But developers shunned it in its earliest form because it was difficult to develop for, and it offloaded display list calculation to the CPU. Now they they have a full on-chip hardware display list engine and have robust API support, the PVR arcitecture has finally proven itself.

------------------
This Signature Space FOR SALE / RENT



[This message has been edited by MadCat (edited 16 June 2000).]

[Himself]

I think I first heard about the PowerVR chip around 4-5 years ago, the S3 virge was just past it's "glory" days. I was using an Amiga back then of course. I'm impressed that they stuck with it and finally have something to offer the PC market.

[The Rock]

In some kind of twisted way, it seems actually good that earlier incarnations of it failed. From everything I've read, it seems that tile-based rendering is a hard thing to make work properly, so maybe now they've worked out the bugs, and tossed in the Z buffer. Correct me if I'm wrong, but game developers don't have to anything special in their programming to take advantage of this rendering technique right?

The Rock
www.3dforce.com

[Arsa]

Hm... didn't m3d use PowerVR? ;-)
They had problems with drivers...
Noy KYRo... ok... we'll see... ;-)

BTW 3dfx bought Giga3D... that's the same (tiling), but thay clame to have no problems with OpenGL and DirectX...

[Mark F]

The KYro(w), humm?

------------------
OH NO, my retractable cup holder swallowed a DVD...
and burped out a movie

[ET]

I just love "What You See Is What It Renders". Kind of true for all chips, isn't it? And unfortunately, sometimes what it renders is complete buggy garbage.

[franksch3]

I am well aware that this rendering technique has been around longer. The NEC 250 (PowerVR2) also did tile based rendering, however it never seemed to come even close to a Voodoo3, TNT2 and G400.

However, this third revision seems to work considerably well if you look at the scores. Also by adding an external z-buffer, compatibility problems with most games have disappeared.

The beta board is running at 110 MHz, has only 2 single texturing pipelines. It uses 143 MHz SDRAM.

On the other hand we have a Matrox G400 MAX, running at 150 MHz, also two single texturing pipelines and SGRAM memory running at 200 MHz.

Third contender is the Geforce 256 with SDRAM (not DDR). It's core runs at 120 MHz, with 4 single texturing pipelines. I don't know how fast the SDRAM memory is on a Geforce.

So the theoretical fillrates of each board are as follows:

KYRO: 110 MHz * 2 pipelines = 220 MPixels/s
G400 MAX: 150 * 2 pipelines = 300 MPixels/s
Geforce 256: 120 MHz * 4 pipelines = 480 MPixels/s

Now looking at the various Quake 3 high quality scores at say 1024x768x32, the G400 MAX gets 'spanked' by both the KYRO and Geforce. Also the KYRO leaves the Geforce behind. Very well done, considering that the theoretical fillrate is much less than the two other cards.

The KYRO's scores are not that far behind the Geforce 256 DDR. And the final silicon should run at 125 MHz or even higher. A jump from 110 to 125 is a jump of 12% extra fillrate, so it might also beat the Geforce 256 DDR.

In other words, while the previous two revisions weren't impressive at all, this third revision clearly demonstrates that tile based rendering is very powerful. I would be surprised that within the year all new cards use a tile based architecture as memory never seems to keep up with the required memory bandwidth for the traditional architectures.

And I haven't even started about the extra precision which can be achieved with tile based rendering. ;-)

[raptor666]

You know the bad news about the liscence agreement with SDRAM etc (Toshiba and Rambus). this will unfortunately be another advantage of the KRYO price wise as the royalties off DDR are probably gonna be bigger than SDRAM . Bad news for DDR ram companies with 64mb etc on board.

Pete
I hate patents like this!!!

[Joel]

What about it's D3D performance? That was another downfall of the two pervious generation chips. Also I don't know if I can give up dual-head mode for using two monitors or being able to output to TV via s-video.

Joel

[RSP]

Thanks for all of your responds.
Tile Rendering might be quiet dificult way to do.
But the KYROS's Z-buffer stuff is a very nice way to solve bandwidth problem especially with high res and 32-bit render mode. That's really interesting me.
I read some infos that in standard rendering process: polygon and textures are processed first (in system memory by processor) then selected which one to go (depth test). What a waste. And the worst part is the processor was greatly involved in this job.

I have an idea..
KYRO do their rendering process in full 32 bit mode. I think its better if 3D card can select which mode to render a texture, and this can be done by detecting the texture's format such 4444RGBA-16 bit or 8888RGBA-32 bit, than 3D card can choose proper rendering mode. So 3D scenes created consist of both 16 bits and 32 bits process. This technique will save some bandwidth.

Finally: What You See Is What It Renders..
will be the main challege for every 3D card manfacturer besides of pushing silicons to fire killer fill rate.
Game developers will have some difficulties if they apply some new rendering techniques but if it can make their games looks better, run smooth and faster with non GTS2 systems...so it's worth to try.

I'm a satisfied G400 user and I'd love to buy next Matrox Fast and Cheap products.

Until that I still siting behind my Drum Set practicing some Dream Theater techniques.

Thank you for your attention.

[Himself]

Joel, no dual head, but it does have EMBM, and super high 16 bit quality.