Kyro 2 - Part 2

[noko]

Awesome. Found the way to benchmark the demo, it uses the same commands as VilliageMark.
At 1024x768x32, Tri-linear filtering, Compressed textures, Max Multi-Texture used 3+3, FSAA off, I get 68.5 FPS. Smoking! I will do more and post latter.

<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 05/05/01 01:48 AM.</EM></FONT></P>

 

[noko]

Here some links to the Temple Demo test I've done showing the results .

<A HREF="http://home.cfl.rr.com/noko/Kyro2bench.jpg" target="_new">http://home.cfl.rr.com/noko/Kyro2bench.jpg</A>

<A HREF="http://home.cfl.rr.com/noko/Kyro2bench2.jpg" target="_new">http://home.cfl.rr.com/noko/Kyro2bench2.jpg</A>

<A HREF="http://home.cfl.rr.com/noko/Kyro2bench3.jpg" target="_new">http://home.cfl.rr.com/noko/Kyro2bench3.jpg</A>

 

[phsstpok]

Noko, what is the link to download this demo and how <b>do you run</b> the benchmark?

Thanks in advance.

 

[noko]

Download from here:

<A HREF="http://pvr.gamestats.com/files/templedemo.zip" target="_new">http://pvr.gamestats.com/files/templedemo.zip</A>

Command line to do the benchmark is -bench, <b>example: </b>

"D:\IntDownloads\Demos\Temple Demo.exe" -benchmark

Couldn't do different benchmarks since the command lines of villiageMmark doesn't seem to work except for -bench. Hopefully someone will enlighten us in the different options for this benchmark. I will do some experimenting to see if I can derive them myself.

 

[Guest]

the problem with geforces with a game/demo (like temple demo) with many textures layers is image quality..
kyro will allways offer better image in those situations...

 

[noko]

Huh? Can you show that?

 

[noko]

Looks like doing 6 textures on a Radeon is a breeze in the park buddy. The Radeon fillrate even with 6 textures is more then sufficient while the Kyro2s continues to degrade on each additional texture.

 

[Guest]

are you saying that radeon fill rate would not degrade?

what would be the point to build a benchmark that would hit kyro 2 the most ?

kyro 2 fill rate is with an average game (3-4 times overdraw) like:
175x2(pipelinesx4(overdraw)=1400 true fill rate (and that is true fill rate) it would need a traditional card with these kind of number to get kyro 2 fill rate
let do some math...
imagine a radeon at 200 mhz
2 pipelines and 3 TMU
200x2x3=1200 fill rate
hum good... but less than kyro 2 numbers...without even mentioning the bandwidth issues of radeon would get. (that would also kill some of thee fill rate)

kyro can do in a single pass 8 textures layers not using too much valuable bandwidth because of that...

the better picture that kyro 2 gets with more texture layers maybe is because all the blending is done in a tile buffer (very different from geforce process), But I don't know much about it...
all I Know is that it gets better quality


<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 05/05/01 05:03 PM.</EM></FONT></P>

 

[noko]

powerVR simmer down, the Radeon doesn't have a problem doing that benchmark is the point. The Radeon seems to be able to plow through 6 textures just fine. Now can you give me some command lines so that I can put my Radeon through some real tests like 1600x1200 where the bandwidth does come into play. I am sure the Kyro2 will do fine in the PowerVR Temple Demo and should do better than the Radeon, at least I hope so.
Well to determine better quality you need to be able to compare something, so post your Kyro screenshots of the Temple Demo and I will post mine hopefully looking at the same thing.
<P ID="edit"><FONT SIZE=-1><EM>Edited by noko on 05/05/01 06:16 PM.</EM></FONT></P>

 

[Guest]

I am too lasy..
so I take this screenshot from the web...
some hungarian site, I think ...


here are the shots from kyro I with default settings:
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0001.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0001.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0002.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0002.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0003.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0003.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0005.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0005.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0006.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0006.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0007.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0007.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0008.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/screen0008.jpg</A>

With 4X FSAA and anisotropic filtering
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0000.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0000.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0001.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0001.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0002.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0002.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0003.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0003.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0004.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0004.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0005.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0005.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0006.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0006.jpg</A>
<A HREF="http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0007.jpg" target="_new">http://www.hwsw.hu/letoltes/kyro/TempleMark_shots/ani_fsaa/screen0007.jpg</A>

let's go comparing...

<font color=red>Now can you give me some command lines so that I can put my Radeon through some real tests </font color=red>
hum...
I don't know more about these benchmark than you..
if I find somthing I will tell you !!

 

[Guest]

I have taken this from this hungarian forum site
<A HREF="http://www.hwsw.hu/ubb/Forum17/HTML/000437.html" target="_new">http://www.hwsw.hu/ubb/Forum17/HTML/000437.html</A>

look at this that I saw there:

<font color=red>Radeon ddr athlon 1350 mhz</font color=red>
Temple Demo 1.0.2 -Benchmark-
Benchmark, Frame
Driver, ATI3DRAG.dll
Description, Radeon DDR
Version, 4.13.2143
HAL, ON
T&L, ON
Trilinear, ON
Texture compression, ON
Max. Multitexture used, 3+3
Max. Texture used, 1024x1024
CPU Description, AMD K7 (Athlon)
CPU Speed, 1350 Mhz.
Screen Res, 1024,768,32
Average FPS, 47.708907

<font color=red>Kyro I at 115 mhz with ahlon 700 mhz</font color=red>
Temple Demo 1.0.2 -Benchmark-
Benchmark, Frame
Driver, pmx2hal.dll
Description, PowerVR KYRO 32MB/64MB
Version, 1.00.07.0089
HAL, ON
T&L, OFF
Trilinear, ON
Texture compression, ON
Max. Multitexture used, 6
Max. Texture used, 1024x1024
CPU Description, AMD K7 (Athlon)
CPU Speed, 700 Mhz.
Screen Res, 1024,768,32
Average FPS, 43.055475

Funny those radeon scores were lower than your's, maybe some old radeon driver, or some lower setting ...



<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 05/06/01 05:51 AM.</EM></FONT></P>

 

[Guest]

FSAA on vs no FSAA
<A HREF="http://pvr.gamestats.com/kyro/pics/tubes/pipes001.jpg" target="_new">http://pvr.gamestats.com/kyro/pics/tubes/pipes001.jpg</A>

taken from <A HREF="http://pvr.gamestats.com/start.shtml" target="_new">http://pvr.gamestats.com/start.shtml</A>

 

[Guest]

there is a nice benchmark that gives what would happen on a intensive overdraw game
<A HREF="http://hem.passagen.se/emiper/GL_EXT_reme.zip" target="_new">http://hem.passagen.se/emiper/GL_EXT_reme.zip</A>

noko you could test it with your radeon...

there is on beyon3d a topic with many results of that peculiar benchmark...
<A HREF="http://www.beyond3d.com/forum/noncgi/Forum1/HTML/002724.html" target="_new">http://www.beyond3d.com/forum/noncgi/Forum1/HTML/002724.html</A>

 

[noko]

Humus did an excellent job on that unique test and I saw Teasy's results there. Very impressive, especially with 8 overdraw factor in which his Kyro took virtually no hit. I have SP (special purpose drivers) in both WinMe and W2K in which the hierarchialZ is not functional, meaning I get lousy results from that benchmark at the moment except for the T&L tests. How about you posting here your results from Humus's benchmark. Also I notice that the Temple Demo is version 1.02, so where can I get that version? I have version 1.01. Those links does show very good quality in the rendering of the Kyro but I notice Anisotropic mentioned on one of them, is there a way to turn that on in the demo/benchmark? Now to clarify something, on Temple Demo 1.01 I get 68.5 FPS with the 8-bit stencil buffer off in the Radeon Drivers and 45.5 FPS with the 8-bit stencil buffer on. I didn't know about the effects of the 8-bit stencil buffer until after I experimented with the setting in the drivers. The quality is better with the 8-bit stencil buffer and obviously a hit in the frame rate. Still no command lines, I am sure there are some, we just have to wait until they are released.

 

[noko]

I was wrong, hierarchialZ does work on the 3132 drivers for the Radeon in W2K so here are my results for my Radeon. Notice the front to back benchmark where hyperZ does much better especially with a overdraw of 8, over 100% better. HyperZ does work well on the Radeon if objects are rendered front to back which the game developers are starting to do. (PowerVR post your results from your Kyro.)

Results:
--------

Overdraw/HSR:
-------------
Overdraw factor 3, back to front: 112.02 fps
Overdraw factor 3, front to back: 174.14 fps
Overdraw factor 3, random order: 139.08 fps

Overdraw factor 8, back to front: 46.42 fps
Overdraw factor 8, front to back: 104.96 fps
Overdraw factor 8, random order: 80.29 fps

Fillrate:
---------
Pixel fillrate: 203.25 MegaPixels / s
Texel fillrate: 605.09 MegaTexels / s

T&L/High polygon count static display list:
-------------------------------------------
Pure transform: 6479705 vertices / s
2 point lights: 5543256 vertices / s
8 point lights: 4941373 vertices / s
2 directional lights: 5555948 vertices / s
8 directional lights: 5537326 vertices / s

High memory bandwidth load/texture cache efficiency:
----------------------------------------------------
One 1024x1024x32 texture: 212.28 fps
Four 1024x1024x32 textures: 203.64 fps

 

[Guest]

Results:
--------

Overdraw/HSR:
-------------
Overdraw factor 3, back to front: 180.76 fps
Overdraw factor 3, front to back: 182.45 fps
Overdraw factor 3, random order: 181.21 fps

Overdraw factor 8, back to front: 157.34 fps
Overdraw factor 8, front to back: 157.87 fps
Overdraw factor 8, random order: 157.22 fps

Fillrate:
---------
Pixel fillrate: 155.44 MegaPixels / s
Texel fillrate: 227.89 MegaTexels / s

T&L/High polygon count static display list:
-------------------------------------------
Pure transform: 4524654 vertices / s
2 point lights: 1022425 vertices / s
8 point lights: 350457 vertices / s
2 directional lights: 2124122 vertices / s
8 directional lights: 1423200 vertices / s

High memory bandwidth load/texture cache efficiency:
----------------------------------------------------
One 1024x1024x32 texture: 111.45 fps
Four 1024x1024x32 textures: 109.78 fps

duron 800

 

[Guest]

I saw this in another forum (beyond3d HUMUS topic) !
Humus is the guy who wrote the above benchmark!
looks like NVIDIA is using Rambus tactics...
let's hope that ATI and kyro could do something against the "SATAN"

<font color=red>
Radeon supported GL_EXT_vertex_array_range (which is nothing but GL_NV_vertex_array_range by another name) in earlier drivers but it seams that it has been removed. I guess nVidia didn't like it, it seams to be nVidia proprietary according to extension registry at SGI's site ... which is annoying, every useful extension that is invented at nVidia seams to end up being nVidia proprietary. They are the only one doing this. They don't seam to understand what the "Open" part of "OpenGL" means.
</font color=red>

hum closedGL


after these bad moves from Nvidia I will not buy a single product from them anymore !!!
they were the ones that fight against Glide (proprietary API), now they are trying to make opengl like nvidia glide ?
FUxK tHEM !!!!<P ID="edit"><FONT SIZE=-1><EM>Edited by powervr2 on 05/07/01 01:39 AM.</EM></FONT></P>

 

[noko]

Kyro struting it stuff on those overdraw benchmarks, A+. In T&L now we are comparing the cpu to the GPU there. Nvidia better watch it, trying to corner a market can back fire pretty bad. Look at Rambus and how many enemies they created. The more Nvidia trys to dominant the more resistance they will get. Pretty soon they become the sole target for everyone. Maybe I am exagerating a little.

 

[noko]

Even with HyperZ the Radeon was no match to the Kyro at 8 factor overdraw. Pretty impressive. Overdraw just does not exist on the Kyro period.

 

[Guest]

That is the kyro 1?

 

[OzzieBloke]

Maybe not... if you spread manure in a room, and you corner yourself, the only way out usually is to get covered in sh*t.

Why someone would manure their room is beyond me, but that is what nVidia seems to be doing... manuring themselves into a corner. Strong-arm business tactics require you to have a strong arm, and nVidia's arm isn't that strong, all things considered.

Cow with legs spread wide either dead or playing 'cello.

 

[Guest]

Found a result posted by ram at Beyond3D for a Kyro 2:

Results:
--------

Overdraw/HSR:
-------------
Overdraw factor 3, back to front: 254.97 fps
Overdraw factor 3, front to back: 256.48 fps
Overdraw factor 3, random order: 256.58 fps

Overdraw factor 8, back to front: 222.65 fps
Overdraw factor 8, front to back: 223.00 fps
Overdraw factor 8, random order: 222.74 fps

Fillrate:
---------
Pixel fillrate: 215.81 MegaPixels / s
Texel fillrate: 346.86 MegaTexels / s

T&L/High polygon count static display list:
-------------------------------------------
Pure transform: 2477112 vertices / s
2 point lights: 529660 vertices / s
8 point lights: 186912 vertices / s
2 directional lights: 1205433 vertices / s
8 directional lights: 708764 vertices / s

High memory bandwidth load/texture cache efficiency:
----------------------------------------------------
One 1024x1024x32 texture: 166.80 fps
Four 1024x1024x32 textures: 162.46 fps

Looking pretty good

 

[noko]

LOL :lol: .

 

[noko]

Thats pretty good except for the T&L portion which shows the capability of a cpu vice a GPU.

 

[Guest]

there will be more board makers producing kyro 2 !

<font color=red>Powercolor have created a new company upon which non nVidia based VGA boards will be released. I guess this is a handy way for them to get around their Powercolor = nVidia only stance.</font color=red>
<A HREF="http://pvr.gamestats.com/start.shtml" target="_new">http://pvr.gamestats.com/start.shtml</A>


<A HREF="http://www.vgano1.com.tw/" target="_new">http://www.vgano1.com.tw/</A>

I never saw this name before...
must be equal to Low prices