Kyro 2 - Part 2

HolyGrenade · Apr 30, 2001

more like 43% to 44%.

"My name is Ozymandias, King of Kings:
Look on my works, ye Mighty, and dispair!"

noko · Apr 30, 2001

Another strong point for the Kyro2 is the FSAA. The performance hit is not as bad as for the GF2s or Radeons.

Edited by noko on 04/30/01 11:13 AM.

Guest · Apr 30, 2001

I've been away for a few days thats why I haven't participated here for a while, anbyway here goes.

<<<<<<<<Once again is the Kyro2 sacraficing quality for speed that is beyond what is significant>>>>>>>>

I have to say Noko this is as far from the truth as possible (not saying your lying just saying your very wrong here). Kyro II does exactly the opposite of what your saying in every case. Where have they sacrificed IQ for speed?, even the anisotropic is 16 tap. If they'd wanted to sacrifice IQ for speed they'd have used 8 tap anisotropic and it would have been in 1 cycle and almost free baring texture bandwidth issues. What about Internal true colour?, it sacrifices speed for image quality. Because of Internal True Colour the Kyro II's 16bit IQ is almost exactly the same as Geforce and Radeons 32bit. If they hadn't used this method then Kyro II would be loads faster when running in 16bit mode so I ask again where have they ever made IQ cutbacks to get better framerate?

Lets look at filtering for a second. The reason the Kyro II loses so much speed when going from bilinear to trilinear is because of texture bandwidth, I know this to be a fact because the Kyro II can do 2 trilinear filtered pixels in 1 cycles so it doesn't take a fillrate hit when going from bilinear to trilinear. Thats why TC makes it free because it lowers texture bandwidth. Its 32bit were trilinear takes a big hit over bilinear and also anisotropic is allot faster in 16bit because the texture banwidth is halfed in 16bit over 32bit. So with the Kyro II you can run at 16bit, get great 32bit like image quality and half texture bandwidth meaning higher filtering won't lose hardly any performance (Anisotropic will still lose allot because of 2 cycles but nowhere near as much as that bench shows). Then add TC to the 16bit rendering and trilinear is totally free and even faster then bilinear in some cases and anisotrpic loses about 50% because of halving the fillrate which is still allot but would still see the Kyro beating the GTS and radeon in those GP benches because in 16bit with TC there'd be just about no texture bandwidth issues so anisotropic would be at most half the speed of those bilinear scores because of its taking 2 cycles instead of 1 for biliear and trilinear.

IMO 16bit benches between the Kyro II and other cards aren't even fair. Radeon, GTS, and all the others are sacrificing image quality BIG time for extra speed, although Radeon doesn't seem to gain any extra speed for 16bit, why is that? Its 16bit rendering is terrible compared to its 32bit so why no performance advantage for 16bit over 32bit?

Also for the record those anisotropic scores are strange and don't hold true to other reviews I've seen showing Kyro II with anisotropic filtering. It does lose allot with anisotropic but it shouldn't lose anywhere near that sort of performance. Personally I think that result was a mistake as it just doesn't fit.

m_kelder

<<<<<<The card isn't even out yet and problems with it are already being found.>>>>>>>

Your totally exagerating this, its not a fualt, merely a weekness and this weekness can be remedied with 16bit and forced texture compression. The Geforce isn't yet released and it has worse fualts IMO. Its 16bit mode (just like all cards but the Kyro and Kyro II) is unusable because it simply looks awefull, obviously you can goto 32bit with the Geforce 3 or Geforce 2 or Radeon but then you could say that 16bit benches for those cards are pointless because there sacrificing to much quality for extra speed.

<<<<We’ve already seen that the chip can’t be run any faster than it can now so we won’t see an ‘ultra’ version. Looks like we’d have to wait and see if kyro III can perform at the high end. It is sad, but 90% of the people out there only care about the highest frame rate. That is why nvidia has the market share they do. The kyro II is nice but when it comes down to it people will pat STM on the back and say good job but then turn around and by a geforce2 or radeon instead.>>>>>

I already pointed you guys to the Beyond3d review which clocked a 32mb Vivid!XS Kyro II from 175mhz too 190mhz and this board doesn't even have a fan, it only has a passive heatsink. So there's no way you should judge the Kyro II chip on Herc 4500 boards that aren't final.

Edited by Teasy on 04/30/01 02:45 PM.

noko · Apr 30, 2001

So are you saying to use the superior method of filering (Anisotropic) the Kyro2 to get any kind of performance must go to 16bit and then that is not sacraficing quality for FPS???? It is a design flaw and you might as well admit it.

Why havn't AnAndTech produced the second review as promised some time ago? Well here is why:

As I mentioned at the start of this editorial, the Kyro II isn’t without its problems. We have been working closely with Hercules to get the latest driver set and board revision of their Kyro II based 3D Prophet 4500 unfortunately we haven’t seen anything since our original review of the card. We are holding off on a follow-up until we get a final board as well as final drivers since there are a number of issues with the current hardware.

For starters, compatibility with i815 platforms seems to be relatively poor with the board we have as well as boards that have been sent to other reviewers. We would experience 2D artifacts and random lockups on our i815 test beds, regardless of the motherboard used. Even using Intel’s own i815 board did not solve the problems, although we did not experience any of these on our Athlon test bed. It was a bit of a shock to see this, especially after remembering the days when AGP cards would have compatibility problems with AMD systems yet worked perfectly with Intel systems. How times have changed, luckily it only seems prevalent with the pre-release Hercules 3D Prophet 4500 that we have in lab. Hercules has assured us that they are working on fixes for these problems, although until we see a final board we’re going to hold off on accepting that statement just yet.
<A HREF="http://www.anandtech.com/showdoc.html?i=1461&p=5" target="_new">http://www.anandtech.com/showdoc.html?i=1461&p=5</A>

Well wel, wait until the board hits the street there buddy and the real issues start to be broadcast onto the web. Right now I cannot recommend a Kyro2. If you have a Intel 815 chipset forget it, save yourself the nightmare. What other chipsets won't work with the Kyro?

Edited by noko on 04/30/01 03:53 PM.

HolyGrenade · Apr 30, 2001

Is it true that the Kyro II's sent for review are all 175MHz with 5ns memory while the release ones will be 166MHz with 5.5ns memory?

"My name is Ozymandias, King of Kings:
Look on my works, ye Mighty, and dispair!"

noko · Apr 30, 2001

I am beginning to believe that the Kyro2 will be further delayed due to the hardware problems noted above. We will see.

noko · Apr 30, 2001

The released versions will be 166mhz and not the 175mhz the reviewers got and tested??? This is getting funnier by the moment. Obviously the reviewers got hand picked overclocked Kyro2's. So the real performance of the Kyro2 will be less than what has been tested on many sites. This will not sit to well with some people after being falsely lead by benchmark results. I just hope you are wrong and the 3dProphet 4500 is clocked at 175mhz. Still no one has reported or reviewed any final boards with the Kyro2 so whats up with that? Obviously the many problems are not fixed.

I am not sure of the release versions of the Kyro2 166 or 175mhz. Plus I missread your post, sorry about that I think PowerVR cleared it up below.Edited by noko on 04/30/01 06:32 PM.

Guest · Apr 30, 2001

you could be right on the memory, but kyro II will be 175 mhz !!!

Guest · Apr 30, 2001

nah kyro 2 will be 175 mhz for sure !!!
please holygrenade give us a link that says otherwise !!

about kyro design flaws :
kyro 16 bits is like geforce 2 at 32 bits (take my word on that), so there will not be a quality/performance compromise !

Nevertheless I think the texture compression would be enough for good performance in 16 anisotropic filtering...
😉

noko · Apr 30, 2001

No I messed up my reply and hoygrenade never said that the Kyro2 will be at 166mhz. Maybe the non Hercules will be at 166mhz.

HolyGrenade · May 1, 2001

I read somewhere, I can't remember, may be some website or magazine (I'll try to find it), that the first samples created for developers and reviewers was released in a rush. They had to make some last minute changes which included increasing the clock from 166MHz to 175MHz.

Also, there was some rumours about it not working properly at that frequency. I'm not sure what the problems are and whether they were just rumours. Thats why I was asking. I thought one of you might be able to fill me in.

"My name is Ozymandias, King of Kings:
Look on my works, ye Mighty, and dispair!"

Guest · May 1, 2001

even videologic will be at 175 mhz !!!
(I am not sure that they will put a fun like hercules)

noko · May 1, 2001

Thats good. Now are there any reviews based on the actual cards that will be shipped supposenly this month?

Edited by noko on 04/30/01 05:30 PM.

Guest · May 1, 2001

you are right, initially it was intended for 166 mhz but for boards without a fun (only a heatsink), hercules tested kyro 2 and saw potencial for 175 .
but hercules loves stability and because of this they will put a heasink coupled with a fun, this decision will increase the price, but it's ok for me..
😉
there are no problems at all with kyro 2 at 175 mhz !
teasy already said that a videologic kyro 2 with a heatsink without a fun can get to 190 mhz !
So there is no issue at all.

Guest · May 1, 2001

wait some more days!
(maybe a review is possible this week)
<A HREF="http://www.beyond3d.com" target="_new">http://www.beyond3d.com</A> have an hercules final card...

noko · May 1, 2001

Finally!!!! Maybe they will go on sell this month. I hope so, it sure did scare nVidia and I am sure ATI as well.

Guest · May 1, 2001

Some of your comments are getting icreasingly bias against the Kyro II. Sweeping statments like this one "If you have a Intel 815 chipset forget it, save yourself the nightmare. What other chipsets won't work with the Kyro?". (BTW don't you mean !815?, there are other 815 boards like 815E, the beyond3d review used this chipset with its Kyro II and had no problems at all.)

Go to any message board like this one and you'll see loads of people saying "my Geforce, Radeon, Kyro, whatever isn't working on this mobo", this doesn't mean there's some wide spread problem, and you must know this for youself. The other day someone with almost the exact same system as me could not get the Kyro 1 to work does that mean that the Kyro 1 has a problem with my system?, no it was most likely some other problem in the system and for all you know thats whats happening here or it could be a problem with his card. I have never heard this problem talked about in any other review "luckily it only seems prevalent with the pre-release Hercules 3D Prophet 4500 that we have in lab", a problem with one card and already your saying the card is a nightmare for this system which is a rediculess attitude to have and is fanboyish IMO.

Are you seriously suggesting that the anisotropic hit is somesort of fualt that can be fixed either in the drivers or hardware? because if you are your dead wrong. I'm not trying to deny that the Kyro II takes a large hit for anisotropic because that would go against its design, clearly looking at the specs it need 2 cycles per pixel for anisotropic and therefore it loses half it fillrate. Since the card uses just about its full fillrate cutting its fillrate in half shows in game performance. Although those benches don't seem to fit the loss Kyro II should take, I can only assume thats a texture bandwidth problem, what else could it be after all.

<<<<So are you saying to use the superior method of filering (Anisotropic) the Kyro2 to get any kind of performance must go to 16bit and then that is not sacraficing quality for FPS>>>>

No I gave information for a remedy that would help anisotropic filtering performance, either use TC or 16bit or both for even better performance, have you seen 16bit on the Kyro II, its hardly much of a IQ loss going from 32bit to 16bit on the Kyro II.

<<<<It is a design flaw and you might as well admit it.>>>>

Noko I'm sorry but you don't know what your talking about here, a design flaw is an error, something that went wrong when designing the chip, something that wasn't supposed to happen. There's no design flaw causing the anisotropic hit. Its merely a fillrate and texture bandwidth problem anyone can see that by looking at the Kyro II's specs and its design. Unless you'd like to actually explain yourself rather then stating that *fact* that I must apparently admit without any real technical facts backing it up.

noko · May 1, 2001

Kyro2 does 2 trilinear filtered textures in one clock in hardware per pass (GOOD!) but by design it has to use two clocks per each anisotropic filter texture per pixel (BAD!!!)*, 4 textures would require 8 clock cycles for each anosotropic filtered pixel, 6 textures 12 clocks! and 8 advertised textures would use a incredible 16 clocks!! per pixel. Yeah man, thats not a design flaw, right!! Considering the rather limited fillrate the Kyro2 has to begin with how is it going to perform when doing 4-6 anisotropic filtered textures?? Bad, really bad is the answer. No, thats not a design flaw, hell yeah it is!! Do I need to go back to fillrate calculations to show the dismal fillrate limitation that the Kyro2 design imposes when using anisotropic filtering?

*Note: the funky way that the Kyro2 does multiple textures of up to 8 textures per pass doesn't help, each texture requires an additional clock causing the texture fillrate ability to decrease as the number of textures goes up.

AnAndTech is the one describing the problems they noted, obviously the 815 motherboards are not faring to well, what makes the 815E's any different? If the problem was fixed wouldn't Hercules immediately respond and let AnAndTech know that the problem was fixed? Maybe they just don't care (which I doubt) or more likely the problem is still not fixed.

Edited by noko on 04/30/01 09:06 PM.

noko · May 1, 2001

Yeap! the design of the Kyro2 has an inherit design limitation (flaw) when Anisotropic filtering is used. Maybe an after thought to incorporate that ability for what ever reason, in use it will limit the Kyro2 performance dramatically. Like in 1/2 the speed when used*, now that is pretty dramatic. Now if you want me to talk about the Radeon design flaws such as in Pixel Shaders, 16 bit dithering hardware design or lack of to make you feel better fine.

*The Kyro2 design requires twice the number of clock cycles to do Anisotropic Filtering, meaning 1/2 the output. No other cards at Ace's Hardware exhibit the dramatic decrease in performance when performing Anisotropic filtering. When the Kyro2 was tested at Ace's hardware it didn't drop the madatory theoretical 50% of its design going from Tri to Anisotropic filtering but 77%! The Radeon dropped roughly 1% going from Tri to Anisotropic filtering the best of the bunch. Obviously the Radeon Anisotropic filtering capability was properly designed and implemented in the chip. Sounds like a pretty big judgement error in designing the Kyro2 here, meaning design error. Edited by noko on 04/30/01 09:37 PM.

Negaverse23 · May 1, 2001

<<you could be right on the memory, but kyro II will be 175 mhz !!!>>
I remember before that Hercules increased (overclocked)the core clock speed to 175Mhz from 166Mhz to improve 2D quality. But that was under the 5ns memory. Can the Kyro II still run at 175Mhz if the memory clock and core clock is synchronous under the 5.5ns memory? (175/175Mhz)

=
Are you just lazy or incredibly stupid? -King of the Earth

Guest · May 1, 2001

Look at this:
<A HREF="http://www.aceshardware.com/Spades/read.php?article_id=25000228" target="_new">http://www.aceshardware.com/Spades/read.php?article_id=25000228</A>

The ATi Radeon does very well, thanks to its Hyper Z technology, and at 225 MHz (an SE version will apparently clock at 230 MHz) with trilinear filtering disabled, it is almost as efficient as the Kyro at 175 MHz. With trilinear filtering enabled, however, the Radeon's performance takes quite a dive. It seems this is because the third texture unit is no longer available to apply the third texture layer, and as such, performance drops dramatically. The GeForce 3 is clearly more efficient than the GeForce 2, but the ATI Radeon's Hyper Z seems to save more bandwidth. It is possible, of course, that Villagemark, being a somewhat older benchmark, is not able to use the full potential of the GeForce 3. 

hum it looks that radeon have a hit in trilinear ... maybe even more with anisotropic (for sure) if the fill rate is limited instead of bandwidth.
interesting to see this...

Obviously the Radeon Anisotropic filtering capability was properly designed and implemented in the chip

😉 it have 3 TMU unlike 1 TMU of kyro 2!!!
thus kyro 2 needs more clock cycles.
obviously it is not a design problem.
it a design option!
they could go into the brute force way... killing the smart SMART DESIGN !!!

Sounds like a pretty big judgement error in designing the Kyro2 here.

like radeon in trilinear (more on anisotropic)! when in a fill rate intensive game?

 With trilinear filtering enabled, however, the Radeon's performance takes quite a dive

half the fill rate on a complex game (4-5 overdraw, on future games) is more than enough to beat the "traditional" cards ...

don't judge the performance of kyro 2 anisotropic filtering by one benchmark !
we could do the same in the trilinear/anisotropic filtering on radeon ...

Edited by powervr2 on 05/01/01 05:24 AM.

Guest · May 1, 2001

I remember before that Hercules increased (overclocked)the core clock speed to 175Mhz from 166Mhz to improve 2D quality. But that was under the 5ns memory. Can the Kyro II still run at 175Mhz if the memory clock and core clock is synchronous under the 5.5ns memory? (175/175Mhz)

kyro only works synchronous with the memory !
yes with 5.5 ns it is possible to go into 175 mhz clock... if not they have to put 5 ns ...

HolyGrenade · May 1, 2001

In the review you posted, if you look at the comments regarding the Anisotropic filtering in Formula 1, you will no doubt see this line:

"In the case of the KYRO II, I have to wonder whether this was a good choice, as performance simply hits rock bottom."

especially when the performance graph looks like <A HREF="http://www.aceshardware.com/articles/reviews/GF3vsK2/F1A1333Filtering.gif" target="_new">this</A>. And this is running on an Athlon t'bird 1.33GHz. Needless to say, one of the fastest platforms available.

Also, the benchmark showing the <A HREF="http://www.aceshardware.com/articles/reviews/GF3vsK2/GiantA1333_32bit.gif" target="_new">minimum framerates in Giants</A>, shows the kyro not doing too well. I thought you said the Kyro 2 will always have a better minimum framerate than the others, but it doesn't seem so.

Also, this article says that the Kyro 2 bump mapping in this game was added later via a patch, because otherwise the game will only apply this to T&L cards. This is an indication of things to come. Developers are starting to put more emphasis on the T&L features of games over the normal ones.

Finally, Kyro 2 isn't the only card to have a fixed internal colour processing of a precision of 32 bits. The Radeon does the same.

"My name is Ozymandias, King of Kings:
Look on my works, ye Mighty, and dispair!"

Guest · May 1, 2001

Also, the benchmark showing the minimum framerates in Giants, shows the kyro not doing too well. I thought you said the Kyro 2 will always have a better minimum framerate than the others, but it doesn't seem so.
first we must not forget that he used directx8 (there is a bug to be corrected by microsoft on their's next directx8 release)!
Teasy with their's kyro I have more 10 fps in directx7 in giants !

especially when the performance graph looks like this. And this is running on an Athlon t'bird 1.33GHz. Needless to say, one of the fastest platforms available.
the problems were not because of the system...
the problem is an evidence that this game processes many textures so kyro 2 texture bandwdith is streessed to a point that 16 tap anisotropic would get a hit.
(I already mention the solution to this on kyro 2)

Also, this article says that the Kyro 2 bump mapping in this game was added later via a patch, because otherwise the game will only apply this to T&L cards. This is an indication of things to come. Developers are starting to put more emphasis on the T&L features of games over the normal ones.
that doesn't make any sense, only apply bump mapping to T&L cards? what have this bump mapping to do with T&L?
😉

Finally, Kyro 2 isn't the only card to have a fixed internal colour processing of a precision of 32 bits. The Radeon does the same.
So why the bad picture quality of radeon in 16 bits ?

Edited by powervr2 on 05/01/01 05:49 AM.

HolyGrenade · May 1, 2001

Your comment regarding the missing Kyro bump mapping in Giants:

that doesn't make any sense, only apply bump mapping to T&L cards? what have this bump mapping to do with T&L?
😉

This is the information to which I was referring. This came from the <A HREF="http://www.aceshardware.com/Spades/read.php?article_id=25000229" target="_new">link</A> you provided.

Giants now supports bump mapping on Kyro II when using the latest beta patch, but this was not possible on the demo I used. Giants normally uses the T&L unit for bump mapping and if no T&L unit is available, bump mapping is disabled.

Well, Haven't I allways said there are a lot of duplications when creating something for both the GPU and the CPU. I guess more people have already started to concentrate more on the GPU than the CPU. So to answer your question, any complex function has a lot to do with T&L.

"My name is Ozymandias, King of Kings:
Look on my works, ye Mighty, and dispair!"

Kyro 2 - Part 2

Distinguished

Distinguished

Guest

Guest

Distinguished

Distinguished

Distinguished

Distinguished

Guest

Guest

Guest

Guest

Distinguished

Distinguished

Guest

Guest

Distinguished

Guest

Guest

Guest

Guest

Distinguished

Guest

Guest

Distinguished

Distinguished

Distinguished

Guest

Guest

Guest

Guest

Distinguished

Guest

Guest

Distinguished

Share this page