Kyro 2 the killer of nvidia ???

noko · Apr 15, 2001

I too enjoy your posts and Serious Sam sure flys on the KyroII. Indicating very good OpenGL drivers both in Win9x and W2k. I did some T&L test and will post latter today. It is very interesting and accurate and reflects future games. Just too tired now to go into the details. I think we concluded that the KyroII fill rates will not be limiting until 6-8 textures are being used in very high resolutions. Then again like Serious Sam the number of textures of an object can vary depending on the object requirements. Meaning that a game could use 8 textures on some objects and 3 on others which would mean the KyroII fillrates would indeed be sufficient virtually in any number of textures from 1 to 8 as long as the textures application is selective. Good night and we can pick up this discussion tomorrow, that is if you wish. Thanks.

Edited by noko on 04/15/01 11:32 AM.

Guest · Apr 15, 2001

A new review on kyro !
<A HREF="http://www.acidhardware.com/reviews/3dprophet4500/index.shtml" target="_new">http://www.acidhardware.com/reviews/3dprophet4500/index.shtml</A>

where is the tomshardware review IN ENGLISH ?????

hum even in a low test system kyro is doing great...

Test System

* Asus A7V
* AMD Duron 650MHz
* 256MHz PC-133 SDRAM
* Quantum LM ATA/66 15GB HDD
* SBLive! Value
* AOpen 53x CD-ROM
* Windows 2000 Professional
* 7.51 Hercules Kyro II Drivers
Edited by powervr2 on 04/15/01 12:45 PM.

Guest · Apr 15, 2001

Nvidia drones???

Youre just jealous, cuz you cant afford a GeForce 3 and i can!

Who is General Failure and why is he reading my hard disk?

Guest · Apr 15, 2001

if you earn the money to buy a geforce 3 in a hard way (working 8 hours a day, many days)
You wouldn't buy one even if you actually could afford it !!

if it's your's dad money try to think what it is to work 8 hours a day 5 days a week to buy a geforce 3 ...

you should start thinking that for the price of a geforce 3 you could buy 4 kyro II ... and what you could get NOW with geforce 3 is "playing" some nvidia demos...

Guest · Apr 15, 2001

3 Days work.

Who is General Failure and why is he reading my hard disk?

Guest · Apr 15, 2001

one rich boy !
here in portugal, we would have to work (in average) a full month to buy a geforce 3 and that is without eating paying the electricity bills etc...
😉
for a kyro 2 one week should be enough (5 days )
😉

Negaverse23 · Apr 15, 2001

If I remember correctly, games such as Quake3 have no previously compressed and specially processed textures in standard delivery. (Guess the GTS and Radeon do have this option.) As for the bad sky problem, 6.47 version of NVIDIA's Detonator fixed this problem. Currently, Nvidia has forced DXT5 in Quake III. The DXT5 unpacking algorithm gives the best graphics quality over the other 4 algorithms.)

=
Ran out of bullsh!t to feed to the flies.

noko · Apr 15, 2001

Lets get some quotes out of the way:

The darkest, meanest, and most entertaining action game in years is almost here,..... We traveled to the frozen gloom of Helsinki, Finland, to play through this underworld action opus - and we're back, we're thawed out, and we're here to tell you that this long-simmering crime noir will be one of the most fun and influential action games ever made. May Issue of PCGAMER 2001

The Game, Max Payne

As you both unload a hail of fire, you'll hit just about everything in the hallway - linoleum tiled floor, plaster walls, light fixtures, glass-covered posters, you name it. Everything will react individually to the bullets. The hallway will be a chaos of flying plaster, chunks of linoleum, shattering glass, and shredded paper (which will drift slowly to the floor, while the heavier debris falls fast).......You've never played action sequences with this level of demolition.May Issue of PCGAMER 2001

This is one of the best game reviews I've ever read and I highly recommend anyone who is into actions games to buy this issue now.

The Engine is from Remedy Entertainment next door to MadOnion. 3dMark2001 game 3 uses the Max Payne engine in which you can benchmark your computer's graphical power. Plus you can use it as a test tool to see the affects of hardware T&L compared to software T&L in a modern future DX8 engine. A reflection of things to come in the gaming world. I will also be testing the affects of AGP2x versus AGP4x since the Kyro2 is only a AGP2x board and the large amount of polygon data may saturate 2x AGP speeds when doing software T&L.

My gig: IWILL KK266 650@866 Thunderbird, Radeon 64 VIVO Retail, 512mb pc133 cas2 Crucial, Leadman PowMax 400w p/s, 20gb IBM drive, 12x DVD, Pacific Digital 4x/4x/24x burner, Linksys ethernet card and hub, 3Com cable modem, HP 20" Trinitron, Microsoft digital Sound sys 80 sound/speaker system, Microsoft optical mouse, Win2k/WinMe dual boot

TESTS:(All tests conducted in W2k with ATIs official 3102 drivers, Radeon at stock 183/183 clocks, 3dMark2001.)

. . . Game 3, MAX PAYNE, AGP 4x

. . . . . 800x600x32 (HardWare T&L)/(Software T&L). . . . HWT&L/SWT&L
. . . . . . . Low Detail: 68.2FPS/53.1FPS . . . HWT&L is 28% faster
. . . . . . . HighDetail: 31.6FPS/24.1FPS . . . HWT&L is 31% faster

. . . . . 1024x768x32
. . . . . . . Low Detail: 58.2/50.3 . . . HWT&L is 16% faster
. . . . . . . HighDetail: 28.9/23.8 . . . HWT&L is 21% faster

. . . . . 1152x864x32 (This is my normal gaming resolution)
. . . . . . . Low Detail: 49.5/46.2 . . . HWT&L is 7% faster
. . . . . . . HighDetail: 25.4/22.5 . . . HWT&L is 13% faster
. . . . . . . Note: This may not be my gaming resolution for this awesome game.

. . . . . 1280x1024x32
. . . . . . . Low Detail: 39.9/41.4 . . . SWT&L finally surpasses HWT&L by 4%
. . . . . . . HighDetail: 21.5/20.3 . . . HWT&L by 6%
. . . . . . . Note: This is the transition point between HWT&L and SWT&L for my setup. Notice also
. . . . . . . that HWT&L when the details is higher still over takes SWT&L.

. . . . . 1600x1200x32
. . . . . . . Low Detail: 28.6/30.9 . . . SWT&L is 7% faster
. . . . . . . HighDetail: 15.7/15.9 . . . SWT&L is 1% faster
. . . . . . . Note: Finally SWT&L takes the lead in both cases.

Conclusion: In the playable resolutions of this game T&L makes a significant performance increase but demishes quickly after 1023x768x32 but still significant at 1152x864x32 in obtaining better performance.

My next post will explore the effects of AGP4x and AGP2x on this high poly game test.

Edited by noko on 04/15/01 01:28 PM.

Guest · Apr 15, 2001

good work noko...

but kyro don't need even a agp bus... it would do just well on a pci ... because of the Tile base rendering there are less triangles need to feed the card, because of the "z-buffer" it will never get out of local memory even in higher depths, because of the 8-layer support it will be required to feed this card with the triangles only once in a 8 cycle time with 8 layer games...
😉
yes T&L is a good feature that would help any card at lower depths...
kyro could do just fine without that T&L, but if he gets T&L (kyro 3 ) it will benefit kyro 3 in ALL depths ALL !!

Guest · Apr 15, 2001

Yeah your right I can't afford a geforce 3, oh and why do you assume I mean owners of Nvidia cards when I say Nvidia drones?, I've owned a TNT2 and a Geforce DDR and I don't consider myself an Nvidia drone, I was referring to people that just won't listen to anything thats not totally praising Nvidia, they just don't want to hear it unless your saying what they've been lead to beleive by King Nvidia, now you obviously consider yourself to be an Nvidia drone or you wouldn't have posted such a ill thought out post, after all for all you know I could have been a millionaire that could buy as many Geforce 3's as I wanted, even if I did buy a Geforce 3 why should that change my opinion of the technical aspects of any other graphics card?

noko · Apr 15, 2001

Well when you are doing Software T&L then the AGP bus does come into play and it would affect the performance of not only the Kyro but any AGP optimized card. On the pci bus there will be contentions with everything else, USB, sound card, ethernet etc. AGP it is a much faster direct port to the cpu which doesn't have to wait in line for access. Now I redid the benchmarks above but used AGP 2x instead by reconfiguring my bios and confirming AGP was at 2x by using PowerStrip3.0 build 125 and Wcpuid Version 3.0 by H-Oda. Seeing if the slower AGP speed would affect software T&L. Well this is what I found:

TESTS:(All tests conducted in W2k with ATIs official 3102 drivers, Radeon at stock 183/183 clocks, 3dMark2001.)

. . . Game 3, MAX PAYNE, AGP 2x

. . . . . 1024x768x32
. . . . . . . Low Detail: 58.3/50.8 . . . HWT&L is 15% faster
. . . . . . . HighDetail: 28.8/23.9 . . . HWT&L is 21% faster

. . . . . 1280x1024x32
. . . . . . . Low Detail: 40.2/41.2 . . . SWT&L is 2% faster
. . . . . . . HighDetail: 21.4/20.4 . . . HWT&L is 5% faster

Conclusion:
There was no real significant difference in performance running in AGP 2x versus 4x. Appears that AGP 2x speeds are sufficient for this Game test and was not the bottleneck. At the middle to lower resolutions the CPU is limiting when doing SW T&L while at the higher resolutions the bandwidth of my video card became more limiting for HW T&L. I also conclude that the Kyro 2x AGP limitation will not adversely affect its performance in games (at least in Max Payne). Actually a higher speed cpu could transfer more T&L information than my 864mhz cpu in which case the 2x AGP speed may become the bottleneck. I have no way in testing this to see if that would be the case or not.)

Next series of tests I will degrade my CPU speed down reflective of a lower speed systems, something which the KyroII is targeted at, showing the T&L difference between the CPU and GPU.

Edited by noko on 04/15/01 06:40 PM.

Guest · Apr 15, 2001

Nice work noko, actually the Kyro II is a AGPx4 chip, Hercules have limited there board to 2xAGP because some ALI mobo's apparently have instibility with the Kyro II in AGPx4 mode, personaly I think hercules should have left the choice of AGP modes to the user so that people without ALI mobo's could use AGPx4, as I said Hercules have limited there board to AGPx4 but this doesn't mean every board maker will choose to do this, the box for the Videologic Vivid!XS (Kyro II) has AGP x2/x4 written on it (I can give you the pic if you want) and the specs on there site say the same so it looks like there Kyro II board is AGPx4, the Kyro 1 board that I'm using right now is in AGPx4 mode and the Kyro II is simply a Kyro 1 on a 0.18 micron process and a slight redesign (the addition of 3 million transistors) to help with the higher clock speed.

I'll do some similar tests on my Kyro soon, could you do some FSAA benches in that game?, because I only have a 14" monitor so I can't go above 1024x768x32 in my benches.

phsstpok · Apr 15, 2001

Interesting.

I have a question. (My ignorance is really going to show). What is the relationship between FSB speed and AGP speed? I understand that AGP 1X is normally 66 mhz. 2x and 4x are then obvious. Here is where I am confused. This implies the AGP 4x would theoretically be 4 x 66 mhz or 266.7 mhz (66 mhz is really 66.6666...). How can AGP operate this fast if FSB is only 133 mhz?

noko · Apr 15, 2001

Thanks for the update, it really doesn't appear that AGP 2x/4x makes that much of a difference. Since I don't have access to a Intel platform (noted better AGP) then obviously I can't be totally accurate. I Will do FSAA test as you requested also remember the Radeon is the worst card for performance hit in doing FSAA. Something which the KyroII shines at. I am pretty confident to say that the KyroII card besides the GF3 card would be the best card period for flight simulators due to its superior FSAA.

Edited by noko on 04/15/01 03:39 PM.

noko · Apr 15, 2001

Actually the AGP bus stays at 66mhz and is is doubled pumped at 2x and quad pumped at 4x. Kinda like DDR ram. Max transfer rates are as follows 1x - 256mb/sec - 2x - 512mb/sec - 4x - 1024mb/sec directly to and from the cpu. Now the max transfer rate of the pci bus is 128mb/sec which is shared with IDE controller, usb, com ports, serial port, sound card, ethernet etc...

phsstpok · Apr 15, 2001

Thanks but now I am even more confused.

IF AGP is clocked at 66mhz and double-pumped, how is the effective data rate 256mb/sec? (By the way, is that megabits or megabytes?)

I still have a problem with my original question even though my facts may be in error. If the FSB is clocked at 133 mhz and double pumped at 266 mhz, what is the maximum rate which data can effectively passed over the AGP bus. It seems to me that FSB limits everything. What is the data rate of the FSB relative to the AGP's data rate, I guess is the question I am really getting at?

Guest · Apr 15, 2001

I've read all the posts in here (I know, I have too much time on my hand) and I didn't think anyone here is totally obsessed with nVidia. The only person that was totally obsessed was powervr2, when he was claiming the kyroII is the king of the crop and the "killer of nvidia". He started to reason later.

You too were also making some claims like nVidia and ATI having no HSR, and the GF2 and Radeon T&L becoming redundant in Microsoft Direct X 8, until noko and holygranade said it wasnt true.

I dont blind my self from other peoples views however dim they may be. I dont think I am one of those so called drones. With these lables, you are becomming just as bad as AMDMeltdown and fugger.

You are right. You could have been a millionaire. But, I was making my assumptions from what I read here. You were not showing appreciation for technology but a obsessive passion for the cheaper option.

Who is General Failure and why is he reading my hard disk?

noko · Apr 15, 2001

Sorry for the confusion, at 1x speed it is (66mhz x 32bits) x byte/8bits = 264mbytes/sec (small correction to max transfer rate above.)
2x speed is twiced that 524mbytes/sec (2x is doubled pumped)
4x speed is four times, 1056mbytes/sec (4x is quad pumped)
I hope this clear things up. At Intel if you plug in "AGP" in there search engine you can bring up the AGP white papers which goes into great detail about AGP.

All right the FSB is the CPU interface speed with the motherBoard and usually what the memory runs at but not always since Via allows you to run a FSB of 100mhz while the memory can run at 66mhz, 100mhz or 133mhz lets say. Also Via allows a FSB of 133 while the memory can be configure to 100mhz with the KT133A chipset. While the AMD CPU FSB maybe 100 or 133mhz its effective speed is twiced that, 200mhz or 266mhz due to being doubled pumped itself. Confusing, yes but not to hard to understand. Now the FSB with the via chipsets can be tweaked by some motherboards from 100mhz to 200mhz in 1mhz increments. Like my IWILL KK266 mobo. More confusing, yeap. Can someone please clear this up better or clean up my ugly explanation.

The pci bus runs at 33mhz normally untill you start to overclock the front side bus.
AGP runs at 66mhz normally until you start to overclock the front side bus.
Edited by noko on 04/15/01 04:48 PM.

Guest · Apr 15, 2001

I started to reason later ???
that title of this topic gived more than 1000 views....

do you think that a card pricier and with lower bandwidth is a good thing ???

noko · Apr 15, 2001

I hope we can keep this objective and show the strengths and weeknesses of the KyroII. KyroII maybe the best card for someone while another card would do better for someone else. Onto more tests....

Edited by noko on 04/15/01 04:56 PM.

Guest · Apr 16, 2001

No I wasn't making claims that the Radeon and Geforce HW T&L unit would be redundant until noko and holygrenade came along. I never changed my opinion and I still have the same opinion as my first post, it happened that holygrenade misunderstood what I wrote (which was partly my fualt). What I said was that the Radeon and Geforce is not anymore DX8 compliant then the Kyro II, and I was saying that when developers fully embrase DX8's vertex shaders those games will need to be transformed and lit by the system CPU and the Geforce and Radeons hardwired hw T&L unit will be useless unless those games also incorperate a hardwired T&L engine in the game. Which they won't do for long IMO with all new dx8 compliant cards being released.

Also the Geforce doesn't have HSR at all in the sense of getting rid of hidden surfaces before they've been textured and shaded. The Radeon has a limited version of HSR (which I originally didn't know it had) that requires the game to be rendered front to back, e.g. say there is a monster hidding behind a wall, if the game is rendering back to front which is the standard way its done, it first renders the monster and then it realises that the wall is infront of the monster so it renders the wall overtop of the monster because obviously the wall is what your going to end up seeing on the monitor and the card had no way to know that the wall would be infront of the monster when it was rendering the monster. Now if its rendered front to back it first renders the wall and then when it comes to the monster it realises the monster is behind the wall and so doesn't render the monster. Its still not 100% efficient even if the game is rendered in this way because it only performs per polygon HSR not per pixel HSR. It cuts the scene its working on up into tiles and stores a z-value from the polygon in that tile thats furthest away from the screen, so if its working on the wall first it will know latter that the monster is further away from the screen then the wall and so it knows that the wall should be infront of the monster and it doesn't need to render the monster. But if small polygons are used the card cannot be sure if any 1 polygon is representative of that whole tile so in this case it doesn't enter the value for that tile into the low res z-buffer, it leaves this tile to the normal res z-buffer and the tile is rendered just like a Geforce would (basically render everything and then throw it away afterwards), so its allot less then 100% efficient. In some games it can be 20-30% efficient or maybe a little more and in some games its 0% efficient (although the other hyper-z features like z-compression will still help in these games), but anyway at least it saves some overdraw.

Now with the Geforce it doesn't matter which way around the game is rendererd its always going to draw both the wall and the monster and with the Kyro II again it doesn't matter which way the game is rendered its only going to render the wall and none of the monster will be rendered because of its per pixel HSR.

I never said there were any Nvidia drones in this thread, I wasn't pointing that comment at anyone in this thread.

too OzzieBloke, I'm glad you enjoy my posts, BTW what are carriage returns?, I'll be glad to put them in in any further posts if you tell me what they are😱)

noko · Apr 16, 2001

Now configuring my CPU speed to a more typical system that a Kyro owner would have I did the following:
-Changed my cpu multiplier to the minimum of 5x and since I have my FSB configured to 133mhz my cpu speed became a 665mhz T-Bird. 200mhz slower then the previous tests. This system would be approximately as fast as a 700-750mhz Duron with pc133 ram. Still much faster than the typical 500-800 celeron system out there but good enough to see the affects of a lower speed cpu on comparing HW T&L to SW T&L on a modern day DX8 game. I also went back to AGP4x which really shouldn't make a difference on these tests. I included the normal lower resolutions because a lower system would probably be using them exclusively as the tests will indicate.

TESTS:(All tests conducted in W2k with ATIs official 3102 drivers with AGP at 4x transfer rates, 3dMark2001.) CPU speed lowered to 665mhz.

. . . Game 3, MAX PAYNE AGP4x

. . . . . 640x480x32 (HardWare T&L)/(Software T&L). . . . HWT&L/SWT&L
. . . . . . . Low Detail: 70.3FPS/48.0FPS . . . HWT&L is 46% faster. OUCH!!
. . . . . . . HighDetail: 29.3FPS/21.3FPS . . . HWT&L is 38% faster. Whosh!!

. . . . . 800x600x32
. . . . . . . Low Detail: 63.4/46.8 . . . HWT&L is 35% faster. Creamed
. . . . . . . HighDetail: 28.2/20.9 . . . HWT&L is 35% faster. Creamed again!!

. . . . . 1024x768x32
. . . . . . . Low Detail: 53.0/44.4 . . . HWT&L is 19% faster.
. . . . . . . HighDetail: 25.8/20.7 . . . HWT&L is 25% faster.
. . . . . . . Note: Even at this higher gaming resolution the HW T&L is kicking some 19-25%
. . . . . . . major butt. Lets just jump to 1600x1200x32.

. . . . . 1600x1200x32
. . . . . . . Low Detail: 28.4/30.5 . . . SWT&L is 7% faster.
. . . . . . . HighDetail: 15.5/15.3 . . . HWT&L is 1% faster.
. . . . . . . Note: Even at this very high resolution when the complexity goes up the T&L engine
. . . . . . . can keep up with the overly loaded cpu.

Conclusion:
With advance game physics such as particle physics which Max Payne uses the CPU is heavily loaded to begin with. Then to stack on the software T&L calculations on top of that brings the CPU to its knees, adversely affecting the frames per second in this modern DX8 game. I conclude unless you have a 1.1ghz or greater processor, games like Max Payne will greatly benefit using hardware T&L.Edited by noko on 04/15/01 10:05 PM.

phsstpok · Apr 16, 2001

Thanks again Noko. I am starting to get a better feel for this whole thing. It is much easier when one looks at data rates as opposed to bus frequencies.

I hope someone responds with info on FSB data rates and answers my inquiry, "Does a 133 mhz FSB (266 mhz DDR) limit the throughput of the AGP bus?" I'm wondering if this might be the reason we don't see any signficant difference between AGP 2X and AGP 4X. The old argument of "there just aren't a lot applications which fully utilize AGP 4x" doesn't sound correct any longer. There are plenty of benchmarks and games with high polygon counts which should be significantly improved if AGP 4x was truly twice as fast as AGP 2X. Yet, we are not seeing this. I'm just trying to find an explanation that I can believe.

Thanks again.

Guest · Apr 16, 2001

yes. With the xbox nvidia was able to get rid of the bus system of computers and that is why the geforce3 in the xbox will do better. The perfect world would to have on motherboard with everything integrated into it. of course, you couldn't upgrade or anything like that so we stick to the separate components.

phsstpok · Apr 16, 2001

I guess, in a "perfect world", the CPU, GPU and memory would all be on one chip leaving a slower bus just for peripherals. For big performance gains we would just replace the one chip. No worries about which of the three is the bottleneck. All three would always be optimized as a set.

I can dream can't I?

Kyro 2 the killer of nvidia ???

Distinguished

Guest

Guest

Guest

Guest

Guest

Guest

Guest

Guest

Guest

Guest

Distinguished

Distinguished

Guest

Guest

Guest

Guest

Distinguished

Guest

Guest

Splendid

Distinguished

Distinguished

Splendid

Guest

Guest

Distinguished

Guest

Guest

Distinguished

Guest

Guest

Distinguished

Splendid

Guest

Guest

Splendid

Share this page