Kyro 2 Ultra/ Kyro 3 Details

[bananaboy]

For those of you who don't know, the Kyro 3 will feature 4 rendering pipelines, hardware TnL, fabbed at .13 micron and will be clocked at 250mhz with DDR memory. All I can say is that is a Geforce 3 Ultra killer, *with the Kyro 3 supposedly coming out next year. With 3 times the performance of the Kyro 2 you can expect any non tilebased chips to be a thing of the past as I don't think its possible to squeeze any more bandwidth out of them.

The Kyro 2 Ultra, by whatever name its given will be clocked at 200mhz instead of Kyro2's current 175.
Source Link

Pretty much means the end of Nv and ATIs dominance once this thing is out unless they are already working on a tilebased architechure themselves.

 

[jobberd]

good to know that things are starting to really heat up in the video card market again. I cant wait to see what the other companies are planning in retort to this.

 

[Leon]

<< All I can say is that is a Geforce 3 Ultra killer, supposedly coming out next year >>



Your logic is flawed. GF3 Ultra coming this fall, NV-30 coming early next year....

Of course, all &quot;release dates&quot; are nothing more than speculation at this point.

Leon

 

[Rand]

It should also be kept in mind that this is still a rumor and not confirmed.

 

[bananaboy]

Its reminiscent of the day 3dfx came up. Nv always had mediocre products from what I remember (i was always partial to 3dfx over nv so beware), with no innovation just hype about whatever feature they supported (32bit color), not saying thats the case now (besides their near useless hardware TnL) but they didn't explode on the scene like 3dfx did. A smart marketing company but not revolutionary, more evolutionary.
I guess I get the same feeling from kyro I did from my voodoo back in the day. All hail the evil kyro!

 

[Remedy]

Where is your link leon?

 

[jobberd]

hes referring to Nvidia's 6 month cycle. every 6 months they bring something new on the market. once the tnt2 came out, 6 months later the tnt2 ultra came out. 6 months after the geforce2 gts came out, the geforce2 ultra/mx/pro came out.

 

[BFG10K]

All I can say is that is a Geforce 3 Ultra killer

<rolleyes>

Let's not get so dramatical about it. After all, the Kyro2 was supposed to be a GF2 Ultra killer but it ended up being approximately equal to a regular GTS.

 

[PotNoodle]

AFAIK nv25 is slated to come in at .13um silicon processes, and there are currently issues surrounding that process from TSMC, this being tha case don't be surprised if nv25 doesn't make it out until 2002, and not late 2001.

NVIDIA has reached the point that their need for smaller silicon processes has outstriped the development of them - this was already evident from the delay to NV20 (it was supposed to be here last year), and it will probably be so in the future. Its quite likely that the fall refresh this year will not be another chip but an 'Ultra' version of nv20 (GF3), much like they did with GTS.

BFG,

&quot;Let's not get so dramatical about it. After all, the Kyro2 was supposed to be a GF2 Ultra killer but it ended up being approximately equal to a regular GTS.&quot;

Was it? I tend to looks at it the other way around - it was designed to be an MX competitor, that has ended up looking like a GTS competitor. Img were never shooting for the high end with it.

 

[PotNoodle]

However...

&quot;The Kyro 2 Ultra, by whatever name its given will be clocked at 200mhz instead of Kyro2's current 175.
Source Link

Pretty much means the end of Nv and ATIs dominance once this thing is out unless they are already working on a tilebased architechure themselves. &quot;

I wouldn't go that far just yet - the ST / PowerVR combination still has a long way to go to prove that they are serious in this business, and not just a one hit wonder. ST has 'dabbled' with video for ages (their previous relationship with NVIDIA being proof of this), however never have they sustained an 'attack' on the market place for any period of time. To my mind they need to have STG5500 and STG6000 series underway before we can truely state they are a real.

Also, I wish they wouldn't bother mucking around with 'Ultra' versions of KYROII; I'd far rather them do something serious and get STG5500 out the door, with full T&amp;L and other features, rather then buggering about on pointless KYROII refreshes.

I do think a KYROII Ultra would be a little pointless as well. I've had mine up to 190Mhz previously and the performance increase over the stock speed isn't much to write home about, so I have a hard time seeing the and extra 25Mhz or so for an 'Ultra' justifying the added RAM costs. KYROII is already transform limited in many cases, and an extra 50Mpps (or whatever it will be) is hardly going to make much difference with FSAA, so I have a hard time seeing how a KYROII Ultra will really get much benefit.

Like I said - Do something serious ST, get STG5500 to us.

 

[Pocatello]

If the Kyro III don't fully support DX 8, it's still behind the GeForce III. It's still amusing to see that people bashing nvidia's hardware T&amp;L, calling it &quot;useless in the future&quot;, while declaring the Kyro III's T&amp;L as a miracle. Static T&amp;L doesn't cut it anymore, that was hot 2 years ago (GeForce DDR).

 

[BFG10K]

I tend to looks at it the other way around - it was designed to be an MX competitor

I tend to think that the Kyro1 was supposed to be an MX competitor. After all, when the Kyro1 was released the MX was pretty much the standard in consumer 3D acceleration.

 

[PotNoodle]

BFG,

?I tend to think that the Kyro1 was supposed to be an MX competitor. After all, when the Kyro1 was released the MX was pretty much the standard in consumer 3D acceleration.?

I don?t think KYROI was supposed to be anything ? that was a foot in the door to see where things fall.

KYROII is less complicated than MX, based on the same silicon process (.18um) from the same manufacturer (TSMC) hence it wouldn?t surprise me if the chip unit price is actually lower than that of MX; its just the nature of its architecture that afford competition with GTS rather than MX (this is why the vendors like it so much ? they have much better margins). In all likelihood, because of the larger die size and hence wafer price, KYROI is probably more expensive per chip than KYROII ? the difference being though is ST produce the KYROI chip themselves and not TSMC, so they are probably running thinner on their own margins to it. It wouldn?t surprise me to see KYROII chips replace KYROI sometime down the line when stocks of ST?s .25um chips have depleted.


Pocatello

?It's still amusing to see that people bashing nvidia's hardware T&amp;L, calling it &quot;useless in the future&quot;, while declaring the Kyro III's T&amp;L as a miracle.?

Who?s doing that?

?Static T&amp;L doesn't cut it anymore, that was hot 2 years ago (GeForce DDR).?

If static T&amp;L doesn?t cut it then why are NVIDIA still basing chipsets on it (nForce) and saying that GF3 style T&amp;L won?t be integrated for some two years! Does nForce not cut it either???

IMO ?static? T&amp;L will see its hey-day in 2002 with titles such as Unreal 2 etc. Look at how long its taken to support static T&amp;L to any degree, and then apply that to vertex shaders. Xbox may accelerate Vertex Shader support a little for the PC, but developers still have to code to the available hardware, most of which is still likely to be MX level cards next year ? vertex Shader enabled boards are likely to be the vast minority for a few years yet.

 

[bananaboy]

If you do the math, there is no way even what comes out after the Geforce3 will have enough power.. correction, bandwidth. I dont think my statement about it being a Geforce3 killer dramatic at all. The kyro2 has a theoretical fill rate of 350megapixels per second, while the effective fill rate is 350mpps.. take any nvidia cards theoretical fill rate and roughly a 4th of that is what is usuable because of its overdraw.
Source link

So imagine a card with a effective fillrate of 1000mpps like the Kyro3.. were talking next year for this to, it will be a long time till the big NV can get that kind of bandwidth. Kyros tilebased architechure has proven itself to be real, while hardware TnL people are still debating about whether its going to be next year or the year after it starts helping.

 

[Pocatello]

Take a look of this post here
Link

 

[Soccerman]

LOL, BFG..

After all, the Kyro2 was supposed to be a GF2 Ultra killer

nope, it wasn't supposed to be anything! it was simply there.

I honestly can say I didn't see any market hype due to the manufacturers of cards using this chipset!

I tend to think that the Kyro1 was supposed to be an MX competitor. After all, when the Kyro1 was released the MX was pretty much the standard in consumer 3D acceleration.

partly true, this card was supposed to be released around the time we had Voodoo 3s and TNT2s, but Dreamcast took precedence over this, so I'm guessing they released Kyro 1 simply becuase they had spare chips or something.. It didn't sell too well I don't think.

If the Kyro III don't fully support DX 8, it's still behind the GeForce III. It's still amusing to see that people bashing nvidia's hardware T&amp;L, calling it &quot;useless in the future&quot;, while declaring the Kyro III's T&amp;L as a miracle. Static T&amp;L doesn't cut it anymore, that was hot 2 years ago (GeForce DDR).

hmm.. first do you know if the Kyro 3 has static T&amp;L?

I'm worried about the DX8 features too (not that I care for DX, it's just that people wine and moan about that crap.. Multiplatform API's all the way!!).

BTW, Static T&amp;L wasn't even hot.. it turned out to be essentially useless (giving a small performance boost in games that used it, and in MDK's case, a fairly large boost).

as for doing the math, I don't expect the Kyro series to continue being nearly 100% efficient for fillrate..

EVEN if they did, nVidia has very capable engineers (they bought 3dfx/Gigapixel did they not?), so they could release drivers that did more bandwidth saving measures then they already have on the GF3 (trust me the GF3 has some bandwidth saving measures built in, at least into the drivers, I don't know about the hardware itself).

Kyro will not have an easy fight, even with the potential of this chip.

 

[Teasy]

<<<as for doing the math, I don't expect the Kyro series to continue being nearly 100% efficient for fillrate..>>>

Why wouldn't it continue to be 100% efficient? Kyro III will have the same fillrate to memory bandwidth balance that Kyro II had, Kyro II has a 175mhz GPU with 2 pixel pipes and 175mhz SDR ram and Kyro III will have a GPU with 4 pixel pipes (double the amount of pipes) at 250mhz and 250mhz DDR ram (double data rate)..so it'll have exactly the same balance between fillrate and memory bandwidth as Kyro II.

<<<EVEN if they did, nVidia has very capable engineers (they bought 3dfx/Gigapixel did they not?), so they could release drivers that did more bandwidth saving measures then they already have on the GF3 (trust me the GF3 has some bandwidth saving measures built in, at least into the drivers, I don't know about the hardware itself).>>>

There's not much chance that Nvidia are going to use any sort of software HSR or something in the drivers because that would take up allot of system CPU cycles and make the Geforce 3 more CPU limited then other cards. I don't think Geforce 3 has any bandwidth saving techniques in the drivers, it uses what Nvidia call crossbar technology which is basically cutting the 128bit DDR memory bus up into 4 seperate 32bit DDR buses which it more efficient then having one big 128bit bus. But don't think its even anything like as effecient as TBR because its not in the same league at all. Just look at it this way, currently if we look at raw specs and ignore TBR Kyro II is as fast as a TNT2 Ultra, it has the same raw specs. Yet it not only crushes the TNT2 Ultra, it also crushes the Geforce 2 MX and challenges the Geforce 2 GTS and thats all because of TBR, do you think crossbar technology could turn a TNT2 Ultra into a Geforce 2 GTS?...not a chance, it couldn't even make it close to a MX. Now lets compare the Kyro III specs to an Nvidia card. 4 pixel pipes at 250mhz and 250mhz DDR ram and HW T&amp;L, thats the same raw spec as a Geforce 2 Ultra, so the Kyro III will be to the Geforce 2 Ultra what the Kyro II was to the TNT2 Ultra, now your getting the picture The Geforce 3 will be crushed by Kyro III and I have no boubt about that, Nvidia are going to have to make sure that the card that follows the Geforce 3 is significantly faster then Geforce 3 in order to keep up with Kyro III.

 

[rickn]

Teasy: I dont think k3 is gonna have 250mhz RAM, I think it will use 200mhz DDR (400mhz theoretical)

Maybe kyro 3 will use asyncronous clock rates. 250mhz core/200mhz (400mhz) DDR

I think 3.5ns memory would be cost prohibitive, Imagtec has always tried to produce high perf-low cost cards.

 

[jbirney]

I thought that the GF3 uses 4-64 Bit memory interface. I could have swore (note I usually do....heheheh ok back to the topic) that its a 256 bit memory interface...


I also don't see any reason why the would lose any efficiency unless there is a difference on how it handles doing TnL on a per tile bias? Or there is a slow down when the different pixel pipes try to access the main memory? No doubt that if its close to target, it will be a nice product to own. I am not going to say its goanna be an GF3 killer last time I said that, the card I was talking about never saw the light of day..cough cough Rampage cough cough

 

[Teasy]

<<<Teasy: I dont think k3 is gonna have 250mhz RAM, I think it will use 200mhz DDR (400mhz theoretical)

Maybe kyro 3 will use asyncronous clock rates. 250mhz core/200mhz (400mhz) DDR

I think 3.5ns memory would be cost prohibitive, Imagtec has always tried to produce high perf-low cost cards>>>

IMGTEC have never went asyncronous so I doubt they'd do it with Kyro III, they also like a cetain balance between fillrate and bandwidth, there whole philosophy is efficiency so IMO they'll keep the synchronous system.

<<<I thought that the GF3 uses 4-64 Bit memory interface. I could have swore (note I usually do....heheheh ok back to the topic) that its a 256 bit memory interface...>>>

Nope its a 128bit DDR bus, so Nvidia call it a 256bit bus because its 128bit+DDR = 256bit (it doesn't really work out like that though, its probably more like the equivalent of 220bit SDR or something like that). So it uses 4 32bit DDR buses which people call 64bit because there DDR.

 

[PotNoodle]

Teasy,

? I don't think Geforce 3 has any bandwidth saving techniques in the drivers, it uses what Nvidia call crossbar technology which is basically cutting the 128bit DDR memory bus up into 4 seperate 32bit DDR buses which it more efficient then having one big 128bit bus.?

You forgot the early Z rejection and Occlusion detection. Early Z get most benefit when front-to-back rendering is on place, and occlusion detection needs to be coded for.


rickn,

?Maybe kyro 3 will use asyncronous clock rates. 250mhz core/200mhz (400mhz) DDR?

Would the results justify the cost? IMO no.

Asynchronous memory clocks are really rather hard to implement, and to-date the only gfx manufacturer to do it very successfully is NVIDIA. Is it worth Img doing this when the results are likely to be minimal? KYRO is already texture bandwidth limited, and so increasing the core speed will likely make little difference.

I think we will see a straight synchronous clock, however small die sizes allow for great speed range flexibility. I think you are correct that the ?stock? version will be a low speed one to catch the mainstream market, but then they also have options for higher cost ?Ultra? specs at higher speeds for the higher end as well.

However, given the breadth and diverse technology solutions ImgTech has it wouldn?t surprise me if they have made use of Asynchronous clock elsewhere in the business, so who knows?

jbirney,

?I thought that the GF3 uses 4-64 Bit memory interface. I could have swore (note I usually do....heheheh ok back to the topic) that its a 256 bit memory interface...?

No ? it?s a cheat term.

They call it a 256bit bus because it?s a DDR bus; i.e. because DDR passes two chunks of data per clock it has the ability to pass 256bits per clock, but it isn?t actually 256bits wide. The memory controller on GF3 is actually 4 32bit DDR controllers.

 

[rickn]

Well, if they are gonna do syncronous there is no way it will be clocked at 250mhz. 64mb of 3.5ns DDR memory will still be expensive. Imgtec has bigger roots in europe, Brits get mouthy if they have to pay over $100 GBP for a videocard, there is no way a Kyro 3 with 64mb of 3.5ns DDR is gonna come in under $200 GBP

 

[Teasy]

<<<You forgot the early Z rejection and Occlusion detection. Early Z get most benefit when front-to-back rendering is on place, and occlusion detection needs to be coded for.>>>

Yeah I didn't mention them because as you say without front to back sorting Early Z is totally hit and miss and occlusion detection needs special coding.

<<<Well, if they are gonna do syncronous there is no way it will be clocked at 250mhz. 64mb of 3.5ns DDR memory will still be expensive. Imgtec has bigger roots in europe, Brits get mouthy if they have to pay over $100 GBP for a videocard, there is no way a Kyro 3 with 64mb of 3.5ns DDR is gonna come in under $200 GBP>>>

Where do you get that from?, any product in the U.K is allot more expensive then in the U.S, every new graphics card released in the U.K (until recently with Kyro I and II) has been very expensive, I had to pay £250 for my Voodoo5 and the Geforce Ultra cost about £400 here when it was first released and people still bought it.

 

[jbirney]

Doh, thanks for the clairifications. I work in a large CE company. When we say 256bit data bus we mean we have 256 runners to worry how to route Dam their marketing! Thanks again

 

[PotNoodle]

&quot;64mb of 3.5ns DDR memory will still be expensive. Imgtec has bigger roots in europe, Brits get mouthy if they have to pay over $100 GBP for a videocard, there is no way a Kyro 3 with 64mb of 3.5ns DDR is gonna come in under $200 GBP&quot;

Img / PowerVR have very little say in the matter. Once the chipset is licensed to ST it is completely in ST's hands as to what is achieved with that design - Its ST that negociates with board vendors, its ST that sets the specifications for those board vendors, its ST that gets the moment from the board vendors. If ST feels they can make money on a chip with 250Mhz RAM they will sell a chip that requires 250Mhz RAM - if Img don't like it then Videologic will simpley sell a lower speed chip, but that is basically all they can do about it. KYROII wasn't driven by Img / PowerVR, it was driven by ST - they are the ones that did the die shrink for a design they already have the rights to, they are the ones that calculed the clock speed that KYROII would run at; they are ultimately the ones that got Hercules on board.

Obviously the interplay and relationship between ST and Img is intertwinned a little more closely, but its an illustration that once ST licenses the design its theirs to do with as they please. If STG5500 is has clockspeed flexability then I fully expect to see a range of speeds released; Videologic just have to pick which price range they go for (purely low end, or the lot).