Kyro Review At Firingsquad

[BFG10K]

There is a Kyro review at Firingsquad. Correct me if I'm wrong but I thought tile-based rendering was supposed to be really good? That's not what I'm seeing with the Kyro. It must be its mediocre fillrate holding it back.

At the highest setting of 1600 x 1200 x 32 the GF2 MX is demolishing the Kyro in Quake 3 and Evolva. That's what the Kryo is good at right? Eliminating memory bandwidth stress at high resolutions by not rendering what is not required? Even in the Quake 3 FSAA tests it comes last most of the time.

The only thing it wins is the Evolva FSAA (which is not high enough to play at anyway), the highest 3D mark setting (I don't buy video cards based on this) and UT (which was slightly faster). Also remember that this thing has 64 MB of RAM while the GF2 MX has only 32 MB.

Basically whenever the Kyro wins it's unplayable.

 

[Taz4158]

You're right it's a strange little card.

 

[legion88]

The Kyro is showing its superiority--if you know where to look. The Kryo only has a fill rate of 250 MPixel/sec while the MX has a fill rate of 350 MPixel/sec. The use of HSR/Tile rendering gave the slower card the edge where fill rate becomes an important factor (e.g. 1600x1200x16). The "faster" MX loses in the fastest settings in Quake III where single-texturing is used.

On the other hand, the Kyro does have some weaknesses to it. When switching to "normal" and "high quality" settings in Quake III, multi-texturing is now being used. Since the Kyro only has one texturing unit for each pipeline, its fill rate gets cut in half when single-pass dual-texturing is used. The only way to "overcome" this and keep up with the competition is to have a faster core speed to increase the fill rate.

I noticed that the MX and V5 do not suffer too much in 32-bit color at high resolution. It looks like the reviewer left texture compression on again. You need about 22MB of memory to store the frame buffer at 1600x1200. That leaves only 10MB for the textures so the MX and the V5 should be suffering some problems at that resolution while the 64MB Kyro should be rolling along.

 

[EvilDonnyboy]

Kyro has about a third the memory performance, half the fill rate, but almost par performance in most areas.

Tile rendering IS good. I can't see how you're not impressed by it. Imagine a GF2 or Radeon with tile based rendering (well the radeon does render similarly to the kyro, but not quite).

 

[shawn_16]

I had a link to that in my KYRO post but its way at the bottom
I'm satisfied with my powercolor evil kyro

 

[RoboTECH]

legion88, you made a good point re: TC being left on.

however, there's not really much of a reason to turn it off on a 4500 tho, is there?

 

[JayPatel]

the kyro is a good card...it just needs better drivers...then again which videocard doesnt need better drivers..imagine how fast this thing would be if games were planned with kyro technology in mind.

 

[BFG10K]

It looks like the reviewer left texture compression on again.

Of course he left it on. What reason does he have of turning it off (no Q3 image quality doesn't count)? Whether or not the Kyro has texture compression is irrelevent.

 

[r0d3nt]

If you want a review of the KYRO with DXTC on you can go to the swedish site www.sweclockers.com
Just look att the pictures.
You arent supposed to be able to use DXTC with the current drivers but they did a reg hack.

 

[MustangSVT]

so with performance like that Kyro should be a cheap card around $150 retail atleast, right? but i doubt it'll be that cheap.

 

[legion88]

BFG10K wrote:

<< Of course he left it on. What reason does he have of turning it off (no Q3 image quality doesn't count)? Whether or not the Kyro has texture compression is irrelevent. >>

So you like apples-to-oranges comparisons in your benchmarks, right?

I know a web site that had three chances to review or preview the Voodoo5 5500 (a V5 preview, a review of the AGP and anothe review of the PCI). Each time, the site failed to inform their readers that when they used the maxed out or &quot;high quality&quot; settings of Quake III, the V5 is incapable of doing trilinear. In fact, the author even hinted that trilinear is there, the implementation just &quot;sucks&quot; in his words. In short, the author wanted to convince his readers that not only can the V5 do trilinear but it can do it fast. The &quot;sucks&quot; implementation is what gives it its speed. You have to watch out for the spin that some reviewers put in their writings.

In any event, it is a well known fact that texture compression will not help 64MB cards in current games used in benchmarks (e.g. 1999 games like Quake III Arena). I have a 64MB Radeon. We all know that the Radeon is capable of S3TC in OpenGL. S3TC does not help the performance of the Radeon 64MB cards but it can help the 32MB variety a lot. I do not like reviewers setting up the testing conditions to make certain cards (in this case, the MX and the V4) perform better WITHOUT telling the readers why he/she did it. An explanation would go a long way.

A possible explanation would be something like this:

The fill rate of the Kyro is in the neighborhood of cards released last Spring 1999 (e.g. TNT2 Ultra, Voodoo3). NVIDIA and 3dfx released products with fill rates that are also similar to cards released last Spring 1999. So it makes sense to compare the Kyro to these cards. Unfortunately, the Kyro has a definate edge over the MX and V4 when a lot of space for textures and frame buffer is required (i.e. Kyro has 64MB of memory). To help negate the advantages of 64MB of memory, texture compression is used. This way, the reader can see how beneficial HSR/Tile Rendering can be without the results being influenced by the advantages of more memory.

An explanation like the above is all that is needed.

 

[Killrose]

Unless this thing is going to retail at about $75.00, who's going to buy it?
And maybe thats not cheap enough.