Kyro2 No faster then Mx! (???) :(

phsstpok · Jun 20, 2001

I have two times more frames than a geforce 2 mx 400 (those minimum don't count cause I have only 128 mbs)

powervr2 wrote 

"Dunes

35,2 fps average
35,5 fps average (without excessive peaks)
46,6 high peak
23,7 low peak
41,5 high sustained (24 frames in 0,8 seconds)
28,2 low sustained (33 frames in 1,2 seconds)

my disk was very quiet here...(less memory required for this test? )"

Noko wrote 

"Dunes

. . 16 bit avg/hi/lo
. . . 1024x768 - 54.1/105/35.1
. . . 800x600 - 63/120.6/40.3

. . 32 bit
. . . 1024x768 - 40.9/76.6/27.2
. . . 800x600 - 56.4/105.2/37.6"

It looks like your Kyro II has a little lower performance at 1024x768x32 than Noko's Geforce 2 MX-400.

noko · Jun 20, 2001

Well the first test PowerVR did was with V-sysc enabled the second is the one to go by:

Dunes

54.4 fps average
56.1 fps average (without excessive peaks)
122.2 high peak
19.4 low peak

In which case it does beats by MX400. If I enable V-sysnc on my MX400 it has a average way lower then 40FPS for Dunes.

Well to eat your C

kie and have it too, gotta get Rade

n II

phsstpok · Jun 20, 2001

Wow, that's a big difference but strange. When I find a game or a benchmark where peak framerates are well below my monitor frequency it doesn't make any diffence if I disable vertical sync. The numbers come out the same because vsync doesn't limit anything.

Another thing, with V-sync enabled Powervr2's high peak only reaches 46.6. This is strange because if a video card is fast enough peak framerates should be exactly the monitor frequency. This is a "brickwall" if you will.

Case in point, Serious Sam, Dune demo, on my Geforce256 SDR

V-Sync enabled: 30.3/53.9/20.5, Avg/Hi/Lo
V-Sync disabled 29.3/60.6/19.0

Not much difference.

Anyway, I'm still having a hard time trying to decide on which video card to replace my Geforce256. I want to be able to play today's games and maybe look for a Geforce 3 in a year. I was leaning toward a Kyro II but problems, incompatibilities and sketchy performance reports from real people have me hesistant.

Would you please do a quick check for me with the Radeon? Could you clock it down to the equivalent of a Radeon LE and run a quick benchmark on Serious Sam and maybe Quake III?

phsstpok · Jun 20, 2001

Sorry, Powervr2. I did not see your revised benchmarks.

Have you worked out all of the Kyro II's problems? Did DX8.1 beta fix the performance issues in 3DMark2001? Are you finding any game compatibility problems? Other people have.

noko · Jun 20, 2001

Radeon 64 clocked at 146mhz (underclocked), T-Bird 1.4gh, W2K, Driver 3218:

Serious Sam, 102c

. Dunes
. . 1024x768x32 avg/hi/lo
. . . 57/172.5/33.3

QuakeIII, 129g, demo Four, Max, with compressed textures, no anistropic filtering, with enhanced weapons effects (the new beta patch deletes the old demos)
. . 1024x768x32 avg/hi/lo
. . . 57.3 FPS

Well to eat your C

kie and have it too, gotta get Rade

n II

phsstpok · Jun 20, 2001

Quake III, Demo four with compressed textures probably is isn't a useful comparion since I only have Quake III Arena Test, a demo version.

Nice numbers on the underclocked Radeon, at least from my point of view. Is that with both the GPU and memory underclocked?

I forgot you had the Radeon vid card with a 1.4 Ghz Athlon system but at 1024x768x32 the Radeon should be the limiting factor and those numbers do look good. A Radeon LE is looking better and better for my Duron system.

Thanks, Noko. (That was fast).

p.s. Running my little tests today I just realized the Serious Sam was running smoothly with the Extreme Quality Addon, even at 1024x768. This didn't work when I first installed Serious Sam. I used to get minor hesitations. I wonder which of the many changes I've made corrected that problem. Was it the BIOS update, Geforce driver update (12.40), Geforce Tweak update, or more memory? Probably it was more memory, although the hesitations did not seem to be related to disk drive activity (as in virtual memory access). The added memory sure speeded up game loading, though. This is with Windows 98SE.

noko · Jun 20, 2001

Those where numbers with both core/mem set to 146mhz on the Radeon. Serious Sam loves memory and a fast cpu.

Well to eat your C

kie and have it too, gotta get Rade

n II

Guest · Jun 20, 2001

when I test serious sam (the first time) I had external z-buffer with V-synv enabled that would kill kyro 2 performance a little... (the kyro 2 internal z-buffer will allways be faster)
this external z-buffer doesn't give any better quality it's only there for some cases (some few old games) that requires this external z-buffer... (this external z-buffer requires more cpu power)
please remember that my low 128 mb ram is killing also kyro average results...

remember that kyro 2 drivers chooses the best options for all the games... (I choose to tweak to see what I would get)

I didn't find any compatibility issues with kyro any..
and I have lots of games..

with my older games I always enable 4XFSAA .. this games don't let me choose above sometimes 800x600 and this 4xFSAA give me a boost in quality..
that is a big point with kyro...

only voodoo5 and geforce 3 FSAA might be better than kyro 2...Edited by powervr2 on 06/20/01 04:46 AM.

Guest · Jun 20, 2001

<<<I know the that the Kyro II does better with CPU scaling than other video cards but this is mostly at high resolution, is it not? The reviews and benchmarks that I read seemed to indicate that the Kyro II doesn't show much advantage until 1280x1024. At higher resolutions the Kyro II seemed to challenge Geforce 2 Ultras in some benchmarks but at lower resolutions, including 1024x768, the Kyro II seemed to perform on par with only Geforce 2 MXs.>>>

I wouldn't say that there's many games that the MX can compete in with the Kyro II at 1024x768x32.

<<<If I am correct about this there is a dilema with the Kyro II. With a low price it would seem to be an ideal card for a budget system. However, because budget systems are likely to have less than state-of-the-art processors and 17" monitors(maybe even smaller, like me) that aren't suitable for resolutions higher than 1024x768 the Kyro II does not seem to be a good fit.>>>

Your forgeting two things, firstly a Duron 700 is a budget CPU and can be easily overclocked to 900+ MHZ and there's no need for a big monitor when you have FSAA. FSAA on Kyro II is even better then high res because its incredibly efficient. On a traditional card like the Geforce 2 1024x768x32 4xFSAA needs a 2048x1536x32 front buffer and a 1024x768x32 back buffer. That’s 15.5mb for those 2 buffers, then there's the 2048x1536x32 Z-buffer which is 12.5mb so for 1024x768x32 4xFSAA the Geforce 2 or Radeon needs 28mb just for the framebuffer and Z-buffer. When the Kyro II performs FSAA it downsamples each FSAA'd tile before it goes into the framebuffer in ram, therefore it only needs one 1024x768x32 framebuffer which takes up 3.1mb of video ram, and of course it has no Z-buffer so in total for 1024x768x32 4xFSAA the Kyro II only needs 3.1mb of ram. So just to recap when using 1024x768x32 4xFSAA the Kyro II needs 3.1mb of ram while the Geforce 2 or Radeon needs 28mb of ram and this is totally forgetting texture storage.

noko · Jun 21, 2001

Your also forgetting that the Radeon slows down to a craw when doing FSAA especially at 1024x768x32. The Radeon FSAA is only useful at 640x480 and 800x600. Usually I just up the resolution higher on the Radeon because it does much better at higher resolutions then lower ones with FSAA. Also, note my MX400 FSAA is so limited that it is totally worthless, unless you want to play slide show games. I will run some benchmarks and show why latter.

Well to eat your C

kie and have it too, gotta get Rade

n IIEdited by noko on 06/20/01 07:55 PM.

phsstpok · Jun 21, 2001

I wouldn't say that there's many games that the MX can compete in with the Kyro II at 1024x768x32.

I saw that using the links that Powervr2 provided. I guess I was remembering early Kyro II reviews. I should have realized that things would improve with each driver revision.

Thanks for reminding me about video memory consumption. Manufacturers of nVidia based cards must have discovered the shortcomings of only 32mb of video. More and more inexpensive models are coming with 64mb of RAM. Can the NV15 address more than that? Will we see 128mb video cards?

As for FSAA, it's Kyro II does a nice job with this but it isn't as efficient as I was first led to believe. If I had my choice I would take higher resolution over lower resolution with FSAA but that choice doesn't come cheaply.

OK, you may have swayed me back to Kyro II but I will wait and see if the DX8.1 issue is resolved first. Plus, maybe the price for the Kyro II will come down while I wait.

Guest · Jun 21, 2001

directx8.1 beta build 620 have already enabled rendering into textures with kyro...

That was the single feature of kyro that went disabled in directx8

phsstpok · Jun 21, 2001

Cool! Does this improve 3DMark2001 scores? If not what do you think is holding back the Kyro II in this benchmark?

Guest · Jun 21, 2001

yes it improves 25%, 30% more or less with my duron at 900
more than 1700 with win2k

what is holding back kyro 2 with 3dmark2k1?

1.cpu power...(for software T&L)

those theoretical tests with 3dmark2k1
2. fill rate (there are no overdraw there like with any real game)
3.high polygon count
there are no bandwidth stress (not much fill rate required)in there (like in a real game) so those T&L engines of some cards can pump lots of polys.

any game with a benchmark built in it would be better for comparing kyro with others cards.
(real world situations)

Guest · Jun 21, 2001

I forgot to mention that there is still a bug with kyro 2 with 3dmark2k1 (not sure if it is because of directx8.1 beta)

the pilars in the high detail lobby test are transparent...
that is killing the overdraw with kyro ... so kyro will get better performance when this is corrected...

Edited by powervr2 on 06/21/01 00:36 AM.

phsstpok · Jun 21, 2001

I'm curious how Kyro II will do with Max Payne when it is released. Max Payne use the graphics engine, Max-FX, which 3dMark uses.

...but I'm glad things continue to improve.

phsstpok · Jun 21, 2001

Powervr2, you probably already know about these but GameSpyDaily pointed me to some new Kyro II drivers. I'm just passing on the info.

Get them here

Win 9x/ME
<A HREF="http://www.fileplanet.com/index.asp?file=61697" target="_new">http://www.fileplanet.com/index.asp?file=61697</A>

Win2K
<A HREF="http://www.fileplanet.com/index.asp?file=61698" target="_new">http://www.fileplanet.com/index.asp?file=61698</A>

Edited by phsstpok on 06/21/01 01:50 PM.

Guest · Jun 22, 2001

thee is no problems with lobby low detail (with my duron 900) more than 60 fps average..
thanks about the drivers...
a new set every week..

phsstpok · Jun 22, 2001

What about the other parts of the 3DMark benchmark? They MAY all be important for future games. Granted games need optimizations for playability where a benchmark does not.

Weaknesses in the benchmark may help pinpoint video card areas that need "tweaking" or driver refinement.

I think this claim has validity because I found this quote in the 3DMark2001 Readme file.

"3DMark2001 uses a real game engine (MAX-FX Technology™, by Remedy Entertainment) in all tests. This means that the obtained performance level in the game tests is accurate. If additional time would be spent on implementing more content, interactivity and game storylines, all game tests could be made into real games. We have proven this by making a game demo of the first game test. The scene is the same as in the test, but this time you are behind the steering wheel."

phsstpok · Jun 22, 2001

Noko, you're a video guru and I was wrong again!

I thought I had disabled Wait for V-Sync but I did not know that Serious Sam overides my video driver settings. Making things worse, in the Advanced Rendering settings the Monitor Refresh rate was set to 60 Hz.

With these things corrected this is what I get for Serious Sam, Dune Demo, Extreme Quality Add-on, Anisotropic filtering=none.

52.1/104.5/33.7, Avg/Hi/Lo

where before I was getting

30.3/53.9/20.5

How the hell does Serious Sam achieve such good framerates, with so much detail and activity?

noko · Jun 22, 2001

Very good programming, it uses an advance method of changing model complexity as the distance increases a less complex model is used (lower polygon), saving tremendous amount of bandwidth allowing many more models to be active on the screen at once. A form of Level of Detail but for models. I to am very impressed with Serious Sam game engine and I hope it is used more in future games.

Well to eat your C

kie and have it too, gotta get Rade

n II

phsstpok · Jun 22, 2001

I didn't actually expect an answer. You are a guru!

More games with the Serious Sam engine would be amazing!

jmc2 · Jun 22, 2001

Duron750/900,128meg
Abit kt7, Dx8.0
Kyro2 103 drivers (DEFAULT settings)

I've patched Kyro2 and SAM to latest(mid june)
Anisotropy OFF REALLY makes a difference as
you all have mentioned.

I'm going to try to test(later) the different
compressions offered and see if it makes a
difference (wish the demos did NOT vary from run
to run)
Serious Sam Ver 1.0
Quality setting plus EXTREME High Quality ini addon
demo FPS (peak)
Karnak Ave Low High

640x480 19 4 70
800x600 16 4 58
1024x768 10 3 51(pre 102patch&Kyro2 1.22 drivers)
11 7 31(256Megmem,first run,AFTER patch)
11 7 32(second run)
21 8 68(Anisotropy OFF)

1600x1200 4 2 21

(Looks like the patch 102 drops the HIGH peaks
(and 256Meg memory ups the LOW peak a LOT)

Memphis
640 27 9 63
800 23 9 48
1024 15 5 42 - PRE PATCH 128Meg
13 4 36 - (128Meg)Sam102patch+newest K2 Drv-1.22
15 6 34 - second run of demo(128Meg Memory)
16 9 36 - AFTER PATCH 256Meg,first run
16 9 40 - Second run
26 12 59 - Anisotropy OFF Post patches,EXTREME HQ
High Sustained49 Low S 17

42 19 82 - DEFAULT QUALITY "ADD ON" INI
1600 6 4 28

Dunes
640 47 18 113
800 31 15 88
1024 20 11 66
50 24 115 - NO anisotropy Post Patches
1600 8 5 11
AVE LOW HIGH"PEAK"

Seems 256Meg memory helps the LOW "PEAK' a lot
and the AVE a little. 5-10 percent
******
PIXEL FORMAT
number 6(type_RGBA)
detected... Power VR Kyro 1.2.1
draw to windows support opengl
double buffer swap exchange
colorbits 32 8;8;8;8
depth 32 (0 for stencil)

jmc2 · Jun 22, 2001

Duron750/900,128meg
Abit kt7, Dx8.0
Kyro2 103 drivers (DEFAULT settings)

Serious Sam Ver 1.0
Quality setting plus EXTREME High Quality ini addon
demo FPS (peak)
Karnak Ave Low High

640x480 19 4 70
800x600 16 4 58
1024x768 10 3 51(pre 102patch&Kyro2 1.22 drivers)
11 7 31(256Megmem,first run,AFTER patch)
11 7 32(second run)
21 8 68(Anisotropy OFF)

21 4 41 - POWERVR2'S RESULTS
1600x1200 4 2 21

(Looks like the patch 102 drops the HIGH peaks
(and 256Meg memory ups the LOW peak a LOT)

Memphis
640 27 9 63
800 23 9 48
1024 15 5 42 - PRE PATCH 128Meg
13 4 36 - (128Meg)Sam102patch+newest K2 Drv-1.22
15 6 34 - second run of demo(128Meg Memory)
16 9 36 - AFTER PATCH 256Meg,first run
16 9 40 - Second run
26 12 59 - Anisotropy OFF Post patches,EXTREME HQ
High Sustained49 Low S 17

42 19 82 - DEFAULT QUALITY "ADD ON" INI
1600 6 4 28

Dunes
640 47 18 113
800 31 15 88
1024 20 11 66
50 24 115 - NO anisotropy Post Patches
1600 8 5 11
AVE LOW HIGH"PEAK"

******
PIXEL FORMAT
number 6(type_RGBA)
detected... Power VR Kyro 1.2.1
draw to windows support opengl
double buffer swap exchange
colorbits 32 8;8;8;8
depth 32 (0 for stencil)

My Kyro2 is model hercules prophet 4500 OEM)

256 Mem definalely drops the disk accessing.
Short brief blips instead of a good run on the
Hard drive. raise the LOW peak a LOT!

jmc2 · Jun 22, 2001

hmm, you beat my ave every where but KARNAK
you 15 ave. me 21. Must be my 256 memory.
you LOW 1 me 8 (3 with 128 meg memory)

The 102 patch does seem to drop the HIGH PEAKs
a good bit.

There are several SAM "dump" commands in the consol.
I can get only ONE to work...
DumpDemoProfile():
-
Dump extensive demo profile to file 'temp\DemoProfile.lst'.
(output rate is controled thru dem_iProfileRate)

REALLY annoying!!! that I can't "select" text in the console.
The other dump commands are...
MakeWorldStatistics():
-
Dump statistics about current level to file.

DumpProfileToConsole
DumpProfileToFile

Can't get the above three to function. They seem
to want somekind of (?)input with their use.
ANY ideas out there???

Thanks,
jmc2

Kyro2 No faster then Mx! (???) :(

Splendid

Distinguished

Splendid

Splendid

Distinguished

Splendid

Distinguished

Guest

Guest

Guest

Guest

Distinguished

Splendid

Guest

Guest

Splendid

Guest

Guest

Guest

Guest

Splendid

Splendid

Guest

Guest

Splendid

Splendid

Distinguished

Splendid

Distinguished

Distinguished

Distinguished

Share this page