KYRO enhanced T&L

[Joe DeFuria]

Typedef,

If you buy this thing over a Ti200, you just *must* want to buy anything other than nVidia...This seems like a product without a market.

I agree and disagree. First for some assumptions to set the tone for the rest of this post, which may or may not turn out to be accurate, but are my best guess based on what we know:

1) K2SE is priced a little below the GeForce4 MX 440 and Radeon 7500 levels.
2) K2SE performs a little below those cards as well.

Now, let's get your nVidia-centric thing out of the way. If we assume the street price of the K2SE is within $50 of the Ti-200 AND Radeon 8500, (like the MX 440/460 is) then I would say for a GAMER, the K2SE is not interesting at all. It's not a case of not wanting to buy nVidia, it's a case of wanting to get the best performance / features for the dollar, at a reasonable absolute price.

Note that this also means, that the nVidia GeForce4 MX 440, and certainly the MX 460 are also "products without a market" by your definition.

I basically consider all cards within the price of one game (about $50 U.S.) to be in "competition" for the gamer's dollar. So I do agree that any gamer should find a way to come up with $50 to "step up" to a Radeon 8500 or Ti-200. Sacrifice the purchase of one game, to get a better overall experience with the games you do buy.

But there obviously is a "market" for the K2SE, just as there is a market for the GeForce4 MX. IMO, though, that isn't a market for most gamers.

 

[mboeller]

A couple in my definition means two (2) not more.

By the way I thought spring ends in Germany around the same time as here? hehehehe

Ahh.. I see.

Well my englisch seems not good enough. I translated couple = a few (the dictionary says the same!); and so with the "IMG delays everything" in mind I thought march + 3-5 ( = a couple ) + 2-3 (IMG delay(TM) ) = I got september as the earlist date.

 

[Teasy]

Well my englisch seems not good enough. I translated couple = a few (the dictionary says the same!

Yeah thats the informal meaning for a couple but when being strictly accurate a couple means two.

 

[Joe DeFuria]

For those interested, here's the "Fu Interview" that's being talked about:

http://www.extremetech.com/article/0,2299,s%3D201%26a%3D2529,00.asp

At the time, June 2001, it was expected that the 4800 would be out "in a couple months." That has turned out to be close to a year. (We're at 10 months now, and the product is not on the shelves yet.)

At the time, he also anticipated the 5500 (Kyro III), to be out by the end of 2001. If we delay that launch by the same amount of time the 4500 was delayed (10 months), that would but the KyroIII out this fall...September. Certainly, that method of estimation is crude.

It's also interesting that spec wise, the 4800 was exactly what he said it would be back in June: 200 Mhz respin of the 4500. So perhaps we can do the same with the KyroIII:

• 0.13 micron
• "3 times" the performance of the "4000" series.
• 4 pixel pipelines
• 250 Mhz
• Hardware T&L, no mention of "programmable" T&L. (Hints to DX7 compliant part.)
• Max of 64 MB DDRAM

Price wise, if this part were to be released today, this part should probably be somewhere around Readon 8500 and GeForce Ti-200 levels: $150 or so U.S.

Performance wise? My guess would be somewhere between a Ti-200 and Ti-500. (Based on a quick look at KyroII benchmarks at fill-rate limited situations over at Anandtech...) That would still be a hard sell for this part if the Kyro lacks DX8 pixel shaders.

 

[Dave Baumann]

Performance wise? My guess would be somewhere between a Ti-200 and Ti-500. (Based on a quick look at KyroII benchmarks at fill-rate limited situations over at Anandtech...) That would still be a hard sell for this part if the Kyro lacks DX8 pixel shaders.

Probably not.

Most likely KYROIII will be going against the GF4 MX's and Radeon 7500's of this world, both of which are DX7 compliant thmselves. They all may have a tougher time against the lower end DX8 cards of Radeon 8500LE and GF4 Ti4200's (remembering that Ti200's are likely to be phased out once stocks die down).

 

[king_iron_fist]

Does anyone know what's happening over at cebit at the moment?

BTW I think the kyroIII would be at least GF4ti levels. The sustained fillrate would be over 1000Mpixels which is more than a GF4 and it would be backed up by a very fast T&l engine.

I'm determined to make this card my next upgrade.

 

[Dave B(TotalVR)]

Price wise, if this part were to be released today, this part should probably be somewhere around Readon 8500 and GeForce Ti-500 levels: $150 or so U.S.

Price wise yeah.

Performance wise? My guess would be somewhere between a Ti-200 and Ti-500. (Based on a quick look at KyroII benchmarks at fill-rate limited situations over at Anandtech...)

Its not just gonna be a 4 piped 250Mhz KYRO II y'know.

That would still be a hard sell for this part if the Kyro lacks DX8 pixel shaders.

Unless of course it were to come with some frivelous and useless feature such as totally free full scene anti-aliasing or something equally as irrelevant as that.

As for you crude estimation of a september release for the KYRO III, well its rediculous to assume that because the KYRO II SE was delayed 10 months (not that it was ever officially announced before yesterday) that the KYRO III will be. If the KYRO III were to come out in 10 months then it would be another budget card and I dont think thats what Via wants, do you? its just as likely that they held back the KYRO II SE because the K2 was selling so well and concentrated their work on the KYRO III, in fact I would say its a site more likely.

Dave

 

[Joe DeFuria]

Most likely KYROIII will be going against the GF4 MX's and Radeon 7500's of this world, both of which are DX7 compliant thmselves.

Feature wise, yes, the KyroIII would be best matched against those parts. But price wise, I'm betting on closer to the Radeon 8500/Ti-200. Mostly because of the memory interface, which AFAIK, still dictates the majority of the cost of these products. Here's a snippet of all the current, relevant chips:

KRYO III: 250 Mhz DDR
Readon 8500: 275 Mhz DDR
Radeon 8500 LE: 250 Mhz DDR
Radeon 7500: 230 Mhz DDR
GF4-MX 460: 275 Mhz DDR
GF4-MX 440: 200 Mhz DDR
GF3-TI-200: 200 Mhz DDR
GF3-TI-500: 250 Mhz DDR

Chip complexity is another cost factor of course, and I'd guess a 4 pixel pipe, DX7 style Kyro chip will sit somewhere between the complexity of the 2 pixel pipe competing DX7 chips, and the 4 pixel pipe DX8 chips.

So looking at all those numbers, I'd say the price should come somewhere about a Radeon 8500 LE, or TI-200.

all may have a tougher time against the lower end DX8 cards of Radeon 8500LE and GF4 Ti4200's

Agreed...at least for the gaming market. Once the DX8 cards hit the $150 barrier, it becomes hard to justify a similarly performing DX7 card if there's less than $50 or so price difference.

remembering that Ti200's are likely to be phased out once stocks die down).

That's a good point. Of course, it's being phased out to introduce the better performing Ti-4200 with MSRP of $199...and who knows where the street price will settle at by the time KyroIII is released....

 

[Teasy]

At the time, he also anticipated the 5500 (Kyro III), to be out by the end of 2001.

Actually AFAIR he said late 2001 or early 2002.

that method of estimation is crude.

Yep because the release of a product does not have to be based on when nother product is released. The estimated release of late 2001 to early 2002 was unlikely to be like that as to be a certain amount of time after Kyro IISE. It's more likely estimated at that time simply because that's when they expected the chip to be ready (possibly thats when they expected TSMC's 0.13 micron to be finalized). Also the company making those announcements will no longer be residing over the launch of Kyro III so all of what he said pretty much goes out of the window when trying to figure out when Kyro III will be out.

It's also interesting that spec wise, the 4800 was exactly what he said it would be back in June: 200 Mhz respin of the 4500. So perhaps we can do the same with the KyroIII:


0.13 micron

"3 times" the performance of the "4000" series.

4 pixel pipelines

250 Mhz

Hardware T&L, no mention of "programmable" T&L. (Hints to DX7 compliant part.)

Max of 64 MB DDRAM

The "3 times" the performance of the "4000" series is not really valid considerin FU also mentioned that Kyro II was around the performance of a Geforce 2 MX instead of a Geforce 2 GTS as it actually is. Also the specs suggest 4-5 times the performance of 4000 when looking at pure bandwidth/fillrate without considering any other optimisations and features like HW T&L and possibly free AA. I think FU probably meant 3 times the performance of the 4500 series which would make sense when looking at pure specs.

Price wise, if this part were to be released today, this part should probably be somewhere around Readon 8500 and GeForce Ti-500 levels: $150 or so U.S.

A DX7 chip with static HW T&L and 4 pixel pipes, a normal single 128bit DDR bus and 250mhz ram on 0.13 micron.. I don't see why that should be the same price as a Radeon 8500 which has pixel shaders and vertex shaders on a larger process as well as an expensive segmented memory bus. IMO its very unlikely to sell for $150 online. Its reccomended price will be $150 AFAIK (thats according to IMGTEC), meaning actual online prices of more like $120 or less are likely.

 

[Joe DeFuria]

Its not just gonna be a 4 piped 250Mhz KYRO II y'know.

I know...but the performance estimate according to Fu is "3X" that of the 4000 series. That's what I based my estimation on.

Unless of course it were to come with some frivelous and useless feature such as totally free full scene anti-aliasing or something equally as irrelevant as that.

That's a good point. Something we've been asking the PowerVR team for since, well, forever.

As for you crude estimation of a september release for the KYRO III, well its rediculous to assume that because the KYRO II SE was delayed 10 months (not that it was ever officially announced before yesterday) that the KYRO III will be.

Yeah, it's about as rediculous to estimate a March / April release of the Kyro II SE when Fu said it would be a "couple months" from June, right?

I would wager that if the KIII was "ready for the shelves" a couple months from now, it would have been debuted at CeBit, along with the 4800.

In any case, if you want a more "based" reason for the September estimate, I'll repeat what I've said since the beginning: all the chip makers seem to have trouble getting a 0.13 part out this spring, and appear to have pushed 0.13 products out to the fall.

If the KYRO III were to come out in 10 months then it would be another budget card and I dont think thats what Via wants, do you?

Is that what STM wanted for the 4800? What a company "wants" is not always what they get. Did STM "want" the 4800 coming out this year in April? It's now "just another budget card".

its just as likely that they held back the KYRO II SE because the K2 was selling so well and concentrated their work on the KYRO III, in fact I would say its a site more likely.

Heh...well, you believe what you want....I don't know why on one hand you say Via "doesn't want just another budget card", and then on the other, you feel it's "likely" that STM "delayed" the KIISE to put it exactly in that position...

 

[Dave Baumann]

Chip complexity is another cost factor of course, and I'd guess a 4 pixel pipe, DX7 style Kyro chip will sit somewhere between the complexity of the 2 pixel pipe competing DX7 chips, and the 4 pixel pipe DX8 chips.

The effect on price that 'chip complexity' has is usually down to wafer size, in which case the process factors in largely. If KYROIII is .13um then it will have a wafer (per transitor required!) advantage over all the current chips .

Also, despite GF4MX 'only' having two pipes the new AA and other factors increase its complexity quite considerably.

 

[Teasy]

Feature wise, yes, the KyroIII would be best matched against those parts. But price wise, I'm betting on closer to the Radeon 8500/Ti-200. Mostly because of the memory interface, which AFAIK, still dictates the majority of the cost of these products. Here's a snippet of all the current, relevant chips:

Your forgetting that all those other chips have segmented memory buses which is more expensive then a single bus.

Chip complexity is another cost factor of course, and I'd guess a 4 pixel pipe, DX7 style Kyro chip will sit somewhere between the complexity of the 2 pixel pipe competing DX7 chips, and the 4 pixel pipe DX8 chips.

Kyro was 12 million transistors, I'd be suprised if Kyro III was anything over 25-30 million transistors, all the chips you mentioned thier are in the 40-50 million transistor range (appart from the Geforce 4 MX which is cheap and DX7 anyway) and are on larger processes then Kyro III. Plus IMGTEC have already given a reccomended price of $150, which means less then $150 online at release.

 

[Joe DeFuria]

The effect on price that chip complexity has is usually down to wafer size, in which case the process factors in largely. If KYROIII is .13um then it will have a wafer (per transitor required!) advantage over all the current chips .

And who's to say that by the time Kyro III is released, that nVidia / ATI won't have 0.13 micron variants of their value parts as well? Isn't that something that they traditionally do?

Problem is, we're comparing "future" Kyro, with "present" ATI/nVidia ones. It's not really valid to consider the costs related to the fab process of a chip that already on the market, with one that is yet to materialize. You would be right if the KyroIII was shipping today.

Also, despite GF4MX only; having two pipes the new AA and other factors increase its complexity quite considerably.

And we know nothing of any "other" features of KryoIII that could increase it's complexity considerably compared to the KyroII. What would "free FSAA" cost in complexity, if implemented? Will they add dual-head support? Alter their memory interface?

 

[Joe DeFuria]

Your forgetting that all those other chips have segmented memory buses which is more expensive then a single bus.

True, but do we know that KyroIII doesn't have a segmented bus?

all the chips you mentioned thier are in the 40-50 million transistor range (appart from the Geforce 4 MX which is cheap and DX7 anyway)

Well, isn't that what KyroIII is suppossed to be? "Cheap DX7"? Thing is, the most expensive of the "cheap DX7" (MX 460) parts are very close in price to the cheapest of the most expensive DX8 parts.

So it comes down to this -- do you think KyroIII will be priced on par with the "cheap" DX7 parts, or the "most expensive" DX7 parts?

Plus IMGTEC have already given a reccomended price of $150, which means less then $150 online at release.

Again, the key question is WHEN? Even if it's $100 street , if it comes out in september, who knows what the lanscape will be when we currently have sub $150 DX8 cards...TODAY.

 

[Dave Baumann]

And who's to say that by the time Kyro III is released, that nVidia / ATI won't have 0.13 micron variants of their value parts as well? Isn't that something that they traditionally do?

Thus far NVIDIA haven't done that with the lower end variants - they have in terms of process, but not in terms of product. The 'MX' product lasted in excess of a year and one whole generation for GF2; personally I don't see NVIDIA releasing another MX class product until the second generation of NV30 type products, probably on .10um/.09um. ATi may well have pushed R200 down to .13um but then I'd still expect that to be considerably more complex as well.

What would "free FSAA" cost in complexity, if implemented? Will they add dual-head support

Considering they have levered free FSAA into the low cost / complexity model such as MBX I'd guess not a lot! Remember, NVIDIA are basically using more die space for FSAA to save bandwidth by doing more AA operations on chip; the very nature of PowerVR's architecture already has this one solved.

 

[Teasy]

And we know nothing of any "other" features of KryoIII that could increase it's complexity considerably compared to the KyroII. What would "free FSAA" cost in complexity, if implemented?

I wouldn't think it would be much, it should basically be MSAA. As for other features.. hmm, well thier will be cube mapping I suppose, but again this shouldn't up complexity very much.

Will they add dual-head support?

Unlikely

Alter their memory interface?

No, a single 128bit DDR bus is what I'd expect.

I'd say when you take Kyro III's specs into account it should be no more complex transistor wise then Geforce 4 MX (less IMO) and considerably less complex then the other chips you mentioned, put that together with a less complex memory bus and you have a cheaper product. $150 reccomended is likely to be its price and that is a long way below a Radeon 8500's recommended price or a Geforce 3's recommended price.

True, but do we know that KyroIII doesn't have a segmented bus?

It wasn't mentioned by fu what bus config it has, but then if it was segmented I'd expect it to be mentioned, after all he's not going to give a spec that says "not a segmented bus" is he. Also considering a tilers bandwidth advantages its unlikely to be top of their list of needed features.

Well, isn't that what KyroIII is suppossed to be? "Cheap DX7"?

that's a fair assumption yes, but its likely to have the feature set of a Geforce 4 MX 460 at a slightly cheaper price (no need for the bandwidth saving tech or complex AA or segmented bus, with slower ram of 250mhz rather then 275mhz and on a smaller process) and much faster.

So it comes down to this -- do you think KyroIII will be priced on par with the "cheap" DX7 parts, or the "most expensive" DX7 parts?

I think it'll be a little cheaper then the most expensive DX7 parts and a whole lot faster, especially with free AA (if implimented).

Again, the key question is WHEN? Even if it's $100 street , if it comes out in september , who knows what the lanscape will be when we currently have sub $150 DX8 cards...TODAY.

Even if its released at a time when Geforce 4 and Radeon 8500 are on 0.13 micron (if that happens anyway, it certainly won't happen for Geforce 3) then it still looks to have a price advantage transistor and mem wise over those chips and so could be sold cheaper if needed.

 

[Dave Baumann]

Will they add dual-head support?

Unlikely

IIRC the block diagram of series 4 showed two RAMDAC's.

 

[Joe DeFuria]

Teasy,

The "3 times" the performance of the "4000" series is not really valid...

I think ther's just confusion about my use of the "4000 series". I do not mean the STG 4000 part, as I don't think that's what Fu meant. I mean the 4XXX series of cards...4000, 4500, 4800. For my performance estimate (somewhere between the Ti-200 and Ti-500) I used 3X FPS of the 4500 that Anand used in his benchmarks.

 

[Joe DeFuria]

$150 reccomended is likely to be its price and that is a long way below a Radeon 8500's recommended price or a Geforce 3's recommended price.

Again...right now. I don't expect the graphics landscape to stand still while waiting for the KyroIII release.

Thus far NVIDIA haven't done that with the lower end variants - they have in terms of process, but not in terms of product.

Not sure what you mean by that. nVidia has reduced the process size of the GeForce2 MX. They used to just have one "MX" part, then they went to a new fab and had 2 new GeForce2 MX parts (MX 200 and MX 400). I think it was the same time they introduced the GeForce2 Pro, (Die shrink of GeForce2 GTS) but I'm not sure.

 

[Dave Baumann]

I'm saying the missed out entire generations of their low end product - the original MX didn't occur until after the high end was refreshed to GeForce 2, and the current MX product missed out a the entire GeForce3 line - which I why I don't see them introducing another low (low) end prodct for some time.