150,00$ Kyro II beats Geforce2 Ultra!!!

[Verygreedy]

<i/dominant/block.gif>
Finally! Some more competition in the video market. I am looking to upgrade from a Voodoo 3 AGP card and I think I have found my new card. The new KyroII looks NICE to say the least. Kyro II Review From Anandtech!! I really like what they have done with this card. It even beats out the GeForce II ULTRA? In some instances yes. I refuse to pay over 200$ for a video card. Anyone who pays more on their video card than they do on their processor/mb is nuts! @$149.99 that fits perfectly with what I'd like to spend on a new card. Preliminary results &amp; reviews look good. I think I'm going to pick one up. Unreal Tournament (my fav game) results look especially nice! 270 RAMDAC is a bit weak, although a very welcome comp. to the video card market. Cheers For Kyro II

My video card progression; from 1995-present
Diamond Stealth 1MB Dram
Diamond Stealth VRAM card w/mem upgrade 2mb
ATI 4mb 3d Card
Diamond Viper V330 4mb
Voodoo II card
Diamond Monster fusion ..16mb
VooDoo III AGP card.. 16mb
* Kyro II prophet? New!* coming soon?

Anyone else like what they see? Would
YOU reccommend this card?

 

[Verygreedy]

BUMP .. cool topic

 

[Krioni]

VeryGreedy:

It makes sense that this architecture would do very well in UT. I have an old Matrox M3D based on the original PowerVR chip (no, I don't still use it ) It handled the oringinal unreal VERY good and had damn good visual quality doing it.

 

[BenSkywalker]

Teasy-

&quot;could you explain why you think it would be so hard for the Kyro to make use of the extra poly's that a HW T&amp;L unit would give?&quot;

Binning. I could argue about it until I was blue in the face but instead I'll point to a vocal supporter of deferred rendering and his question to Gary Tarolli-

&quot;14.
Dave
Deferred rendering is well known for the ability to display scenes very efficiently. An example of this is Gigapixel's design and technology. However, in the next couple of years these architectures are going to run into the problem of having to bin large amounts of much geometry data. With 3dfx (presumably) going to this type of architecture, that might bring concerns to the table. Are you confident that 3dfx will be able to get around this issue, or have you already found solutions? Can you elaborate?

We are confident that the GigaPixel design is the best solution even when dealing with the large amounts of geometry data that we expect to see in the future.&quot;

You might remember this interview(linkage for those who don't). I want to know how the he!l they are going to work around this potentialy serious bottleneck(which the PVR chips to date have) in tilers. Having to bin high levels of geometry has been a problem for tilers and to date the only thing I have been able to think of to work around it is a high speed cache of some sort for vertice data, but that would kill off one of the biggest advantages of the technique(not having to use exotic memory solutions).

The way I see it(and I know there has to be a better way) you are going to have to have a ~4MB minimum cache of eDRAM(or the like) to store vertice data, without it your are going to have to continue to shrink the size of the tiles. I'm also not sure how they will handle polygons that overlap into adjacent tiles without having to rehandle the vertice data multiple times. Right now, that isn't a major concern. Probably the highest level of geometry we have now is Giants(still want to see how the Kyro or KyroII fares in that title) in a real game and even that isn't close to straining any of the hard T&amp;L units we have now, not even the Radeon's. When we start to see titles that truly exploit the full power of even current hard T&amp;L the Kyro and KyroII are going to be lagging behind the traditionals from everything that I know.

My big interest is going to be how they work around it.

On the Naomi, I haven't seen a full official breakdown of the hardware, but from what I have seen of the titles it is handling they don't look very impressive to me(in comparison to what is possible with a very robust geometry solution). Check out the &quot;Nature&quot; portion of 3DMark2K1 in the demo, the trees and grass in that are the type of immersive environment I'm thinking about when I talk about T&amp;L being truly exploited. It runs real smooth on my GF1 DDR at 640x480 32bit, though it does choke hard when you up the resolution. For situations like that(massive overdraw due to the scene complexity) I can see how the benefits of a &quot;tiler&quot;, in terms of HSR, could be extremely beneficial. I just can't figure out how they are going to work it out.

I've heard the same thing about the NAOMI's hardware T&amp;L unit being superior to the GeForce/GF2's, I haven't seen anything to back it up other then &quot;it is&quot;. I also haven't seen anything on the throughput of the entire board(not just the T&amp;L engine). It is possible that even with the limitations of the PVR based chips they could have worked around it by doing what I mentioned above, it is an custom arcade design after all

 

[Teasy]

Well as I said the T&amp;L side of things really isn't something I'm greatly familiar with but I can see what you mean, however in the latest Kyro II preview there's a 3dmark2001 bench, in this bench the Kyro II and the Geforce MX both get the same score, 3dmark2001 high detail tests have more poly's in them then any game will for the next year or two IMO so if tilers really suffer so much from this why is not only a traditional but a traditional with a HW T&amp;L engine (the MX) getting the same 3dmark2001 score as the Kyro II in a prog that is totally limited in high detail by poly throughput?, I can see what your describing possibly being a problem with a huge amount of poly's but in a prog with a huge amount of poly's its not showing up, look at most Voodoo5 benches in 3dmark2001 in the beyond3d forum and there lower then that of the Kyro I scores.

As for NAOMI2, I've seen games on that that make the poly count of Giants look low, I haven't seen the 3dmark2001 demo yet because unfortunately Dx8 has a bug in it that makes 3dmark2001 think the Kyro can't render into a texture, this problem has been reported to Microsoft bt IMGTEC and they expect it to be fixed in the next Dx8 release, the 3dmark2001 bench works great though.

Your question is interesting though I just have no idea how to answer it, maybe one of us should bring the post over to beyond3d and see what answers we find there.