PowerVR Series5 still alive and it's "high end"

[demalion]

Hmm? Series 5 isn't for handhelds or the other embedded applications...that product seems already served by other levels of technology. All the possible uses that I understand as being applicable to Series 5 are under "Later", so there is no disassociation between a high end PC part and Series 5 at all presented in that slide. Atleast, as far as my understanding is correct.

AFAIK as well, they already have products suitable for all the "Current" focus uses available to manufacturers.

 

[Kristof]

Sigh... currently just too busy to teasy you guys by spelling out "No comment" using smilies 8)

K-

 

[Entropy]

Sigh... currently just too busy to teasy you guys by spelling out "No comment" using smilies 8)

K-

As long as you're busy bringing the Series5 to market, all is forgiven!

However, if it's another portable gaming platform that you're developing for STM to back away from....

Entropy

 

[Joe DeFuria]

Well, do you really need 30GB+ of bandwidth for a TBDR? It'd kinda be overkill, no?

First of all, who says that GDDR-2 needs to be 30 GB? GDDR-2 doesn't mean 256 bit. Why not 128 bit GDDR2? Is 15 GB overkill?

In any case, 30 GB is not even overkill...as long as the chip has the fill-rate to utilize it. This is what every "I want to see a monster TBR come to market" advocate (including myself) wants to see.

No, 30 GB/sec is not overkill for a TBDR. Just pair that bandwidth up with a TBDR core that has the effective fill rate to utilize it!.

That does of course, beg the question of whether or not such a core can be built using the targted fabrication process.

 

[Simon F]

Back to the topic at hand....

I see nothing to indicate that the "Series5" is actually the achrchitecture that ImgTech is talking about when referring to the PC space,

I don't think I'll get into any trouble by saying "it is".

 

[demalion]

Regarding the prior observation about development costs...I think the nVidia figures are due to depending on absolute bleeding edge process technology for the complexity of their chips. The further from the bleeding edge the process and clock speeds you depend upon for delivering your technology, the cheaper it should be to develop (by a significantly "more than linear" dropoff). But that's just a "reasonable guess", not based on particular first hand knowledge that I have in mind right now.

 

[RussSchultz]

Regarding the prior observation about development costs...I think the nVidia figures ...

I think they're also coming up with the absolute cost of entry. Some of those costs will be amortized across different products.

 

[Joe DeFuria]

Hmm? Series 5 isn't for handhelds or the other embedded applications...that product seems already served by other levels of technology. All the possible uses that I understand as being applicable to Series 5 are under "Later", so there is no disassociation between a high end PC part and Series 5 at all presented in that slide.

You didn't understand what I was saying, or perhaps I wasn't clear enough.

My only point was "SERIES 5" is not mentioned anywhere. Saying that something is coming "later" does not mean that the first expected product will be based on what we have come to know as series 5. At one point the "next" chip was going to be based on series 4. We know how that ended. For all I know, "series 5" (supposed to be out early '03, no...late '03....when?) has been dropped in favor of Series 6...slated for arrival 2H '04.

(Edit...Looks like Simon just answered the "is it Series 5" question with an affirmative!)

In any case, this might be a good time for someone to "summarize" the currently accpeted rumors for what Series5 is supposed to be, and when we are/were supposed to see it.

 

[Joe DeFuria]

Back to the topic at hand....

I see nothing to indicate that the "Series5" is actually the achrchitecture that ImgTech is talking about when referring to the PC space,

I don't think I'll get into any trouble by saying "it is".

Lol...Great!

Now, shall we press our luck? How about saying "when" it's targetted for arrival? You won't get in trouble, from me anyway....honest!

 

[jb]

Now, shall we press our luck? How about saying "when" it's targetted for arrival?

I got this one for ya Simon:

When its done.

Ok sorry that was not very funny.....

 

[Kristof]

http://www.electronicstimes.com/story/OEG20021003S0016

 

[demalion]

Aren't those the types of statements that cause me to *poke* and *prod* in your general direction, Kristof?

*prod* Simon F

Don't want you feeling left out...

 

[Kristof]

*prod* Simon F

Hmmm... think you need to **prod** a bit harder, don't think he noticed... yeah he is getting a day older

 

[micron]

Debut 2003 on a .13micron process?......wOOt!!
I'm going to go reinstall my Prophet 4500 and play with it!

 

[Hellbinder]

There are some very interesting comments in that Article.

Metcalfe said that Series-5 architecture would be based on a primary processing pipeline and a series of hardware accelerators that can be optionally switched in and out to render graphics.

This is pretty interesting... External Geometry Co-Processing??? External Vertex Shader Engine?? I know that Large Geometry was always considered the limiting factor for taking TBR to the next level. Thus perhaps they have seperated the memory Bandwidth requirements somehow?

“As well as class-leading functionality, there will be some unique features enabled by tile-based rendering,â€￾ said Metcalfe.

Really, really, really high levels of FSAA?? perhaps

Metcalfe added that the Series-5 architecture would debut in 2003 in a 0.13-micron process technology. He said that it was not yet decided whether Imagination would get first silicon implementations or test chips made at a licensee's wafer fab or at a foundry wafer fab in Taiwan.

Considering the usually low Transistor count needed by Imagine's Products, and the indications of external coprocessing... At .13Um A 8 pipelined PVR chip should run at 500mhz core easy.

Which makes for an interesting speculation
8 pipelines
500mhz Core
128bit Memory interface
perhaps 4xxmhz DDR with 13ish GB bandwith???

Although... Past generations have always had syncronus Core/Ram. So 4xx/4xx may be more realistic. However looking at PowerVR's past performance models.. 8 pipines at 400mhz with 13ish GB.... could have some seriously SEVERE performance, and FSAA could be beyond anything seen so far.

 

[BoddoZerg]

Seeing as how people keep comparing the Xbox to the Dreamcast (both favorably and unfavorably) I would like to see a PVR chip in Xbox2.

Then nVidia would be free to go to PS3 as everyone is rumoring.

 

[PiNkY]

You can read about the new SIS graphics spinoff, XGI, for instance here:
http://www.xbitlabs.com/news/video/display/20030520142447.html

Look at the numbers mentioned.
I've found it remarkable how Jen-Tsun of nVidia in every public adress I saw referred on the web remarked how high the cost for GPU development had escalated. Now why would he keep pointing out that? Well, the only really valid reason I could see was that he wanted to deter one or more potential competitors from entering the fray.

All other indications I've seen, such as the size of S3, the rough size of XGI, other industry data points, imply that the number you come up with, roughly half a billion dollar, is vastly inflated, and purely derived from such nVidia public adresses.

On the other hand, I have no way to pinpoint what the actual cost would be for ImgTech.

Entropy

Thanks a lot for putting things in perspective, although I want to point out that the mentioned 13M $ don't say much about the ip value of Xabre(I/II), depending whether the company just licenses the chip design from SIS (quite likely IMO, as SIS is propably not willing to loose this IP in case the XGI venture turns out to be a failure and gets subsequently liquidated).

 

[demalion]

*prod* Simon F

Hmmm... think you need to **prod** a bit harder, don't think he noticed... yeah he is getting a day older

So...

is insufficient, and I need to move on to...

 

[Hellbinder]

Oh one more thing for Simon and Kristof.

I really hope your company takes the Handcuffs off your hardware this time around.. and *Seriously* address the High-End. Dont just go with barely enough Core/Ram/Bandwidth to eek out a few wins here and there.

There is no doubt in my mind that you guys could pull a 9700pro on the entire industry if you wanted to. It is obvious that every time you release a new product.. *if* it had been clocked the same as the competition... it would have completely Demilished even the closest competitor.

For once in your companies life.. Make a damn statement for a change. As personally, i think that an 8 piped TBR card clocked at the same MHZ as the Nv35 would be MANY % faster in most benchmarks. i could wager based on past experience 200-250% faster just like the 9700pro was over the GF4.

 

[RussSchultz]

I think part of the problem with PowerVR always being behind the curve is their business model (selling IP).

The product generally to be finished to before anybody will buy it, and by the time somebody does buy the IP, and productize it, and market it, its behind the curve (of a company that can send it to the fab the day it finishes being designed).

Hopefully, something is wrong with what I've said, or they've found some brave souls to step up to the plate without the goods being finished.

I'm tired of a two horse town.